Retire contacts.contact_type; derive Contacts status from the grid (v0.1.0:106)

The Investors/Prospects distinction is now derived live from the canonical
grid (contact_grid_signals -> committed/pipeline_stage), not the mechanically
set contact_type column:

- Desktop Contacts: drop the Investors/Prospects tabs + TYPE badge; show a
  derived Status (existing-LP badge + pipeline stage chip).
- Dashboard: repoint Total LPs / Prospects onto fundraising_investors entities
  (committed>0 vs $0, graveyard + blank-row placeholder excluded); fix a
  total_contacts soft-delete leak.
- Stop reading/writing contact_type across the create/update/import/sync paths.
  The column is left inert in place; a physical drop is deferred to a later
  signed-off table-rebuild migration (SQLite no-drop-column; contacts is
  FK-referenced) -- same retire-then-drop path lp_profiles took.

.claude/launch.json

@@ -0,0 +1,11 @@
+{
+  "version": "0.0.1",
+  "configurations": [
+    {
+      "name": "crm-preview",
+      "runtimeExecutable": "bash",
+      "runtimeArgs": ["-c", "CRM_DB_PATH=/tmp/crm_preview.db CRM_DATA_DIR=/tmp/crm_preview_data CRM_FRONTEND_DIR=$PWD/frontend CRM_PORT=8765 CRM_ENV=development exec python3 backend/server.py"],
+      "port": 8765
+    }
+  ]
+}

.claude/rules/email.md

@@ -0,0 +1 @@
+../../docs/guides/email.md

.claude/rules/matrix-intake.md

@@ -0,0 +1 @@
+../../docs/guides/matrix-intake.md

.claude/rules/migrations.md

@@ -0,0 +1 @@
+../../docs/guides/migrations.md

.claude/rules/nl-query.md

@@ -0,0 +1 @@
+../../docs/guides/nl-query.md

.claude/rules/packaging.md

@@ -0,0 +1 @@
+../../docs/guides/packaging.md

.claude/rules/redaction.md

@@ -0,0 +1 @@
+../../docs/guides/redaction.md

.claude/rules/spark-ingest.md

@@ -0,0 +1 @@
+../../docs/guides/spark-ingest.md

.claude/rules/thesis.md

@@ -0,0 +1 @@
+../../docs/guides/thesis.md

.env.example

@@ -19,3 +19,65 @@ X_API_KEY=
 # ── CRM (ingest opens the SQLite file directly, read-only) ──
 CRM_DB_PATH=./data/crm.db
 CRM_DEV_DB_PATH=./data/crm_dev.db
+
+# ── Daily activity digest (Phase B) ──
+# The daily digest (each team member's activity per investor + a by-investor view,
+# summarized LOCALLY on Spark — never Claude) is controlled from Settings → Admin
+# (stored in the DB). These env vars only SEED the first-boot default before an
+# admin sets it; once the policy row exists, the admin panel wins. The "Send Digest
+# Now" button works regardless. Leave blank to default to off / 6 PM.
+CRM_DIGEST_ENABLED=
+# Local (box-time) hour 0-23. Default 18 (6 PM).
+CRM_DIGEST_SEND_HOUR=18
+
+# ── Daily-digest sender ──
+# The digest mailer prefers Gmail domain-wide delegation (the service account that
+# already powers email capture; its grant includes gmail.compose, which can send) and
+# falls back to SMTP below. For the Gmail/DWD path it sends impersonating this domain
+# user; if unset, it uses the first active admin's email.
+CRM_DIGEST_SENDER=
+
+# ── Daily-digest outbound SMTP fallback (dev override of the per-package mailbox) ──
+# On the Start9 box these are set by the "Configure Digest SMTP" action (written
+# to /data/secrets/smtp/* and exported by docker_entrypoint.sh). For dev, set them
+# here. SMTP_SECURITY is one of: starttls (587) | tls (465) | none.
+SMTP_HOST=
+SMTP_PORT=587
+SMTP_SECURITY=starttls
+SMTP_FROM=
+SMTP_USERNAME=
+SMTP_PASSWORD=
+
+# ── Matrix intake bot (backend/matrix_intake/, runs as its own process on the Spark) ──
+# Parses a typed message in a dedicated Matrix room into a proposed fundraising-grid
+# add/edit (local Qwen via Spark Control above), then writes through the CRM API only
+# after in-thread human approval. Reuses SPARK_CONTROL_URL / CRM_CHAT_MODEL above.
+MATRIX_HOMESERVER=https://<homeserver>
+MATRIX_USER=@intake-bot:<homeserver>
+MATRIX_ACCESS_TOKEN=
+MATRIX_DEVICE_ID=ten31-intake-bot
+MATRIX_INTAKE_ROOM=!<roomid>:<homeserver>
+# Dedicated room for reviewing CRM-drafted email-activity proposals (the proposed grid notes the
+# Email Capture panel shows). The bot posts a review card per pending proposal here and relays the
+# in-thread yes/no/edit back to the CRM, in sync with the web panel. Separate from the intake room
+# so high-volume email proposals don't drown the conversational intake. Leave empty to disable the
+# whole email-review poll loop. The bot must be a member of this room. Needs the server side in the
+# s9pk (≥ v0.1.0:89) and the bot's CRM user set to role 'bot' (see docs/guides/matrix-intake.md).
+MATRIX_EMAIL_REVIEW_ROOM=!<roomid>:<homeserver>
+# Dedicated read-only Q&A room (W2): every top-level message here is answered as a natural-language
+# query (translated on the box's LOCAL model — nothing leaves the box), no '?'/'@bot' trigger needed.
+# The '?'/'@bot' trigger still also works in the intake room. Leave empty to disable the dedicated
+# room (questions then go via the intake-room trigger). The bot must be a member of this room. Needs
+# the server side in the s9pk (POST /api/query/nl) and the bot's CRM user set to role 'bot'.
+MATRIX_QUERY_ROOM=!<roomid>:<homeserver>
+# CRM write-back: the bot logs in as a DEDICATED service user (admin-created CRM user;
+# the CRM has no service-key path, so it uses normal Bearer-JWT auth).
+CRM_API_BASE=http://127.0.0.1:8080
+CRM_BOT_USERNAME=
+CRM_BOT_PASSWORD=
+# Set to false only if CRM_API_BASE is https with a self-signed cert.
+CRM_API_VERIFY_TLS=true
+# Ten31 team-member names (comma-separated), fed to the parser so a teammate's name reads as
+# the person DOING outreach, not the prospect ("Jonathan is chatting with Wyoming" → Wyoming).
+# Optional; first names as actually used in the room. Leave empty to disable the framing.
+INTAKE_TEAM_ROSTER=

.gitignore

@@ -2,6 +2,12 @@
 .DS_Store
 **/.DS_Store
 
+# ── Design provenance: keep the text artifacts (*.dc.html, store.js, *.md, tokens.json)
+#    but not the heavy binaries (the raw .zip export + screenshot/thumbnail PNGs). ──
+design/_imports/**/*.zip
+design/_imports/**/*.png
+design/_imports/**/*.thumbnail
+
 # ── Python ──
 __pycache__/
 *.py[cod]
@@ -13,6 +19,16 @@ __pycache__/
 !.env.beta.example
 data/secrets/
 
+# ── Claude Code (deny by default, allow-list shared wiring — see standards/portability.md) ──
+.claude/*
+!.claude/rules/
+!.claude/agents/
+!.claude/commands/
+!.claude/skills/
+!.claude/settings.json
+# launch.json is a deliberate, secret-free debug config CRM commits
+!.claude/launch.json
+
 # ── Local databases & SQLite sidecars ──
 # CRM is canonical and lives on the Start9 /data volume; never commit live LP
 # data into git (guardrail #1). Dev/synthetic DBs are local-only.

CLAUDE.md

@@ -1,89 +0,0 @@
-# Ten31 Agentic System — Project Memory
-
-This file is the project constitution. Read it first; it states settled decisions and non-negotiable guardrails. Where anything here conflicts with a one-off prompt, this file wins.
-
-## What we're building
-
-Ten31 is an investment platform (bitcoin ecosystem, energy, AI infrastructure, freedom tech) that raises from LPs and deploys into private companies. We are building an in-house system of AI agents to widen the fundraising funnel, sharpen and propagate our investment thesis, and automate marketing/branding. Build agents on the **Claude Agent SDK**, connected to our systems via **MCP**. Frontier reasoning runs on **Claude**; privacy-sensitive and high-volume work runs on **local models on our DGX Sparks**, fronted by **Spark Control**.
-
-Full architecture and rationale: see `@./docs/Ten31_Agentic_Build_Plan.md`.
-Current phase tasks and acceptance criteria: see `@./docs/PHASE_0.md`.
-Embedding/retrieval API contract + ingest recipe (authoritative): see `@./docs/EMBEDDINGS.md`.
-
-**We are in Phase 0.** Phase 0 builds the data + retrieval substrate. There are NO live, outward-facing agents in Phase 0.
-
-## Settled architecture
-
-- **Reasoning model:** Claude via the Agent SDK / API (API-key auth, not claude.ai login).
-- **Local models (Sparks, via Spark Control gateway):**
-  - Chat/triage: Qwen3.6 35B-A3B on Spark 1.
-  - Embeddings: `BAAI/bge-m3` (dense, 1024-dim, L2-normalized) → `/v1/embeddings` (OpenAI shape).
-  - Reranker: `BAAI/bge-reranker-v2-m3` (cross-encoder) → `/v1/rerank` (Cohere shape).
-  - Served by **spark-embed**, a small FastAPI server on Spark 2 (NGC PyTorch image — *not* HF TEI, which ships no arm64 CUDA image). Shipped in Spark Control v0.15.0.
-  - Audio: transcription + diarization + TTS on Spark 2.
-- **Canonical data store:** the self-built CRM on the Start9 server. This is the single source of truth for LP/prospect data.
-- **Vector index:** Qdrant v1.16.0 on Spark 2 (ports 6333/6334). Derived and rebuildable from the CRM (~8–15 min full re-embed) — NOT a second source of truth. But it holds the only *live* copy of the index, so it is never auto-restarted; the ingest pipeline must be idempotent so a rebuild is always safe.
-- **Retrieval:** one orchestrated call, `POST /api/search` (embed query → Qdrant dense+sparse RRF with payload pre-filter → cross-encoder rerank → top_k). The sparse/BM25 leg is generated **client-side** with FastEmbed (`Qdrant/bm25`) at both ingest and query time, with Qdrant applying IDF over our own corpus — so exact entity/name matching is weighted by our term statistics, not bge-m3's pretrained sparse. Authoritative contract + ingest recipe: `@./docs/EMBEDDINGS.md`.
-- **Gateway:** Spark Control (on Start9) fronts all local model services behind one trusted URL with shared TLS, access control, and observability.
-
-## Environment & services
-
-- All local model calls go through **Spark Control**, never directly to a Spark.
-- Endpoints: `/v1/chat/completions`, `/v1/embeddings`, `/v1/rerank`, `/api/search` (orchestrated hybrid retrieval), `/v1/audio/transcriptions`, `/v1/audio/speech`.
-- **Secrets live in `.env` (gitignored). Never commit secret values.** Required variables (names only):
-  - `ANTHROPIC_API_KEY`
-  - `SPARK_CONTROL_URL` — gateway for `/v1/embeddings`, `/v1/rerank`, `/api/search` (reads + dense embeds)
-  - `QDRANT_URL` — direct Qdrant on Spark 2 (`http://<spark2>:6333`) for collection admin + ingest upserts
-  - `X_API_KEY` — the X (Twitter) API key for Scout/Analyst enrichment. **Note:** this is *not* a CRM auth key; the CRM has no service-key/API-key path today (see below).
-  - CRM connection vars:
-    - `CRM_DB_PATH` — absolute path to the SQLite file (default `<CRM_DATA_DIR>/crm.db`). The CRM has **no network DB protocol** — ingest "connects" by opening this file directly (read-only, `mode=ro`), co-located with the Start9 `/data` volume.
-    - `CRM_DATA_DIR` — the `/data` volume root (holds `crm.db`, `backups/`, `secrets/`, `email_attachments/`).
-    - `CRM_BASE_URL` — `http://<host>:8080` (env `CRM_HOST`/`CRM_PORT`), for any HTTP access to the running CRM.
-    - `CRM_SECRET_KEY` — the CRM's own JWT signing secret (set on the Start9 deployment, persisted at `/data/.crm-secret`); only needed if the MCP server authenticates over HTTP rather than reading SQLite directly.
-- A `.env.example` lists the variable names with empty values.
-
-## The agents (target roster — built in later phases)
-
-- **Scout** — monitors public sources (X via API, filings, etc.); flags trigger events. (Phase 2)
-- **Analyst** — builds LP dossiers, enriches records, maps warm-intro paths. (Phase 2)
-- **Architect** — owns/refines the canonical thesis; collaborative copilot. (Phase 1)
-- **Scribe** — distributes the thesis as content across channels. (Phase 1)
-- **Closer** — drafts outreach, nurture, meeting prep. Humans approve/send everything. (Phase 3)
-- **Orchestrator** — schedules and routes work; picks per-agent retrieval modes. (Phase 3)
-
-## Division of labor
-
-- **Spark developer (separate):** TEI serving (BGE-M3 + reranker) and Qdrant on Spark 2, exposed via Spark Control `/v1/embeddings` + `/v1/rerank`.
-- **This repo (Claude Code + the partners):** CRM schema extensions, ingest/sync pipeline, CRM MCP server, retrieval-mode library, and (later phases) the agents.
-
-## Guardrails — NON-NEGOTIABLE
-
-1. **Sovereignty.** Sensitive LP and relationship data stays on our infrastructure (Start9 + Sparks). Send only the minimum necessary, non-sensitive context to the Claude API. Never bulk-export the LP list to any third party.
-2. **CRM is canonical.** Qdrant and any other store are derived. Never treat a derived index as the source of truth; never let them silently diverge.
-3. **No destructive data ops.** Never hard-delete CRM records or history. Soft-delete/archive only. Migrations must be reversible and reviewed before running.
-4. **Human-in-the-loop on anything outbound.** No agent sends email, posts publicly, or contacts an LP/prospect autonomously. Agents draft; a partner approves and sends. (Especially Closer and Scribe.)
-5. **Log every agent action** to the interaction log, for compliance and debugging.
-6. **Compliance gate before Phase 3.** No cold/outbound capability ships until counsel has defined solicitation posture (e.g. 506(b) vs 506(c)), accreditation/QP verification, and recordkeeping rules.
-7. **Secrets never committed.** Use `.env` / a secrets store. No keys, tokens, or credentials in code, configs, or docs.
-8. **Enrichment is one-way and public.** Per-prospect public lookups that write INTO the CRM; never push our data outward.
-9. **Development data handling — keep real LP data out of Claude during the build.** Claude Code (the engineering partner) runs on the Anthropic API, so anything it reads is sent to a third party. Therefore Claude Code works only on **code, the schema, and synthetic or properly-redacted data** — never the real LP list, live records, or raw note/email prose. The real backfill and ingest **run on Ten31 infrastructure** (Start9 + Sparks) via **local models**; sensitive rows are never pasted into a Claude Code session or sent to the Claude API during development. To produce a realistic test corpus, redact/pseudonymize a copy **on the Sparks** (local) — do not hand-feed real records to Claude to "clean up." This is the same sovereignty boundary as guardrail #1, applied to the engineering workflow itself.
-
-## Conventions
-
-*Filled in from the CRM code (2026-06). Full detail: `@./docs/crm-overview.md`.*
-
-- **Language / runtime:** Python 3.11, standard library only at runtime. The CRM is one file, `backend/server.py` (~4.5k lines): a stdlib `http.server.ThreadingHTTPServer` + hand-written `CRMHandler` with manual path dispatch. **Not** FastAPI — `backend/requirements.txt` lists FastAPI/SQLAlchemy/Alembic/Pydantic but **none are imported** (vestigial). The only non-stdlib runtime deps are optional `bcrypt`/`jwt` and (for the Gmail module) `cryptography`.
-- **Storage:** a single SQLite DB (`data/crm.db`), WAL mode, `foreign_keys=ON`, opened per-request via `get_db()`. Two parallel investor models coexist (classic `contacts`/`lp_profiles` + the `fundraising_*` grid) — see `docs/crm-overview.md` §2.3; reconciling them to canonical IDs is the core Phase-0 entity-resolution task.
-- **Migrations:** **additive and reversible only.** Core schema uses ordered `backend/migrations/NNNN_*.sql` files applied once at startup by `backend/core_migrations.py`, tracked in a `schema_migrations` ledger; ship a paired `NNNN_*.down.sql` for rollback. (The Gmail module has its own runner under `backend/email_integration/migrations/`.) SQLite ALTER is add-column/rename only — which enforces the additive guardrail.
-- **Run locally:** `./start.sh` (dev defaults, port 8080). `./start_beta.sh` for a Tailscale/production-mode launch (requires `CRM_SECRET_KEY`). No build step.
-- **Tests / lint:** none in-repo. Sanity-check edits with `python3 -m py_compile backend/server.py`. Verify migrations against a *copy* of `crm.db`, never production.
-- **Production:** Start9 package `ten-database`. **`start9/0.4/` is the live target** (TypeScript SDK manifest under `start9/0.4/startos/`); `start9/0.3.5/` (YAML manifest) is the superseded prior generation. All state on the persistent `/data` volume.
-- **Auth:** username/password → HS256 JWT (Bearer header), two roles (`admin`/`member`), no row-level authorization. `X_API_KEY` (in this file's env list) is the *X/Twitter* key — there is **no CRM service-key path in code**; an MCP/ingest client must read SQLite directly or authenticate as a real CRM user.
-- Prefer clear, reviewable changes over cleverness. Keep the ingest pipeline and MCP server modular so retrieval modes and sources can be added without rewrites.
-
-## First actions for a new session
-
-1. Read `@./docs/PHASE_0.md` and `@./docs/EMBEDDINGS.md` (the latter is the authoritative embedding/retrieval contract and ingest recipe).
-2. Read the CRM source in the repo; produce a short written summary of the storage engine, schema, and API surface, and fill in the Conventions section above and the CRM env vars.
-3. Confirm Spark Control is reachable and `/v1/embeddings`, `/v1/rerank`, and `/api/search` respond (these shipped in v0.15.0; check `GET /api/endpoints`).
-4. Proceed through the Phase 0 workstreams in order. Do not build any outward-facing agent behavior in Phase 0.

CLAUDE.md

@@ -0,0 +1 @@
+AGENTS.md

EVALUATION.md

@@ -0,0 +1,79 @@
+# Evaluation — CRM (Ten31 Venture CRM + Agentic System) — 2026-06-12
+
+Intent: A self-hosted venture-fund CRM for Ten31 (replacing Airtable to keep sensitive LP/prospect data off third-party servers) managing contacts/organizations/opportunities/communications/LP profiles, with a new in-house AI/agentic layer that drafts outreach and sharpens the investment thesis — packaged as a StartOS s9pk, currently Phase 0/1 (agents draft, humans send).
+
+Agents run: evaluator, security-auditor, exerciser, start9-spec-checker. Skipped: reviewer (working tree is clean — no uncommitted diff to review).
+
+## Verdict
+
+This is a functional, deliberately stdlib-only CRM (one ~5,400-line `backend/server.py` monolith over SQLite) with a genuinely well-engineered privacy layer bolted on; it largely achieves its intent, and the redaction boundary that keeps LP data out of Claude is the strongest part of the codebase. **The headline risk is a single P0: an unauthenticated path-traversal in the `/assets/` route (`backend/server.py:1717`) that runs *before* the auth gate and lets any LAN/Tailnet client read arbitrary files — the live LP database, the JWT signing secret (→ forge an admin token), and the Gmail service-account private key.** A separate P1 punches a hole in the privacy premise from the other side: the outreach drafter sends raw email bodies to Claude with the NER backstop disabled, leaking any non-CRM names in cleartext. The project's own test suite is not green today — two thesis tests fail against the shipped v0.1.0:73 seed (stale assertions, not a runtime bug). Net: not ready to trust as "private" until the traversal and the outreach leak are closed; everything else is hygiene or polish.
+
+## Cross-referenced findings
+
+- **Path traversal (the P0).** Found independently by the **evaluator** (P1, `server.py:1717-1718`) and the **security-auditor** (P0, escalated because the same read also yields `/data/.crm-secret` → admin-token forgery and the Gmail DWD key). The **exerciser** ran a path-traversal probe and reported "blocked" — a **false negative**: `curl` collapses `../` client-side, so the test never sent a literal `..`. The two code-reading agents are correct (`get_path()`/`urlparse` does not normalize `..`; a raw client like `curl --path-as-is` reaches it). Merged as **one P0** with the auditor's higher severity, on the strength of the secret/key exposure.
+- **Two thesis tests fail.** The **evaluator** (P1) and the **exerciser** (P2) independently reproduced and root-caused the same failure: `test_thesis_seed.py` / `test_thesis_actions.py` assert the `positioning` variant_group has 2 members, but `ensure_positioning_framings` (added after the tests) seeds 5 more → 7. Merged as one **P2** (no runtime impact; but "all tests pass" is false today).
+- **`X-Forwarded-For` trusted** for rate-limit/ban keying — flagged by both **evaluator** (P3) and **auditor** (P3). One P3.
+- **CORS default `*`** — flagged by both **evaluator** (P3) and **auditor** (P3); both note it's benign today (Bearer auth, no cookies). One P3.
+
+## Priority queue
+
+- [P0] Unauthenticated path traversal in `/assets/` → arbitrary file read (LP DB, JWT secret → admin forgery, Gmail key); runs before auth — `backend/server.py:1717-1732`, `1671-1681` — evaluator + security-auditor (exerciser's "blocked" was a false negative)
+- [P1] Outreach drafter sends raw email bodies to Claude with NER backstop disabled (`Boundary(...)` built with no `ner_fn`), leaking non-CRM names — contradicts documented fail-closed design — `backend/mcp/outreach_agent.py:230`, `106-118`, `151-170` — security-auditor
+- [P1] Soft-deleted contacts and organizations remain fully readable by direct ID (`GET /api/{contacts,organizations}/{id}` omit `deleted_at IS NULL`) — exerciser
+- [P2] Two thesis tests fail against the shipped seed (stale 2-vs-7 member assertion) — `backend/test_thesis_seed.py:50`, `backend/test_thesis_actions.py:40` — evaluator + exerciser
+- [P2] Non-integer query params (`?limit=abc`, `?offset=abc`) raise unhandled `ValueError` and crash the request thread → connection reset, no error body (8 list endpoints) — exerciser
+- [P2] 5,383-line single handler class with 6 near-identical copy-pasted CRUD update blocks — raises cost of safe change — `backend/server.py:1523`, `2267-2293` — evaluator
+- [P2] Frontend loads React + Babel-standalone from `unpkg.com` with no SRI — offline-fragile (contradicts data-sovereignty premise) + unpinned supply chain + in-browser transpile cost — `frontend/index.html:9-11` — evaluator
+- [P2] TLS verification disabled on the scrub-gateway path (`CERT_NONE`, `check_hostname=False`); ships pre-wired, MITM exposes the LP-name dictionary if gateway backend enabled — `backend/ingest/http_util.py:11-17`, `docker_entrypoint.sh:86` — security-auditor
+- [P2] `cryptography==42.0.5` shipped in image carries a known bundled-OpenSSL advisory (used for Gmail RS256) — bump to ≥43 — `start9/0.4/Dockerfile:50` — security-auditor
+- [P2] `assets/ABOUT.md` is stale and user-visible on the box: claims first-boot seeds the volume (removed in v0.1.0:40) — fresh install shows an empty CRM with no explanation — `start9/0.4/assets/ABOUT.md:9` — start9-spec-checker
+- [P2] Hardcoded LAN IPs (`192.168.1.72` Spark, `192.168.1.87` Qdrant) compiled into the s9pk (16 occurrences) — network change forces edit+rebuild+reinstall — `start9/0.4/startos/actions/*.ts`, `docker_entrypoint.sh:85-87` — start9-spec-checker
+- [P3] `X-Forwarded-For` trusted verbatim for rate-limit/ban keying — spoofable to evade ban or poison another IP's bucket — `backend/server.py:1588-1592` — evaluator + security-auditor
+- [P3] CORS default `*` — benign with Bearer auth today, pin in prod — `backend/server.py:81` — evaluator + security-auditor
+- [P3] DB connections closed only on success paths (no `try/finally`) — hygiene, not a leak in practice — `backend/server.py` (many handlers) — evaluator
+- [P3] `get_body` reads `Content-Length` bytes with no max-size cap → memory-exhaustion DoS on write routes — `backend/server.py:1538-1554` — security-auditor
+- [P3] Container runs as root (no `USER` in Dockerfile) — drop privileges — `start9/0.4/Dockerfile` — security-auditor
+- [P3] `requirements.txt` lists ~12 vulnerable/unused deps (e.g. python-jose 3.3.0, CVE-2024-33663) — not imported at runtime, but a trap if ever installed — `backend/requirements.txt` — security-auditor
+- [P3] `gmailResult.gmail_url` rendered into an `href` without scheme validation (server-generated today) — `frontend/index.html:10134` — security-auditor
+- [P3] No uniqueness constraint on `contacts.email` — duplicates silently accepted — exerciser
+- [P3] Create-opportunity does not validate the `stage` field (only the PATCH stage route does) — arbitrary stage strings stored — exerciser
+- [P3] No server-side length limits on text fields (10k-char names accepted, stored, and exported) — exerciser
+- [P3] `POST /api/fundraising/log-communication` ignores a valid `investor_id` and demands `row_id`/`investor_name` — grid-vs-contacts model mismatch — exerciser
+- [P3] Deprecated `datetime.utcnow()` calls emit warnings (will break on a future Python) — exerciser
+- [P3] `npm audit`: fast-xml-parser/-builder transitive advisories (build-time only, via start-sdk) — `npm audit fix` — security-auditor
+- [P3] `anthropic` dependency unpinned — reproducibility/supply-chain gap — security-auditor
+- [P3] Manifest declares `aarch64` but no native arm image is built (runs via QEMU; `fastembed`/`mcp` unverified on arm) — drop the arch or build it — `start9/0.4/startos/manifest/index.ts:23` — start9-spec-checker
+- [P3] `start9/0.4/README.md:39-40` describes the removed seed mechanism (developer-facing) — start9-spec-checker
+- [P3] `packageRepo`/`upstreamRepo` manifest URLs 404 (private/nonexistent) — fine for private use, fails registry validation — start9-spec-checker
+- [P3] Stale `start9/0.4/javascript.tmp.1776377780/` build artifact on disk (gitignored, harmless) — start9-spec-checker
+
+## Scorecard
+
+The evaluator's six-lens table, unadjusted (other agents' evidence reinforces but does not contradict it):
+
+| Lens | Score /5 | Justification |
+|---|---|---|
+| Architecture | 3 | Clean module separation (`redaction`/`ingest`/`mcp`/`email_integration`) + consistent central dispatch, but all CRM logic in one 5,383-line handler — `server.py:1523`. |
+| Security | 2 | Strong auth crypto (pbkdf2 200k, `compare_digest`, pinned HS256) and clean secrets hygiene, undermined by the P0 traversal + the P1 outreach leak the auditor added — `server.py:1717`, `outreach_agent.py:230`. |
+| Performance | 4 | Per-request connections + WAL + 42 indexes + locked abuse state, right-sized for ~400 records / 5 users. |
+| Testing | 3 | 13 fast isolated tests covering the hard parts (redaction leak-hunts, grounding boundary), but 2 are red against the current seed. |
+| Code quality | 3 | Comments explain *why* well; the monolith and copy-pasted CRUD blocks raise change cost. |
+| Documentation | 5 | AGENTS.md + 6 scoped guides are accurate and verifiable (the "FastAPI is vestigial" claim checks out — zero runtime imports). |
+
+Note: the auditor's P0 escalation and added P1 leak both land on the Security lens; they corroborate the score of 2 rather than push it lower (it was already the floor of the table).
+
+## Disagreements & gaps
+
+- **Path traversal — exerciser vs. the two readers.** The exerciser reported path traversal "blocked"; the evaluator and auditor both found it exploitable by code reading. Resolution: the exerciser's tool (`curl`) normalized `../` before sending, so the probe never tested the vuln — it is **real and P0**. Lesson: black-box probes for traversal must use a raw, non-normalizing client (`curl --path-as-is`).
+- **Shared blind spot — the differentiating Phase-1 AI paths are unverified at runtime.** No agent could exercise live Claude/Anthropic calls (no `ANTHROPIC_API_KEY` in env), Qdrant ingest (`/api/index/*`), or Gmail draft creation. So `/api/outreach/draft`, `/api/architect/ground`, and thesis generation were reached only in their degraded/no-key form. The P1 outreach-leak finding is from code reading, not a live capture — confidence is high but a live request would confirm it. This is the one gap every agent shares.
+- **StartOS spec — two UNVERIFIED items.** The spec-checker could not confirm whether 0.4 still requires `instructions.md`/`prepare.sh` (docs pages 404'd) or measure the expanded image size; `start-sdk verify` failed only because the machine has the wrong-era (0.3.5) binary — `start-cli s9pk inspect` succeeds and the artifact is valid. No packaging blocker for private sideload.
+
+## Suggested order of work
+
+1. **Close the P0 traversal first** — `os.path.realpath` + `commonpath` containment check (or stdlib `translate_path`) on the `/assets/` branch in `backend/server.py:1717`, and confirm with `curl --path-as-is 'http://host/assets/../../data/crm.db'`. Until this is fixed the "private" claim is false; do not deploy.
+2. **Fix the P1 outreach leak** — pass `ner_fn=_ner_local` to the `Boundary` in `outreach_agent.py:230` (mirroring `architect_grounding.py`), fail closed if NER is unreachable, add a minimize-first pass. Both #1 and #2 directly protect the LP data the project exists to protect.
+3. **Make the test suite green and authoritative** — fix the two stale thesis assertions (assert structurally, not on an exact count) and add a one-line aggregate runner so "do the tests pass" has a single answer; then the suite can gate the next deploy.
+4. **Fix the two functional bugs** — add `AND deleted_at IS NULL` to the get-by-ID handlers (P1), and wrap query-param `int()` parsing to return a 400 instead of crashing the thread (P2).
+5. **Deploy-prep the package** — update `ABOUT.md` (and README) to current first-boot behavior so a fresh install isn't a mystery empty CRM, and lift the hardcoded Spark/Qdrant IPs into config/env before the box ever moves networks.
+6. **Then verify the live Phase-1 paths on the box** with a real `ANTHROPIC_API_KEY` — the outreach/architect/thesis features that no agent could exercise here.
+7. **Hardening sweep** (P3 batch) — bump `cryptography`, drop the vestigial vulnerable `requirements.txt` entries, stop trusting `X-Forwarded-For`, cap request-body size, run as non-root.

ROADMAP.md

@@ -12,9 +12,6 @@
   - `GET /api/fundraising/backups`
   - `GET/PATCH /api/fundraising/backup-policy`
   - `GET /api/fundraising/relational-summary`
-  - `GET /api/feature-requests`
-  - `POST /api/feature-requests`
-  - `PATCH /api/feature-requests/:id`
 - New DB tables:
   - `fundraising_state`
   - `fundraising_investors`
@@ -22,7 +19,6 @@
   - `fundraising_funds`
   - `fundraising_commitments`
   - `fundraising_views`
-  - `feature_requests`
   - `app_settings`
 - Grid saves/restores now sync into relational fundraising tables automatically.
 - Formula engine is now sandboxed (no `eval`/`new Function`) with expanded function support.
@@ -84,6 +80,444 @@
 3. Email/communication integrations (optional)
 4. Granular permissions (if team grows)
 
+## Backlog (post-Phase-1 agentic)
+
+### Data-model cleanups (deferred from the v0.1.0:104 session)
+
+- **Retire `contacts.contact_type`** (the Contacts Investors/Prospects tabs + TYPE badge). It's a legacy binary that's set mechanically — `'investor'` just means "exists in the grid" (stamped unconditionally by `_upsert_contact_from_fundraising`), `'prospect'` means "imported/added, not in the grid" — and is superseded by the grid-derived signals `contact_grid_signals()` already injects (`existing_investor`/`committed`, `pipeline_stage`). Plan: replace the tabs + TYPE badge with those signals, repoint the dashboard `total_lps`/`total_prospects` counts, then drop the column. Live UI change → its own small design pass. (Grant: "I want to delete it, next session.")
+
+- **Consolidate `contacts` ↔ `fundraising_contacts` into one linked model.** Goal (Grant): everyone in `contacts` maps to a `fundraising_investors` row (an individual maps to their own row). Today `contacts` is the canonical person directory (FK target for `communications`/`opportunities`); `fundraising_contacts.contact_id` (migration `0004`) points INTO it; the mobile Contacts page reads `contacts`. Three populations: **A** linked (grid pill ↔ contact), **B** `contacts`-only (imported prospects / manual adds — need a grid row), **C** pill-only (`fundraising_contacts.contact_id IS NULL` — need a contact row). **Census-first:** before designing any migration, count A/B/C on the box — Grant runs the SQL himself (he is **not** providing a DB copy), so hand him a counts-only script. The census decides whether this is a ~20-row cleanup or a ~300-row structural migration with `communications`/`opportunities` repointing. Then Grant reconciles B (add grid rows/pills) and C (add contact rows) and ensures all are linked. **(v0.1.0:105) A TEMPORARY admin census ships to read A/B/C off the box without shell access: `GET /api/admin/contacts-census` (`handle_contacts_census`) + a Settings → Admin "Run census" button, mirroring `backend/scripts/contacts_census.sql` (counts only). DELETE the endpoint + route + button after the numbers are captured — all tagged `TEMPORARY` in code.**
+
+### Captured tweaks (Matrix, 2026-06-18/20)
+*Small UI/UX + capture-quality items captured via Matrix; not yet scheduled.*
+
+- **[P2] Backup history (Settings) defaults minimized, chevron-expand, pinned to the bottom** — it's rarely viewed, so it shouldn't take prime space. Frontend-only (`frontend/index.html`). (2026-06-18)
+- **[P2] Preserve the active tab across a page refresh** — a refresh currently snaps back to the top/default tab. Persist the selected tab (e.g. localStorage / URL hash) and rehydrate on load. Frontend-only. (2026-06-18)
+- **[P2] Email capture matches an investor on `To:`/`From:` only, not `Cc:`** — today if an investor's address appears anywhere on a message landing in a team mailbox (including when a teammate is merely cc'd on an outbound reply to the investor), it logs a spurious "received from investor" entry. Restrict the investor-link match to the to/from headers so a cc doesn't create a phantom inbound note. `backend/email_integration/` matching (see `docs/guides/email.md`). (2026-06-20)
+
+### Follow-ups/reminders + NL search + bot grid-mutations (agreed plan, 2026-06-18)
+*Agreed with Grant 2026-06-18. Three workstreams, sequenced **W1 → W2 → W3**. **Overarching constraint (Grant):** the dominant risk is **leaking LP data (names, $, notes, contacts) to third-party LLMs — NOT write-safety.** A wrong number is recoverable; investor substance reaching Claude is not. Consequences: W2 keeps LP rows off Claude (only the question text + schema vocabulary leave the box; entity names resolved locally); W3 keeps bot mutation-parsing on local Qwen. Because this DB *logs* commitments/pipeline but doesn't move money, a bot mutation is low-stakes → **any team member may approve one in Matrix**; the guardrail is "the bot can't silently mass-change numbers," enforced by the per-mutation human approval gate, not a tight money gate.*
+
+**W1 — Reminders & follow-ups — BUILT + tested locally 2026-06-18 (v0.1.0:92, deploy pending).** First-class tickler tied to the grid: `reminders` table (in-app migration `0006`; logical FK to `fundraising_investors.id` + denormalized name, like `0005`), full CRUD (`GET/POST/PATCH/DELETE /api/reminders`; soft-delete; status open/done/snoozed/cancelled; assignee; `source` human/bot/automation), a read-only **derived `reminder_status` grid column** (overdue/due_soon/open — injected like `pipeline_stage`, **filterable so the follow-up view can later key off reminders instead of the binary `follow_up` checkbox**, per Grant), an orphan reconciler (`reconcile_grid_reminders` — cancels reminders when their investor leaves the grid, the pipeline reconciler's twin), a **Reminders** nav page (filter/complete/snooze/edit/delete + create), a Dashboard **"Reminders Due"** card, a **"Reminders due"** daily-digest section, and a per-investor **`last_activity_at`** recency rollup (the shared building block W2's "not nurtured" query needs). Pure local CRM — no LLM path, no leak surface. Tests: `test_reminders.py` + digest reminders test; **31/31 suite green, render-smoke green**. **Deploy:** needs an s9pk build + install (version bumped to 92); get authorization first.
+- **W1b (deferred fast-follow):** nurture-gap automation — a daily job flags "committed / in-pipeline + no activity in N days + no open reminder" → auto-suggests a reminder (`source='automation'`, human confirms). Build once the recency rollup is proven in practice.
+- **Left untouched (deliberate):** the grid `follow_up` checkbox + automation list-memberships, and `communications.next_action_date` + `/api/outreach/radar` — reminders are the new richer layer; folding those into it is a later cleanup, not now.
+
+**W2 — Natural-language query (read-only). BACKEND BUILT + tested + validated locally 2026-06-18; web/Matrix UI pending.** = the **"Email/communication search + NL query → item 3 (NL→safe structured query)"** below, now sequenced second and **redesigned** (see below). Subsystem detail: `docs/guides/nl-query.md`.
+
+- **Approach changed from the original "Claude behind redaction + a validated filter-AST" to LOCAL-ONLY + a named-intent catalog (decided with Grant 2026-06-18).** Rationale: (a) the dominant risk is LP data reaching a vendor — running translation on the **local Qwen via Spark Control** keeps the question on the box entirely (same basis as intake/digest), so there is **no Claude path and no redaction boundary** to manage, which is both simpler and safer; (b) a generic SQL/AST compiler was over-built for the real need — instead there are **~12 curated, hand-written, parameterized "named queries"** (`backend/nl_query/intents.py`) each with typed slots, and the **slot validator** (`runner.validate`) is the whole trust boundary (no dynamic identifiers, no raw SQL). The LLM only maps a question → `{intent, slots}`; its output is still validated, so a hallucinated intent is rejected. **Results never go to any model** (deterministic local render). Both design choices were pressure-tested by independent review agents before building.
+- **As built:** `backend/nl_query/` (`intents.py` catalog, `runner.py` validator/executor + audit, `translate.py` local-Qwen translator, `try_questions.py` dev harness). `POST /api/query/nl` (`{question}` or direct `{intent,slots}`) + `GET /api/query/catalog`, `require_bot_or_admin`, read-only, audited (`audit_log` `entity_type='nl_query'`). Soft-delete-correct per table (`fundraising_*` has no `deleted_at` — `graveyard` is the axis; emails via a live `email_account_messages` sighting; reminders/opps/comms via `deleted_at`). Builds on W1's `last_activity_at`. Tests: `nl_query/test_nl_query.py` + `test_translate.py` + `test_nl_query_endpoint.py` (34/34 suite green).
+- **Validation:** the local Qwen translated **12/12 of Grant's real example questions** correctly (right intent + slots, incl. "3 months"→90, sent/received→direction) against the live Spark — settles local-only; Claude not needed. Translation quality on messy/typo/no-match inputs shakes out in live use.
+- **Remaining:** **step 4** = web "Ask" box in the Communications tab (calls the endpoint, renders rows + the interpreted query); **step 5** = Matrix `@bot <question>` (thin client of the endpoint; the 2-admin review room means a full-book dump is acceptable, so no bulk-result cap — only a light anti-flood truncation). Reads need no approval gate. Then deploy with reminders (v92) as **v0.1.0:93**.
+
+**W3 — Bot grid-mutations behind a Matrix approval gate.** Generalize the email-proposal scaffold (`email_proposal_matrix` + propose→post→decide→apply) into one `agent_proposals` table (kind discriminator + JSON payload + target). Bot proposes set-commitment / assign-fund / change-stage / set-reminder; a human approves/edits/rejects in Matrix (**any member**); then apply. **Surgical, version-checked mutations — never blob RMW:** stage rides the existing `opportunities` link + validated stage endpoint; reminders write the W1 table; set-commitment/assign-fund need a version-checked single-cell upsert into the grid blob. Triggers the deferred **scoped service-token** item below (per-mutation-kind allowlist on the bot credential; money/merge/delete always require human approval regardless of scope — the autonomy axis). Parse on local Qwen, not Claude.
+
+
+### Matrix-bridge intake for the fundraising grid — M1+M2+M3 BUILT & LIVE
+*Requested 2026-06-16. **M1+M2 live since v0.1.0:86 (2026-06-17); M3 (business-card photo) shipped & live 2026-06-20** — code in `backend/matrix_intake/`, guide at `docs/guides/matrix-intake.md`. M3 unblocked once the Spark Control daily-driver model became vision-capable: the bot OCRs a card via Spark Control's `/v1/chat/completions` multimodal passthrough (same `CRM_CHAT_MODEL`), then runs the existing intake flow; captures contact name/email/title/city/LinkedIn/phone/mobile (server half of phone/mobile = s9pk v0.1.0:98). Remaining: ongoing on-device card testing (OCR accuracy on small-in-frame cards). Next major build is **Pipeline adoption** (see below).*
+
+Use the **matrix-bridge** repo's pattern to listen on a dedicated ten31-database Matrix room. Send a message (with an optional business-card photo) and a local LLM **via Spark Control** parses it into the fundraising-grid schema and **auto-creates the investor entity + contact row**. For an existing investor, send a meeting note and it **appends an interaction-log entry**. Approval gate: the bot replies in Matrix with the proposed add/edit; the user approves / rejects / edits in-thread before the write commits (keeps the draft→human-approve guardrail).
+- Fits the "grid is canonical" direction (writes land in `fundraising_*`) and the never-send-autonomously rule (in-thread human approval before any write).
+
+**Locked design (2026-06-16, approved) — build now, M1 then M2:**
+- **Separate component, shared scaffold:** new `backend/matrix_intake/` (its own process; lifts matrix-bridge's connect/prime-then-listen/threaded-reply plumbing). `matrix-nio` is isolated to this component's `requirements.txt` — it never enters the stdlib CRM runtime. Keeps the CRM write credential + LP data out of the general-purpose matrix-bridge bot (blast-radius + data-sovereignty), and lets the two iterate independently. Runs on the Spark (placement settled against `standards/guides/placement.md` at deploy).
+- **~~v1 = text-only~~ — M3 business-card photo SHIPPED (2026-06-20).** The Spark Control daily-driver model is now vision-capable (multimodal `image_url` passthrough), so card→text→fields works end-to-end. Transcribe-then-reuse (vision OCRs to text; the existing text extractor pulls fields) preserves the email/phone integrity rules. See the matrix-intake guide.
+- **Parse:** local Qwen via Spark Control `/v1/chat/completions` (temp 0, JSON-only), reusing the existing Spark client pattern (`backend/redaction`/`backend/ingest`).
+- **Approval handshake (the one stateful piece):** in-memory pending-proposal store keyed by Matrix thread root; user replies **yes / edit field=value / no** in-thread. Satisfies never-write-autonomously; exempt from "agents draft, humans send" (internal data entry, like the digest).
+- **CRM-side:** `POST /api/intake/investor` (service-auth) creates a new investor+contact **through the existing grid-save path** (so relational sync + audit + backup-on-write happen as with a UI edit; bot never does whole-blob RMW) or appends a meeting note to the interaction log for an existing investor; `GET /api/intake/match?q=` fuzzy-matches via the existing entity-resolution/email-matcher. New investor needs no fund at intake.
+- **Phases:** M1 = scaffold + parse + in-thread propose, **no writes** (proves Matrix↔Spark). M2 = intake endpoint + match + write-on-approve + tests. **M3 = business-card photo (SHIPPED 2026-06-20).**
+
+**Post-deploy enhancement — fuzzy match + in-thread confirm (Grant, 2026-06-17). DEPLOYED & LIVE 2026-06-17 (v0.1.0:86; box migration chain …85→86 clean, `candidates` endpoint verified); Matrix live-smoke pending.** Today `find_intake_match` is **exact-after-normalization** (`_normalize_text` = lowercase+strip), so near-misses — "Charlie" vs "Charles" (same last name), "Acme Capital" vs "Acme Capital LLC", a one-character email typo — return no match and the bot proposes a **new** investor, risking a duplicate the human approves without realizing a near-match exists. The existing in-thread approval gate is useless against this because the human is never *shown* the near-match. Fix: matcher returns **ranked fuzzy candidates** (deterministic pre-filter: normalized name similarity / token overlap + email edit-distance ≤ ~2), surfaced in-thread for the human to confirm or pick, with the **local Spark LLM optionally re-ranking/judging the shortlist** (good at Charlie/Charles + legal-suffix equivalence; fed only the shortlist, never the whole LP list). Keeps the approval gate but makes it effective against duplicates. Land **after** the live smoke — net-new logic + reply grammar + tests; the current exact match is safe and its failure mode (a duplicate) is recoverable via the existing entity-merge subsystem (`backend/entity_*.py`).
+- **As built:** `find_intake_candidates` in `server.py` (deterministic — stdlib `difflib` name similarity + token-set Jaccard, legal-suffix-aware via `_strip_legal_suffix`, + email Levenshtein ≤ 2; ranked, ≥0.62, top 5). `GET /api/intake/match` now returns `{match, candidates}`. Bot: a new `_stage="disambiguate"` shortlist (`proposals.render_disambiguation` / `interpret_disambiguation` / `attach_to_candidate` / `promote_to_new`) — human picks a number / `new` / `no`. **The optional LLM-judge re-rank was deliberately deferred** (the deterministic filter already surfaces the named cases; an LLM judge is the right *pruner* for shortlist noise — build if the deterministic ranking proves too noisy in practice). Tests: `test_intake_endpoints.py` (server fuzzy cases), `matrix_intake/test_proposals.py` (disambiguation grammar), `matrix_intake/test_crm_client.py` (candidate shape).
+
+**Post-deploy enhancement — conversational (LLM-mediated) edits (Grant, 2026-06-17). DEPLOYED & LIVE 2026-06-17 (bot-side; pulled + restarted on the Spark `modelo32`); Matrix live-smoke pending.** Today an in-thread correction uses a rigid grammar (`edit field=value`). Let a free-form reply that isn't `yes`/`no`/a literal `edit …` be treated as a natural-language revision instruction: send {current proposal + the instruction} back through local Qwen (`spark.py`, the same parse leg — no Claude, no scrub) and re-render the revised proposal card for approval (e.g. "add that we met on June 14" → updated Note). Keeps the draft→human-approve gate (the human still confirms the LLM's revision) and subsumes `edit field=value` as a deterministic fast path. Thread the instruction text into `normalize`'s source so the email-integrity rule still holds (a revised email must appear in the original message or the instruction). Pairs naturally with the fuzzy-match item above — build both as one conversational-UX pass after the smoke. (Parsing of free-form *intake* messages already works today via the Qwen parse leg; this item is specifically about the *edit/refine* turn.)
+- **As built:** `parse.revise` + `_apply_revision` (offline-testable; the approval-stage `else` branch in `bot.py` routes any non-yes/no/edit reply here). `parse_message` now stashes `_source_text` so revise can re-check email integrity against {instruction + original}; the model's email field is never trusted. No-op revisions are caught via `proposals.same_fields` (re-prompt, not a false "Updated"). **Known v1 limit:** revise edits fields but does not re-run the matcher on a mid-thread firm rename. Tests: `matrix_intake/test_parse.py` (revise merge + email integrity + match-id preservation).
+
+**Managed service — DONE (container) 2026-06-17; dashboard card deferred to a spark-control session.** The bot ran as a bare `nohup` process (silently died on a Spark reboot). Now it's a **docker-compose service** (`docker-compose.yml` at the repo root + `backend/matrix_intake/Dockerfile`; `restart: unless-stopped` → survives reboot; image bundles `backend/matrix_intake` + the stdlib `backend/ingest` Spark client; `.env` mounted read-only). Cutover done on the Spark (nohup stopped, container `matrix-intake` up + listening). **Still bare `docker`/SSH-managed** — a spark-control dashboard card (Update/Restart/Stop/Logs tile like `matrix-bridge`) is a separate task in the spark-control repo: see `docs/handoffs/add-intake-bot-to-spark-control.md`.
+
+**Parse mis-identifies the investor when the message names an internal teammate (found in live-smoke, 2026-06-17).** *"jonathan is chatting with wyoming soon about fund commitment"* → the bot picked **jonathan** (a colleague/CRM admin) as the investor and offered a Jonathan/Nathan fuzzy shortlist, when the investor is **Wyoming**. Root cause is upstream of matching: local Qwen has no notion of who's internal, and mis-read the sentence role. **Fix (cheap, high-confidence, near-term):** feed the parse prompt the ~5-person team roster + the frame *"messages are written by a team member about a prospect; a named team member is the person doing outreach, never the investor"* (roster from a config value or a small read — not the admin-gated `/api/users`, since the bot is a member). Offline-testable (stub the model). **Bigger design (deferred, needs more failure samples):** the user's idea of routing inputs through the LLM *with grid context* for entity resolution — feasible (local Qwen, same as the digest, never Claude) but feed a **bounded shortlist, not the full ~400-name grid** (a small model dilutes on a haystack); pairs with the deferred LLM-judge. Also exposes a missing concept: the **internal deal owner** (Jonathan), which the bot doesn't model. Get 3–5 more real intake messages before re-architecting; the roster fix lands regardless.
+
+**Long-term — extract the intake bot to its own repo (recommended, not yet done).** Containerizing from this monorepo is the pragmatic now-state, but the bot is a genuinely separate deployable (own process, own `matrix-nio` dep, own lifecycle); its only CRM coupling is the HTTP API (a clean network contract) plus ~40 lines of stdlib Spark client (cheap to vendor). The tell: the spark-control Update button would run `git reset --hard origin/main` on the **whole CRM clone** — wrong blast radius. `matrix-bridge` is already a dedicated repo; mirror it. The extraction is a migration (new Gitea repo, move code + tests + guide, vendor the client, re-point the Spark deploy), so it's deferred until worth the lift — do it *before* wiring the spark-control card if both land in the same push.
+
+### In-app camera business-card intake — DONE & live (v0.1.0:100, device-confirmed 2026-06-20)
+*Shipped: a camera button in the mobile top bar (left of the quick-log pencil) → take/choose a photo → vision-transcribe → parse → fuzzy-match → edit/approve/reject, surfaced as an inline mobile sheet (`source="app_card"`, form-field edits only, any-authenticated-member). The reusable core is nio-free and already reachable from the CRM (`server.py` imports `llm`; `matrix_intake/parse.py`+`spark.py` import no `matrix-nio`), so it landed as **one endpoint** (`POST /api/intake/card`) + **one mobile component** — no bot refactor / new dep / migration; reuses the New-investor sheet pre-filled + the `.quicklog-btn svg` icon-sizing fix. History: commits `463f624` (feature) / `622d454` (handoff). (Plan doc removed — git history is the record.)*
+
+### Scoped service-credential auth path for automated CRM writers
+*Surfaced 2026-06-17 while deploying the Matrix intake bot. **Decision: defer — the bot uses a dedicated member username/password for now.** The CRM has no API-key/service-token path; its only auth is username+password → JWT. A dedicated **member** login is appropriately scoped against what matters operationally (no admin: can't manage users, reset data, or change settings) and unblocks the live smoke today.*
+
+**Accepted residual risk (why this is worth revisiting):** a member credential is far broader than the bot's actual need (two endpoints: `GET /api/intake/match`, `POST /api/fundraising/log-communication`). A member can **read the entire LP/prospect database** — the exact data this system exists to keep off third-party servers — plus broad member-level *write* within the fundraising domain (could create/append on any investor). The credential lives in a `.env` on the Spark, so a Spark compromise leaks read-access to all LP data. Mitigating context: own-infra, LAN-local; the Matrix bot is the **first out-of-process API writer** (the digest runs in-process with direct DB access), so there is exactly **one** consumer today → building a token-scope framework now is premature (YAGNI).
+
+**Right long-term design:** a hashed, revocable **service token** with a per-route **scope allowlist** (intake-match + log-communication only), minted/revoked from the admin panel, replacing the bot's member login. Revocation then kills the token without rotating a reused human password.
+
+**Build trigger:** when a **second** out-of-process automated writer appears, OR before **any** automated writer is reachable beyond the LAN — whichever comes first. Build it once, properly, at that point.
+
+### Admin-only vs. all-users web-UI surface — audit
+*Requested 2026-06-16 (idea, P2).* Have the **explorer agent** report which web-UI functionality is visible only to admins vs. to all users (member role) — a map of the role-gated surface across `frontend/index.html` and the backend route auth checks. Useful input for the consolidation/permissions work.
+
+### Daily activity digest (email to the team)
+*Requested 2026-06-15. **Phase A deployed** (v0.1.0:76). **Phase B deployed & verified live in v0.1.0:77 (2026-06-16)** — digest content + Spark summarization + daily scheduler + by-investor section + admin-panel control + on-demand send. Auto-send defaults OFF until an admin enables it in Settings → Admin. **v0.1.0:83 (built, deploy pending): in-app windowed preview** — Settings → Admin builds a digest over a chosen window (last 24h or since a date) and shows it before sending (`POST /api/admin/digest/preview`), so the **real Spark summarizer can be verified on demand** even on a quiet day (the fixed last-24h `send-now` couldn't); manual send uses the same window and never touches the daily cursor.*
+
+**Decisions (locked 2026-06-15):** recipients = **all active admins**; summarization = **Spark-LLM narrative** (never Claude — un-anonymized substance stays local); granularity = **grouped by user** (→ per investor).
+
+**Send transport — DECIDED 2026-06-15: Gmail domain-wide delegation** (not SMTP). The box's existing service-account grant (which powers email capture) **includes `gmail.compose`**, which authorizes `users.messages.send` — verified by a token-mint probe **and a live `messages.send` to grant**. So the digest sends through the account the CRM already uses: **no app password, no new account, no admin change.** The narrow `gmail.send` scope is *not* granted, so the sender must request `gmail.compose`.
+
+**Phase A — DONE:** (v0.1.0:75) `configureDigestSmtp` Start9 action + `docker_entrypoint.sh` `SMTP_*` export + `backend/smtp_send.py` + admin `POST /api/admin/digest/test-email` (recipient-restricted to the admin set — not an open relay) + Settings button. (v0.1.0:76, redeploy pending) `backend/email_integration/gmail_send.py` (`users.messages.send` via DWD/compose) + `backend/digest_mailer.py` (**Gmail-DWD preferred, SMTP fallback**); the endpoint + button route through it; sender = `CRM_DIGEST_SENDER` else first active admin. Tests: `test_smtp_send.py`, `test_smtp_endpoint.py`, `test_gmail_send.py`.
+
+**Phase B — DONE (2026-06-15/16):** `backend/digest_builder.py` builds **two sections** — *by team member* (per-user **Spark** narrative + both directions, with a deterministic fallback) and *by investor* (team-wide, inbound + outbound, deduped per email, structured). Soft-delete filtered throughout. `backend/email_integration/digest_scheduler.py` is an always-on daily thread that re-reads a **DB-backed policy** each cycle and sends once/day at the configured hour to all active admins (window cursor in `app_settings`). Control moved out of env into the **admin panel**: `app_settings.digest_policy` + `GET/PATCH /api/admin/digest/policy` + a Settings → Admin **enable toggle + send-time dropdown** (env vars only seed the first-boot default). Plus admin `POST /api/admin/digest/send-now` + a "Send Digest Now" button. Decisions settled: **6 PM default**, **always-send** (empty-day note), **per-user narrative + by-investor section**, **in-app control** (not StartOS). Tests: `backend/test_digest_builder.py`. Detail: `docs/guides/email.md`.
+
+Have the CRM send a **daily digest email** summarizing each registered user's activity — primarily **who emailed which investors and the substance of those emails** — to the fund principal (and eventually other admins). Scales with the synced-user count: 2 users synced today, ~5 eventually.
+
+- **Source data:** the captured email-activity already flowing through the Gmail DWD propose→approve pipeline (`backend/email_integration/`), keyed per registered user → per investor/contact. Optionally fold in other CRM activity (audit feed, automation runs, new opportunities) later.
+- **Send path is NEW capability.** Today nothing leaves the box — the system only *captures* Gmail and *creates drafts*. This needs outbound SMTP. StartOS 0.4 has a system-wide SMTP account (since v0.4.0-beta.9): the user configures it once for the whole server and services read it via `sdk.getSystemSmtp(effects).const()`, which returns a `T.SmtpValue` (`host`, `port`, `from`, `username`, `password`, `security`). Wire the digest sender to that rather than hardcoding any account. *Implementation path (researched 2026-06-15, our SDK pin `^0.4.0-beta.66`):* model a `manageSMTP` action on [gitea-startos](https://github.com/Start9Labs/gitea-startos) / [vaultwarden-startos](https://github.com/Start9Labs/vaultwarden-startos) — a three-way `selection` (system / custom / disabled) built on `sdk.inputSpecConstants.smtpInputSpec`, persisted to `storeJson`, with `main.ts` injecting `SMTP_HOST/PORT/USER/PASS/FROM/SECURITY` env vars into the daemon `exec` block (same shape as the existing `setAnthropicApiKey.ts` action). The Python sender reads them via `os.environ` and opens `smtplib.SMTP`/`SMTP_SSL`. **"Custom SMTP" is a dedicated per-package account, fully independent of the server's system SMTP** — the custom branch never calls `getSystemSmtp`, so the digest can send through its own provider even on a box with no system account configured (confirmed in both reference packages). This is the likely fit here: a digest-only mailbox separate from anyone's Gmail. Note StartOS 0.4 dropped the old `Config`/`Properties` manifest spec — SMTP config is an **action + storeJson**, not a manifest config field. **SDK note (verified 2026-06-15):** our pin `^0.4.0-beta.66` resolves to exactly `0.4.0-beta.66` (caret on a prerelease stays within the `0.4.0` tuple), whose SMTP surface — `getSystemSmtp` → `T.SmtpValue {host, port, from, username, password, security}`, `inputSpecConstants.smtpInputSpec` (providers gmail/ses/sendgrid/mailgun/protonmail/other; selection disabled/system/custom), `smtpShape`, `smtpPrefill` — is **byte-identical** to the 1.5.3 reference packages (verified from published tarballs; repo `node_modules` is absent). Build against beta.66 as-is — **no SDK bump needed** (moving to 1.x is a major-track change with broad blast radius across `startos/`, and nothing about SMTP justifies it).
+- **Analysis runs on Spark, never Claude.** The digest is deliberately **un-anonymized** (real LP names + email substance), so any summarization/analysis must go through **Spark Control to local models** — this is the one path that intentionally bypasses the scrub→Claude→re-hydrate boundary, because keeping the substance local is the whole point. Never route digest content to Claude.
+- **Exempt from "agents draft, humans send."** That rule governs outward LP/prospect contact. This is an internal ops digest to the team's own inboxes — a different category — so an automated daily send here does not violate the draft-only guardrail. State this explicitly at build time.
+- **Scheduling:** a daily cron, naturally co-located with the existing `backend/email_integration/scheduler.py` sync cadence.
+- **Soft-delete:** every aggregate/read in the digest must filter `deleted_at IS NULL` (see the standing soft-delete rule).
+
+Open design questions (settled at build time): send time = **6 PM box-local** (configurable in the admin panel), covering the ~24h window up to send; empty days = **always send** with a "no activity" note; summary granularity = **one per-user narrative** plus a **by-investor structured section** (inbound + outbound, team-wide) added 2026-06-16; enable/time live in the **admin panel** (DB-backed), not StartOS actions.
+
+### Email/communication search + natural-language query
+*Requested 2026-06-16. Three increments, **sequenced 1 → 2 → 3** (1 and 2 first as a quick increment; 3 is a separate, larger build after). Origin: Grant asked whether we can query "emails sent to a specific investor" / "activity by user," and floated NL queries like "existing investors who have committed capital across our funds that we haven't emailed in a while."*
+
+**Status: items 1 & 2 SHIPPED in v0.1.0:83 (built + verified locally 2026-06-16, deploy pending).** The Communications tab now has the structured activity surface (item 1: typed/fixed investor dropdown, mailbox + direction + **date-range** filters, free-text, **click-to-expand full body** via `GET /api/email/detail`) and a **"Search content"** semantic mode (item 2: `GET /api/email/search` over the Qdrant email index). The dropdown-empty bug (the facet only listed grid investors) was the v83 fix — it now mirrors the list across grid/org/contact matches. **Item 3 (NL→SQL) remains** — the larger, separate build below. Detail: `docs/guides/email.md`.
+
+**Context — the data is captured but currently has NO front-end.** The entire Gmail email schema (`emails`, `email_threads`, `email_investor_links`, `email_account_messages`, `email_activity_proposals`, …) exists and is populated by the DWD capture pipeline, but is surfaced **nowhere** in `frontend/index.html` today (only as inputs to the daily digest). So all three items below are about making already-captured data queryable/visible. Email bodies of *matched* emails are already chunked + embedded into Qdrant with `{lp_id, lp_name, doc_type:"email", date_ts}` metadata.
+
+**Caveat that shapes all three — the two-model join.** "Emails to an investor" link to the **fundraising grid** (`email_investor_links.fundraising_investor_id`); "committed capital" lives in the grid too (`fundraising_commitments`, multi-fund). But manually-logged `communications` and `lp_profiles` (single-fund) live in the **classic** model, and the two models are only bridged by fuzzy email/name matching (no authoritative join key). Any query spanning "committed capital" + "email recency" must reckon with this. Prefer the grid side as the higher-signal source (matcher already does).
+
+**1. Activity query endpoints + panel — DONE (v0.1.0:83).** Delivered as the **Communications tab** rather than the originally-sketched `/api/activity` endpoints: `GET /api/email/activity` (`db.query_email_activity`) returns the actual records filterable by investor / mailbox / direction / **date range** / free-text, and `GET /api/email/detail` expands the full body. Answers "emails to investor X" and "what has mailbox Y sent" interactively. Soft-delete filtered throughout; investor identity is typed (`fund:`/`org:`/`contact:`) so org/contact-only matches resolve and are pickable. *(The `collect_user_activity()`/`collect_investor_activity()` digest helpers remain the by-user/by-investor pivot source; a dedicated per-user pivot UI was not needed for the answer Grant wanted, which the mailbox+direction filters already give.)*
+
+**2. Email content search box — DONE (v0.1.0:83).** A **"Search content"** toggle in the Communications tab → `GET /api/email/search?q=` wraps `backend/ingest/search.py:hybrid_search` filtered to `doc_type='email'`; hits are hydrated + soft-delete-filtered against SQLite (canonical) and link back to the full body. Semantic/lexical search over email *content* ("find where we discussed the mining deal"), distinct from item 1's structured filters. 503 (clean "unavailable") when Spark/Qdrant is unreachable.
+
+**3. Natural-language → safe structured query — SUPERSEDED & BUILT as W2 above (2026-06-18).** The design constraints below (especially "LLM = Claude behind the redaction boundary" and the validated-AST shape) were **revisited and changed** during the build: translation runs on the **local Qwen** (no Claude, no redaction), and the safe surface is a **named-intent catalog**, not a generic query AST. See the W2 entry above and `docs/guides/nl-query.md` for what shipped; the original framing is kept here for provenance. _An LLM translates a plain-English question into a **safe, read-only** DB query against the CRM, for relational/analytical questions that semantic search *cannot* answer — Grant's example ("committed across funds AND not emailed in a while") is joins + aggregates + recency, not a text-topic match. Original design constraints (locked at request time):_
+  - **LLM = Claude behind the redaction boundary** (better at text-to-SQL than local Qwen; the scrub→Claude→re-hydrate path already exists for the PII concern). Not Spark — Spark Control offers embeddings/rerank/RAG + local chat, but **no text-to-SQL**.
+  - **Safety is the hard part, not the parsing.** Do NOT hand the LLM open-ended SQL against the live DB (soft-delete leaks, injection, runaway scans). Constrain it: read-only connection/view, a curated/parameterized query surface or a validated query AST, soft-delete-filtered views, row/time caps. Treat as its own designed feature with its own tests.
+  - Must reckon with the two-model join caveat above (capital lives in the grid; recency from email links).
+
+### Consolidate on the fundraising grid as canonical; retire vestigial classic-CRM surfaces
+*Decided 2026-06-16. The CRM carries two stacked models: the original generic CRM (contacts / lp_profiles / opportunities / manual communications) and the fundraising grid + email capture. The team uses the grid; most classic surfaces are un-adopted (verified on the box: Pipeline + Communications empty, Contacts auto-populated from the grid). **Decision: the fundraising grid + email capture is the canonical system of record;** prune or repurpose the rest rather than maintain a parallel half-empty CRM.*
+
+**Retire `lp_profiles` + LP Tracker — DONE & deployed live (v0.1.0:78, 2026-06-16).** 21/21 backend tests green, `py_compile` clean; installed to the box (`installed-version`→`0.1.0:78`, migration chain …77→78 clean, server up on :8080).
+- Removed the orphaned `LPTrackerPage` component + the `lp-tracker`→`fundraising-grid` redirect (frontend).
+- Removed the `/api/lp-profiles*` endpoints (list/get/create/update) and their handlers, the unused `lp-breakdown` report + route, the contact-dossier LP display (frontend + the `lp_profile` block in `handle_get_contact`), and the demo-seed LP block.
+- **Dashboard KPIs repointed:** "Total Committed" now sums `fundraising_investors.total_invested` (the canonical grid rollup), **excluding graveyarded investors** so the headline reflects live committed capital — a deliberate divergence from `/api/fundraising/relational-summary`, which sums all rows. "Total Funded" dropped — the grid has no funded-vs-committed concept and the frontend never rendered it. (If a funded/wired status is wanted later, that's a new grid feature, not a revival of lp_profiles.) Regression-guarded by `test_dashboard_report.py`.
+- **Left in place (intentional):** the empty `lp_profiles` table + index (no destructive drop, per never-hard-delete); the contact-delete soft-delete cascade; the `--reset-all-data` clear; and the inert MOCK_MODE `mockDb.lp_profiles` fixtures (dev-only fallback, never hits the backend — its dashboard mock still reads mock lp_profiles, a known dev-only divergence from the real backend). Updated `test_soft_delete_reads.py` to drop the now-removed `lp_profile` assertions (kept its org `total_funded` opportunities-aggregate checks).
+
+**Adopt the Pipeline — wire it to the grid. — DONE: DEPLOYED & live-smoked 2026-06-18 (v0.1.0:88; migration chain …86→88 clean, `0005_grid_pipeline_link.sql` applied on the box, server up; the full +Pipeline → board → advance-stage → remove round-trip is verified on the box).** *(Was: second build after the Matrix-bridge intake.)*
+- Pipeline (`opportunities`) is fully built and functional but unused. Keep it: it's the one classic surface that tracks something the grid doesn't — a forward-looking deal funnel (stage, `expected_amount × probability`, owner, close date) vs. the grid's actual committed dollars + flags.
+- New idea (Grant, 2026-06-16): let users **flag an investor in the grid as a pipeline opportunity** (a grid column/control) so it **auto-creates / syncs an `opportunities` row** that loads into the Pipeline board. Design the grid↔pipeline link (which fund seeds it? what sets stage/expected amount? keep them reconciled). Turns Pipeline from a disconnected second data-entry surface into a view driven by the canonical grid.
+- Revisit the stray contact-create side-door (the "Create Opportunity" modal `POST /api/contacts`) once the grid-driven flow exists.
+
+**As built (decisions locked with Grant 2026-06-17):** UX = **row action + seed modal** ("Add to Pipeline" per grid row → captures primary contact / target fund / expected amount / stage / probability). The durable link is `opportunities.fundraising_investor_id` (**migration 0005**, additive + reversible); "is in pipeline?" / "what stage?" are **derived from a live opp join**, never a denormalized flag (no drift). **Ownership split:** the grid owns whether the link exists + the seed; the **board owns stage/probability/owner/close/next-step** — a grid save never reseeds a live opp (`POST /api/fundraising/pipeline/link` is idempotent: one live opp/investor, re-link returns the existing one). Contact is **reused from the grid's synced `fundraising_contacts.contact_id`** — the `POST /api/contacts` side-door is **gone**. Grid `lead` → opp owner (fallback acting user). Two **read-only** grid columns (Pipeline action + Pipeline Stage) injected on read; their row values are stripped on write so they never persist or dirty the autosave. **Remove from pipeline** (`POST .../unlink`) **soft-deletes the opp; the grid row is left fully intact** (Grant's explicit ask). Deleting an investor from the grid archives its orphaned opp (`reconcile_grid_pipeline_links`, called after `sync_fundraising_relational`). **Folded in:** the P2 soft-delete leak in `handle_pipeline_report` + dashboard pipeline aggregates (archived opps no longer counted). Tests: `backend/test_grid_pipeline_link.py` (link/idempotent/round-trip/guards/unlink-intact/re-link/orphan/aggregates), 28/28 suite green, render-smoke green. **Deploy:** server-side → needs an **s9pk build + install** (v87); get authorization first.
+- **Follow-up (v0.1.0:88, frontend-only, DEPLOYED & verified 2026-06-18):** retired the Pipeline page's **"+ New Opportunity"** button + its create-by-contact modal — an opportunity is now born **only** from a fundraising-grid investor row ("+ Pipeline"), matching how the team works (they live in the grid). The board is now a view + stage-management surface; button replaced with a muted "Add deals from the Fundraising Grid" hint. Removed the dead handler/state + the page's unused `/api/contacts` fetch.
+- **Deferred (not built):** no write-back of committed dollars into grid fund cells (grid stays canonical for committed $); a graveyarded investor with a live opp still shows its stage (deliberate — a live deal is a live deal).
+
+**Keep the Contacts table — as the read-only per-person directory it already is.** Confirmed 2026-06-16: the grid models **investor entity → many people** correctly today. The grid "contacts" column is a multi-pill editor; each pill syncs to a `fundraising_contacts` row AND its own classic `contacts` row (5-person family office → 1 investor + 5 contacts, linked via `fundraising_contacts.contact_id`, migration 0004). The Contacts page is **read-only for creation** (header: "added from the Fundraising Grid"; no New-Contact button), edit-only via the detail slide-over — the desired flow already holds. Email capture already rolls **multiple people up to one investor** (matcher indexes each pill's email separately, all → same `fundraising_investor_id`; `email_investor_links` records both investor and specific person). No build here — future email-surfacing UI should present comms grouped by investor across all its people.
+
+**Legacy-usage sweep — what's still wired, what's dead, what can be deleted (Grant, 2026-06-19; after mobile feature-complete).** Phase 7's `.badge-*` remap surfaced that several legacy badge classes (lead/outreach/meeting/due-diligence/committed/funded) have no — or near-no — live JSX caller, and Grant flagged other likely-dead legacy bits (e.g. the `investor type` field). Once the mobile redesign is feature-complete, do a systematic pass: for each legacy component / CSS class / endpoint / DB field, determine whether it still renders or runs anywhere (JSX callers, `do_GET`/`do_POST` dispatch, DB reads), then soft-retire/delete the genuinely-dead ones (never hard-delete data — archive per convention; `refactor-scout`/`janitor` can inventory). Start list: the orphaned `.badge-*` classes, `investor type`, and other classic-CRM remnants not reachable from the grid-canonical flow.
+
+### Front-end: pre-compile JSX, drop runtime Babel (optional, larger)
+*Logged 2026-06-16 during the v0.1.0:82 vendor+SRI work. The scoped fix shipped: React/ReactDOM/Babel are now vendored + SRI-pinned and served same-origin, and a jsdom render smoke check gates every build (`docs/guides/packaging.md`). This is the bigger alternative we deliberately deferred.*
+
+Today the app ships `@babel/standalone` (~3 MB) and transforms ~5k lines of inline JSX **in the browser on every page load**. A build step that pre-compiles the JSX to plain JS would (a) eliminate the runtime-transform blank-screen class entirely (no Babel in production), and (b) load much faster. **Cost:** it introduces a build step, which contradicts the current **"No build step"** convention (single `frontend/index.html`, inline-Babel React) — so this is a real architecture change, not a tweak. Weigh only if page-load size/latency or render robustness becomes a felt problem; the render-smoke gate already de-risks the status quo. If taken: keep the source `index.html` editable, emit a compiled artifact into the s9pk, and keep the smoke check pointed at the built output.
+
+### One-off feature batch (Grant, 2026-06-18)
+*Eight one-off ideas, triaged against the backend 2026-06-18. **Cross-cutting guardrail:** anything framed as "auto-add / auto-forward / auto-suggest" lands as a **proposal surfaced for human approval** (reuse the `email_proposal_matrix` propose→Matrix→decide rails), never a silent write — per "agents draft, humans approve." #1 is built (deploy pending); #6 is a spark-control task (→ INBOX); the rest are scoped backlog. #2/#4/#7 reuse existing rails (email-proposal loop + W2 NL-query) — they're "wire a new source into an existing pipeline," not greenfield.*
+
+- **1. Drag-reorder fundraising grid views — BUILT (frontend; deploy pending), 2026-06-18.** The sidebar view list is now drag-reorderable (HTML5 DnD mirroring the column-reorder idiom: `moveViewBefore` + `draggingViewId`/`dragOverViewId` in `frontend/index.html`). Order persists via the grid page's **existing autosave** (`views` is already in its snapshot + deps → `PUT /api/fundraising/state` → `views_json`), the same path rename/delete use — **no backend change.** Render-smoke green; the in-app drag interaction itself not yet browser-tested. **Known edge (same as existing rename/delete):** reordering while *off* the grid page only updates localStorage and is re-hydrated from the backend on next grid mount — reorder while viewing the grid. **Deploy:** needs an s9pk build + install.
+
+- **2. [P2] Suggest new contacts from digested emails (outreach detector).** When a captured *outbound* email goes to an address not already in `contacts`/the grid and looks like outreach, propose adding it as a contact. Hangs off the existing email capture + `email_proposal_matrix` / `/api/intake/email-proposals` review rails — net-new is the detector + "looks like outreach" criteria (exclude vendors / newsletters / internal domains). Lands as a **proposal**, not an auto-add.
+
+- **3. Pipeline stages + investor flags/labels — sharpened into a LOCKED SPEC (2026-06-19).** Was "new pipeline stages"; the design conversation collapsed it into a 4-stage per-investor funnel + auto-derived Existing-Investor flag + staleness overlay/nudge. **Full locked spec: see "Pipeline stages + investor flags/labels — LOCKED SPEC" below.**
+
+- **4. [P2] Squarespace website form-submissions → DB (near-term, high value).** Parse `form-submission@squarespace.info` capture emails — structured Name / Email / Company / LinkedIn / Location / comments (see the website-lead screenshots, Grant 2026-06-18) — and feed them into the proposal flow. Deterministic parser (fixed format) + existing proposal rails = relatively contained. **Guardrail:** despite the "auto-added" ask, land each lead as a **Matrix proposal → one-tap approve**, not a silent insert (same pattern as email proposals). Real leads (e.g. Matt Baas, Vikrum Tatla) are currently only living in an inbox.
+
+- **5. [P3] Matrix voice note → Spark Control transcription → intake.** matrix-nio receives an audio/voice event → download + decrypt the media → **Spark Control** transcription endpoint (Whisper-class — **confirm it exists; external dep**) → feed the text into the existing local-Qwen intake parse + disambiguation. Never call a Spark directly (Spark Control only). Larger; gated on the transcription endpoint existing.
+
+- **6. → INBOX (spark-control repo, not this one). Dashboard card for the crm/intake bot** (Update/Restart/Stop/Logs tile like `matrix-bridge`). Already noted under the Matrix-intake "Managed service" item + `docs/handoffs/add-intake-bot-to-spark-control.md`; captured to `standards/INBOX.md` to confirm/do in a spark-control session.
+
+- **7. [P2] Intake: "query the LLM when the name doesn't match."** Extend the disambiguation grammar (today: number / new / no — see screenshot) with a `search: <text>` option that runs the read-only **W2 NL-query** to locate the real existing investor when the typed name doesn't fuzzy-match a candidate. Builds on the existing NL-query + intake rails; keeps the human approval gate.
+
+- **8. [P2] Email capture learns from approve/reject (scope down to rules v1).** Use the already-logged approve/reject decisions to pre-suggest a decision. **v1 = deterministic, not ML:** detect `List-Unsubscribe` / `Precedence: bulk` (newsletters) + a learned denylist of rejected sender addresses/domains → pre-mark / auto-suggest reject (e.g. recurring non-investor newsletters). Don't build a classifier until the rules prove insufficient.
+
+### Pipeline stages + investor flags/labels — LOCKED SPEC (Grant, 2026-06-19)
+*Sharpened from the inherited 6-stage funnel (lead/outreach/meeting/due_diligence/committed/funded) over a design conversation 2026-06-18/19. Supersedes one-off batch item #3. **Locked — ready to build on green-light.** Grounding (verified): the grid's only labeling today is 3 boolean flags (priority/follow_up/graveyard) + a derived longshot + the `lead` **owner** column; there is **no investor type field**; "existing investor" is implicit in `total_invested > 0`; the 6-stage pipeline lives on classic `opportunities` and only applies to rows explicitly "+Add to Pipeline"'d; saved views are driven off the flags, not stage.*
+
+**Conceptual frame — three orthogonal axes (were conflated):**
+- **A. Relationship** — existing-LP vs prospect → collapsed to a single **auto-derived "Existing Investor" flag** (below). No prospect/lead/advisor sub-types: leads become prospects fast, and there are no advisors in this grid.
+- **B. Disposition flags** — keep **Priority** (the focus set) + **Graveyard** (truly dead). **Drop Longshot** — labeling something longshot is already half-giving-up, overlaps graveyard, and doesn't earn a third bucket. Everything not Priority/Graveyard is the neutral middle.
+- **C. Pipeline stage** — the active-raise funnel (below), per-investor.
+
+**1. Funnel = 4 stages, per-investor, terminal at Commitment:** `Lead → Engaged → Diligence → Commitment`
+- **Lead** — identified + first contact (cold outreach, a logged first meeting, or a website inbound); one-directional so far.
+- **Engaged** — a **two-way** conversation exists (they replied / there's a back-and-forth). *(Boundary confirmed with Grant: two-way, not "a second person at the firm.")*
+- **Diligence** — substantive: follow-up calls/meetings or data-room access.
+- **Commitment** — terminal. On commit → hand off to fund admin + record the $ in the grid fund cell; the pipeline's job is done.
+- **No Funded** (fund admin owns post-commitment; the Existing-Investor flag is effectively the "closed" signal). **No Meeting** (an activity, not a position). **No Lost** stage (the Graveyard flag covers dead).
+- **Start at any stage** — a known LP re-solicited for a new fund drops straight into Engaged/Diligence, not Lead.
+
+**2. "Existing Investor" = auto-derived flag** from `total_invested > 0`, injected read-only like `pipeline_stage` (never a maintained column); rendered as a star/indicator (esp. mobile). Orthogonal to stage — a re-solicited LP shows the star **and** a live stage at once. Lifecycle: prospect runs Lead→…→Commitment → $ recorded in the grid cell → they light up as an Existing Investor.
+
+**3. Staleness — a derived overlay on the stage + a Matrix nudge, NEVER an auto-demotion.** Governing principle: **derive-and-display freely; mutate state only via a human.**
+- A quiet deal does **not** change stage. Staleness shows on the **last-contact recency value** (the grid row's / mobile card's "2d ago"): light-grey when fresh → **amber → red** by days since `last_activity_at`, appending "stale" once it crosses the threshold (e.g. "35d stale"). The stage chip stays clean; the warning rides the recency line. The **same `last_activity_at` source drives the desktop grid and the mobile card**, so both color-code automatically.
+- **Why not auto-flip off Engaged/Diligence:** it re-couples axes B+C, silently destroys information ("stalled mid-diligence" vs "never engaged"), is a silent un-approved mutation (against the human-in-the-loop guardrail), and creates a perverse "log junk to stay alive" incentive.
+- The "auto" part is the **nudge = W1b nurture-gap** (see the W1/W2/W3 backlog; this refines its target set to **Engaged/Diligence**, not Commitment): daily job flags "in pipeline (Engaged/Diligence) + no activity > threshold + no open reminder" → bot **suggests a reminder**, a human confirms → re-engage (logging a comm resets `last_activity_at`) **or** consciously graveyard. The system nudges; the human acts. Deals never silently fall off.
+- **Stale threshold: ONE global threshold (locked 2026-06-19).** Not stage-aware for v1 (Diligence-trips-faster was considered and deferred). Pick the amber/red day-counts at build.
+- **Stale-as-a-view:** also a saved grid view keyed on `last_activity_at` (e.g. >90d, not graveyarded) — distinct from the per-stage overlay; both reuse `last_activity_at`, no new field.
+
+**Accepted tradeoff (per-investor, not per-fund — Grant's call):** re-soliciting an existing LP for a new fund reuses their single opportunity (set fund + reset stage) — you won't see "Funded Fund I / Diligence Fund III" as two simultaneous pipeline entries. The grid's per-fund $ columns remain the record of which funds an investor is in; the pipeline shows only the *current* raise. (Per-fund stage was considered and deferred as a bigger build.)
+
+**Concrete change set (cost asymmetry: labels/overlays are cheap; the enum is the one-time expensive bit):**
+1. **Enum:** `PIPELINE_STAGES = ['lead','engaged','diligence','commitment']` (`server.py:1833`) + the ~8 mirror sites: report ordering CASEs (`server.py:3782/3859`), `nl_query/intents.py:34/37`, frontend kanban (`index.html:4168`, mock `:2174`), opp-form `<option>`s (`:7732`), and the `'funded'/'lost'` filters in `total_funded`/`pipeline_value` (`server.py:2721/3766/3877`).
+2. **Data migration** of existing `opportunities.stage`: `outreach,meeting→engaged`; `due_diligence→diligence`; `committed,funded→commitment`. Reconcile the stray `lost` value (not in the settable enum) to graveyard-flag semantics.
+3. **Existing-Investor flag:** derive from `total_invested > 0`, injected read-only (grid column + mobile star).
+4. **Drop Longshot:** remove the derived `longshot_followup` + its deprecated view filter.
+5. **Staleness overlay:** green/amber/red on the injected `pipeline_stage` by `last_activity_at`, + the stale saved view.
+6. **Nudge:** specialize **W1b** to Engaged/Diligence in-pipeline deals.
+
+Items 3–6 are cheap (derived/read-time/frontend, reuse `last_activity_at`, no migration); items 1–2 are the deliberate one-time enum + migration.
+
+**Card presentation (mobile + grid, locked 2026-06-19):**
+- **Stage chip** = one of the 4 stages, shown **only when the row is in the pipeline** (most grid rows aren't — no chip / a faint "+ Pipeline" affordance otherwise).
+- **Top-right corner** = the **Priority** disposition only (star/pill when flagged, empty otherwise). Graveyard rows live in the Graveyard view / render muted — not a corner badge.
+- **Existing Investor** (auto-derived, `total_invested > 0`) = its own distinct indicator (star by the name or a left accent — **not** a per-card banner; keep it restrained per `design/DESIGN.md`).
+- **Last-contact recency** carries the staleness color (grey→amber→red, "Nd stale").
+- This **replaces the design-mockup's INVESTOR/PROSPECT category chip** — we have no prospect/investor *type*; that two-value badge was the tool deriving committed-$>0, which is exactly our Existing-Investor flag. Feeds `design/BRIEF.md` §3a.
+
+### Mobile-first implementation — backlog (design landed 2026-06-19)
+*The `/design` round-trip is complete: the contract now describes the mobile-first system
+(`design/DESIGN.md` §8 + the `mobile` token group), provenance + per-surface interaction model
+are in `design/_imports/2026-06-19/`, and the input brief is `design/BRIEF.md`. This is the gap
+between that contract and the current desktop-only `frontend/index.html` — the implementation
+backlog. **Scoped 2026-06-19 (plan below); not yet started.**
+The comps are signed-off prototypes, **not drop-in** (Claude Design runtime, seed data) — each
+surface is re-authored in the app's React idiom and wired to the **real API**.*
+
+#### Mobile PWA — installable home-screen app — BUILT 2026-06-20 (deploy pending)
+
+**Option A (iPhone-first, no service worker).** Makes the app installable to the iOS home
+screen and launch **standalone** (full-screen, no Safari chrome, dark themed status bar,
+splash). Shipped: `frontend/manifest.webmanifest` (`display:standalone`, `start_url:/`,
+`theme_color`/`background_color` = the brand base `#0b1118` already reserved for this in
+`design/tokens.tokens.json`); square icons generated from `ten31-app-icon.svg` (full-bleed
+`#0b1118` + white "T31", maskable-safe) → `icon-192.png`/`icon-512.png`/`apple-touch-icon.png`
+(180); `<head>` gains `rel=manifest`, `theme-color`, the `apple-mobile-web-app-*` metas
+(status bar `black` — opaque, so content never slides under the notch), `apple-touch-icon`,
+and `viewport-fit=cover` (so the tab bar's existing `env(safe-area-inset-bottom)` clears the
+home indicator). One pre-auth backend route serves `/manifest.webmanifest` as
+`application/manifest+json` (`backend/server.py`); icons serve via the existing `/assets/`
+handler. **No service worker** — on iOS the install prompt doesn't exist regardless (A2HS is
+always manual via Share), standalone display needs none, and a cache-first SW would reintroduce
+the stale-shell class the render-smoke gate guards against. Verified: render-smoke green +
+live-curl (manifest + icons 200 pre-auth, correct content-types). **Deploy:** ships in the
+next s9pk with the mobile phases.
+- **Known minor:** the iOS status bar is fixed `black` at launch (can't follow the in-app
+  light/dark toggle); a barely-perceptible seam vs the `#0b1118` app. Acceptable; dark is default.
+- **Deferred (not needed for iPhone):** a network-first service worker → Android's "Install"
+  prompt + faster relaunches; the JSX-precompile build-step (ROADMAP below) is the better lever
+  if relaunch speed is ever a felt problem.
+- **Adjacent issue (not PWA, noted while here):** a phone in **landscape** can exceed the 768px
+  breakpoint and render the *desktop* layout; `orientation:portrait` in the manifest hints at
+  this but iOS ignores it for home-screen apps. Revisit if it bites during device-testing.
+
+#### Phase 8 — conform to the FINAL Claude Design mockups (mobile) — **NEXT SESSION (scoped 2026-06-19)**
+
+*Phases 0–7 built the mobile surfaces + light theme. Phase 8 closes the gap to the **final** design + functional parity. Two independent agent passes ran 2026-06-19 (functional-parity + visual-conformance); their findings + the source-of-truth correction below drive this plan.*
+
+**⚠️ Anchor on the `.dc.html` prototypes at their DEFAULT props — NOT the screenshots.** The Claude Design export (`design/_imports/2026-06-19_zip-file/Venture-CRM mobile redesign/`) ships parameterized `*.dc.html` prototypes whose `data-props` **defaults are the landed decisions** (`variant:compact`, `theme:dark`, `font:plex`, **`lpFlag:earmark`** — see `GridApp.dc.html:320`). The PNG `screenshots/` are **option-history** (rejected/stale prop combos: INVESTOR/PROSPECT disposition badges, a 6-stage MEETING/FUNDED funnel, the star flag). **The per-surface build reference is `design/phase8-conformance.md`** — a re-anchored visual-conformance spec (anatomy + deltas + line refs, built against the `.dc.html` defaults); it supersedes the original screenshot-anchored pass. **The functional-parity agent report stands** (built from `store.js` + `.dc.html` wiring, not screenshots). *(General learning promoted to `standards/guides/design.md` Phase C.)* **Per-surface existing-LP signal: Grid + Pipeline = earmark corner-triangle** (Grant override 2026-06-19 — the dc used ★ for Pipeline; unify on the earmark) **· Contacts = avatar ring.** **Contacts (decided):** drop the investor/prospect type tabs (prospect type unused), but **keep a Priority-flag sort** as an enhancement beyond the dc.
+
+**Authoritative final investor card** (from `GridApp.dc.html:84–105`, the `<sc-for>` card; supersedes the screenshot "3-zone" card): card = `--panel` bg, 1px `--border`, radius 10, padding 12×14, `--shadow-card`, gap 8. **Existing-LP = an earmark corner-triangle** top-left (18px `--accent` dog-ear via the border trick), *not* a star, *not* a left-border. Row 1: investor **name** (16px/600, left) · **PRIORITY** badge right (mono 10px pill, **only if flagged — no INVESTOR/PROSPECT disposition badge**). Row 2: committed **`$amount`** (mono 15px, left) · **4-stage pipeline-stage chip** (mono 11px pill, `lead/engaged/diligence/commitment`). Row 3: **recency** "2d ago" (mono 12px). **No contact-name footer band** (that was an older screenshot variant). The current app card is close — 8a is mostly: swap the existing-LP signal to the earmark corner-triangle, move priority to the right-side PRIORITY badge, confirm `$` placement.
+
+**Sequenced plan (each = a reviewable commit; lands on `main` deploy-pending; Grant device-tests the whole set at the end):**
+- **8a — Card re-author (Grid + Contacts)** to the spec above + earmark; **reconcile `design/DESIGN.md` §8 + tokens to this final card** (the contract may still describe the older card). Closes visual G1/G2/C2 (re-anchored).
+- **8b — Detail-surface actions:** Contacts detail → bottom-sheet w/ copy-email, **Log communication**, **Open-in-Grid**, org/stage/committed summary; Pipeline detail → **comms timeline + "+ Log"**; add Open-in-Grid cross-nav to Reminders too. Closes funct GAP 2/3/5, visual C1/P1.
+- **8c — Global quick-log pencil** in the mobile top bar (Grid + Contacts) → pick investor/contact → `log-communication`. Closes funct GAP 1, visual S3.
+- **8d — Sort controls** (Grid + Pipeline) at Claude-design detail (name/stage/amount/staleness/priority). Closes funct GAP 4, visual G4.
+- **8e — Reminders parity:** grid reminder = **read/edit/clear/create** the investor's existing reminder w/ date presets; swipe-snooze → **snooze sheet** w/ presets (not fixed +7d); **investor picker** on the Reminders-tab create sheet — **easy: the `reminders` table already has `investor_id` and `POST /api/reminders` already accepts it (server.py:3761); the standalone create sheet just omits it** (NOT a backend limit — the functional agent mislabeled it); due-chip pill + urgency dot + "Today" bucket. Closes funct DIV 1/2/3, visual R1/R2/R4/R5.
+- **8f — Pipeline card + nav:** labelled `‹ Lead`/`Engaged ›` footer buttons + ★/recency on card; **clickable pager dots**; per-stage segmented tint. Closes visual P2/P3/P4, funct pager-dots. *(Pipeline **accordion mode** explicitly **NOT wanted** — Grant: a prototype variant we never used.)*
+- **8g — Add-investor stage + priority:** the create sheet sets an initial pipeline stage + priority (create → `pipeline/link` at chosen stage; priority on the row). **Grant-approved product decision.**
+- **8h — Loose ends:** grid create-sheet **duplicate matches tappable** → open existing investor; **logging a "Note" resets staleness** (Grant-approved — confirm `log-communication` recency rule does this); remove/handle the **dead `'snoozed'` status** (snooze = keep `open` + push due_date by design, yet still in the edit dropdown).
+- **8i — Shell:** bottom-tab **SVG line icons** (replace emoji glyphs) + the **`·Ten31·` wordmark** in the top bar (currently shows the page title). Closes visual S1/S2.
+
+**Effort note:** almost all of 8a–8i is **client wiring against endpoints that already exist** — only 8g (create→link) and 8e (reminder fields) touch create/link flows, both already-built endpoints. No new backend tables. Verify each phase with the jsdom/CSSOM harness + a commit. To see the prototypes rendered (optional): `support.js` needs `window.React` + a local server — render `*.dc.html` at default props (don't trust the screenshots).
+
+**Prerequisite — inline-style→CSS migration: SCOPED 2026-06-19 — much smaller/divisible than
+the "~1,300 inline styles" framing suggested.** Ground truth from `index.html`: **370** total
+`style={{}}` objects (not 1,300), against an existing **1,861-line `<style>` block** (with
+`:root` vars + ~all the `.nav-item`/`.sidebar`/`.table` classes + **4 media queries already**,
+incl. a `min-width` one) and **1,088 `className=` usages** — the app is already majority
+class-based. Two consequences:
+- **The responsive migration that gates mobile is only ~114 inline styles**, confined to the four
+  mobile surfaces + shell: FundraisingGrid **70**, Reminders **18**, Contacts **17**, Pipeline **7**,
+  App shell **2**. The other **240** inline styles live on desktop-only pages (Settings 104,
+  Outreach/Email/Status 57, Thesis 44, Comms 31, Dashboard 4) that are **absent on mobile**, so they
+  never block it. → **Not a monolithic blocker; it divides per-surface** and folds into each surface's
+  build (no upfront sweep).
+- **Two separable axes, not one.** (1) *Responsive* = layout-bearing inline styles → CSS classes +
+  `min-width` queries (the ~114 above; gates mobile layout). (2) *Theming* = inline **hex → `var()`**
+  so `[data-theme="light"]` can re-bind them — **183 hex literals** in the JSX region, app-wide but
+  mechanical (precedent: the design guide's inline-hex→`var()` field notes); gates the **light theme**
+  only. Sequence them apart.
+
+**Data-layer dependency — the locked pipeline-stages/flags spec** (see the section above) lands
+**first, standalone (Phase 0 below)**: the mobile cards render the 4-stage chip, the auto-derived
+Existing-Investor star, and the staleness overlay, all of which need the stage enum + migration +
+`total_invested>0` derivation + the `last_activity_at` ramp. Building the cards before the data
+layer means hardcoding against a model that's about to change.
+
+**Implementation plan (sequenced; decisions confirmed with Grant 2026-06-19) — fold the per-surface
+migration into each surface's build, behind one shared foundation step. No upfront sweep.**
+
+- **Phase 0 — Pipeline-stages/flags data layer — BUILT + tested locally 2026-06-19 (deploy pending).**
+  The locked spec above. **Shipped:** enum → `['lead','engaged','diligence','commitment']`
+  (`server.py`) + all mirror sites (report CASEs/filters, `total_funded`→`commitment`,
+  `nl_query/intents.py`); reversible migration **`0007_pipeline_stages_v2`** (outreach/meeting→engaged,
+  due_diligence→diligence, committed/funded→commitment, stray `lost`→archived; up+down verified on
+  synthetic data — the live DB has 0 opps so it's a real no-op there); backend injection of
+  `existing_investor` (`total_invested>0`), `last_activity_at`, and `staleness` (`''`/`aging`≥30d/
+  `stale`≥60d, boundaries inclusive) into the grid GET + stripped on write (`_computed_row_values` +
+  frontend `stripComputedRows`); frontend enum sites (Pipeline board, opp-form, mock) + a 4-stage
+  `pipeline_stage` chip with DESIGN tints. **Drop Longshot (spec item 4) was already done** by prior
+  cleanup (vestigial empty column + strip code) — left as-is (still cleans legacy blobs). Tests:
+  `test_pipeline_stages_v2.py` (migration remap + derivation values/boundaries) + updated
+  `test_grid_pipeline_link`/`test_soft_delete_reads`/`nl_query`; **36/36 suite green, render-smoke
+  green, fresh-DB migrate clean**. **Deferred to Phase 3 (co-lands with the mobile cards, where the
+  card design specifies them):** the *visible* desktop rendering of the existing-investor star + the
+  staleness-colored recency column + the seeded "Stale" saved view — the data is injected and
+  test-locked now, so Phase 3 is pure frontend. W1b nudge specialization is a separate fast-follow.
+  **Deploy:** needs an s9pk build + install (**authorize first**).
+- **Phase 1 — Shared mobile foundation — BUILT 2026-06-19 (deploy pending).** Shipped: `:root` mobile
+  vars (`--mobile-tab-bar-h`/`--mobile-touch-target`/`--mobile-input-h`/`--mobile-sheet-radius`/screen-pad +
+  fonts + `--text-subtle`/`--border-strong`); CSS for the safe-area-aware **`.bottom-tab-bar`**, the
+  **`.bottom-sheet`/`.sheet-scrim`/`.sheet-handle`** primitive (styling), and `.mobile-only`/`.desktop-only`
+  utilities — all `display:none` on desktop so **zero desktop change**; the **4-tab bottom bar**
+  (Grid·Pipeline·Reminders·Contacts → `setPage`) + a **mobile account/logout popover** wired into `App`
+  (sidebar already CSS-hidden <768px). Render-smoke green. **Deliberately deferred:** (a) the
+  **`<BottomSheet>` React component + `useIsMobile()` hook** → Phase 2, designed against their first real
+  consumer (no dead code); (b) the **13→15px type bump is per-surface**, not a global body rule — `body`
+  has no base font-size, so it lands as each surface is re-authored (Phases 2–5); (c) the
+  `[data-theme="light"]` block → Phase 6 (dead without the toggle). Browser-interaction (the bar on a real
+  phone) untested, like view-reorder.
+- **Phase 2 — Contacts (pattern-validator spike, BEFORE the Grid) — BUILT 2026-06-19 (deploy pending).**
+  Read-only A–Z directory (sticky letter headers, sorted/sectioned by last name) + segmented
+  All/Investors/Prospects tabs + pinned search → **full-screen read-only detail** (`.fs-detail`, promotes
+  the slide-over: contact info w/ tap-to-copy email, opportunities, communication history) → **sort
+  BottomSheet** (the sheet primitive's first, read-only consumer: Name A–Z / Z–A / Recently-contacted —
+  restores the column-sort the card list loses). Proves the list→detail→sheet pattern + per-surface
+  migration mechanics on the lowest-risk surface before the crux. *(Reordered ahead of the earlier "Grid
+  first" draft.)* **Lands the shared primitives** (deferred from Phase 1, designed against this first
+  consumer — no dead code): **`<BottomSheet>`** (scrim/Escape/**pointer drag-to-dismiss**, mount enter/exit
+  animation, built on the Phase-1 `.bottom-sheet` CSS) + **`useIsMobile()`** (768px `matchMedia`; surfaces
+  swap via a rules-of-hooks-safe wrapper — `ContactsPage` → `Desktop`/`MobileContactsPage`, **zero desktop
+  change**). This surface's **15px body bump** lands on `.mobile-screen`. Writes: **none** — Contacts is
+  read-only on mobile per `BRIEF.md` §3b (create/edit live on the Grid). Grew the `:root`/mobile var set
+  (`--bg-input`, `--accent-light`, mobile card/control radii + card/screen/detail-title fonts) per DESIGN §9.
+  Verified: render-smoke green + a throwaway jsdom interaction harness (mounted the real app at 375px,
+  stubbed `/api/contacts` — list/grouping/sort-sheet/detail/back all asserted, 14/14). **No browser/real-phone
+  check yet** (same deferral as Phase 1 + view-reorder). **Deploy:** folds into the next s9pk build.
+- **Phase 3 — Fundraising Grid (the crux). P3a BUILT 2026-06-19 (deploy pending); P3b (name/pill edit) deferred.**
+  Split confirmed with Grant 2026-06-19: P3a ships the readable + already-write-supported surface now;
+  editing an existing investor's **name + contact pills** is **P3b** (needs a new narrow per-row PATCH +
+  a pill-editor UI — `log-communication` can't rename/edit pills, and the whole-grid PUT is forbidden on
+  mobile).
+  - **P3a (built):** lean **`MobileFundraisingGrid`** (separate component — the desktop grid's debounced
+    whole-grid-PUT autosave would race on every mobile edit, so it's NOT reused; `FundraisingGridPage` is
+    now a `useIsMobile()` wrapper → `Desktop`/`Mobile`, desktop untouched). Card list over the **active
+    view** (ported the desktop view-filter predicate — graveyard/follow-up/lead flags + columnFilters — to
+    a shared pure helper so it can't drift), tappable view-name → **view-picker sheet**, search, the locked
+    **card model** (name · committed $ via `formatMoneyMobile` · stage chip · staleness-colored recency ·
+    Existing-Investor left-accent · Priority corner; graveyard muted). Full-screen detail (read-only:
+    commitments/funds, contact pills, notes) + **edit sheets**: **log a note** (`log-communication`),
+    **pipeline stage** (linked → `PATCH /api/opportunities/{id}/stage` via the new injected `opportunity_id`;
+    unlinked → `pipeline/link` then it; + remove-from-pipeline), **set a reminder** (`POST /api/reminders`),
+    and **`+ New` investor** (`log-communication` + `create_investor_if_missing`, client-side dedup
+    typeahead). **Never whole-grid `PUT /state`.** Backend: one small hook — read-only **`opportunity_id`**
+    injected into grid rows (`opportunity_id_by_source_row`, added to both strip points), so the detail can
+    PATCH the linked opp directly. Tests: `test_grid_pipeline_link` extended (opp_id inject/strip/round-trip),
+    36/36 green; render-smoke green; a throwaway stateful jsdom harness drove the real surface at 375px
+    (view filter, picker, detail, stage-PATCH, log-note, reminder, create+dedup — 18/18). **No real-phone
+    check yet** (same deferral as P1/P2). **Deploy:** folds into the next s9pk.
+  - **P3b (deferred):** `POST /api/fundraising/update-row` (version-safe single-row name/contacts mutation,
+    +test) + the bottom-sheet **pill editor** (add/edit/remove pills, client-side dedup). Then name + pills
+    become editable on an existing investor, completing BRIEF §3a's editable set.
+- **Phase 4 — Pipeline — BUILT 2026-06-19 (deploy pending).** Lean **`MobilePipeline`** (separate
+  component; `PipelinePage` is now a `useIsMobile()` wrapper → `Desktop`/`Mobile`, desktop kanban
+  untouched, just renamed `DesktopPipelinePage`). **Swipe-between-stages:** a count-forward segmented
+  stage control (`.pipeline-seg`) + a horizontal **CSS scroll-snap** container of four full-width stage
+  pages (`.pipeline-swipe`/`.pipeline-stage-page`) + page dots; tapping a segment scrolls to its page,
+  scrolling syncs the active segment/dots. Each card shows opp name · contact·org · expected $, with
+  per-card **‹/› stage move** (`PATCH /api/opportunities/{id}/stage`, disabled at the lead/commitment
+  boundaries) — the kanban "advance" without opening the detail. Tap a card → full-screen `.fs-detail`
+  (read-only `OpportunityDetailPanel`-equivalent fields via `MobileDetailRow` + a `StageChip`) with a
+  **stage-picker `BottomSheet`**. **Opp-centric** (operates on the same `opportunities` rows + stage
+  endpoint as the desktop board and the Grid detail's stage edit), amounts read-only; **no Existing-Investor
+  star** (opps carry `fundraising_investor_id` but not `total_invested`). Removal/deletion stays on the
+  desktop board + the Grid detail's "remove from pipeline" — the Pipeline tab is **view + advance-stage
+  only**. A `reviewer` pass was applied (reset the detail's stage-sheet open-state on back; `moveStage`
+  awaits the PATCH). Verified: render-smoke green + a throwaway jsdom 375px harness drove the real surface
+  (seg counts, stage pages, segment/dot sync, ‹/› move re-bucketing, detail + stage-sheet PATCH, back — 12/12).
+  No real-phone check yet (same deferral as P1–P3a). Reuses the P2/P3a primitives directly; **no backend
+  change.** **Deploy:** folds into the next s9pk.
+- **Phase 5 — Reminders — BUILT 2026-06-19 (deploy pending).** Lean **`MobileReminders`**
+  (`RemindersPage` is now a `useIsMobile()` wrapper → `Desktop`/`Mobile`; the desktop page renamed
+  `DesktopRemindersPage`, otherwise untouched). **Urgency-grouped list** over `/api/reminders`
+  (Overdue → Due soon → Later → Done → Cancelled buckets via `reminderBucket`; group headers carry the
+  overdue-red/due-soon-amber tint) with a compact **Active/Done/All** segmented filter + **`+ New`**.
+  Each row is a **`ReminderRow`** pointer-drag swipe (own per-row drag state): **swipe-left → mark done**,
+  **swipe-right → snooze +7d** (threshold 70px; snooze keeps status `open` and pushes `due_date`, mirroring
+  the desktop's "no wake mechanism" rationale), a **tap → create/edit `BottomSheet`** (title · due date ·
+  investor *(create-only free-text label — PATCH can't change investor, matching the backend)* · assignee
+  *(if `/api/users` is readable)* · details · status *(edit-only)* · Delete). Vertical-dominant drags release
+  to list scroll; non-swipeable (done/cancelled) rows stay tap-to-edit. Added `formatDueShort`/`reminderDueDelta`
+  (local-midnight delta — the desktop `formatDate` mis-renders FUTURE dates). A `reviewer` pass was applied
+  (**`pointercancel` no longer fires a spurious mark-done** — the key fix; stray drag on a non-swipeable row
+  recovers as a tap; cancelled gets its own bucket header). **No backend change.** Verified: render-smoke
+  green + a throwaway jsdom 375px harness (grouping/counts, swipe done + snooze PATCH, tap→edit prefilled,
+  create POST, Done-filter reload — 12/12). No real-phone check yet (same deferral as P1–P4). **Deploy:**
+  folds into the next s9pk.
+- **Phase 6 — Light theme + toggle — BUILT 2026-06-19 (deploy pending).** App-wide light theme
+  behind a `:root[data-theme="light"]` switch; **dark stays the default** (a pre-paint boot script
+  in `<head>` reads `localStorage.venture_crm_theme`, no `prefers-color-scheme` auto-switch). Toggle
+  is **app-wide**: a labeled control in the desktop sidebar footer + a sun/moon icon in the mobile
+  top bar, both driven by one `theme` state in `App` (single source of truth). Authoritative color
+  pairs came from the **full Claude Design export** (`design/_imports/2026-06-19_zip-file/`, incl. the
+  previously-missing `store.js` + the four `*App.dc.html` `DCLogic` palettes) — every stage/recency/
+  note/priority/reminder/money tint uses the designer's exact dark+light values, not guesses.
+  - **Method (zero dark-mode regression by construction):** grew `:root` to a full token set (44
+    themed color slots) whose **dark values equal the originals byte-for-byte**, so `var(--x)`
+    resolves identically in dark — verified (16 exact-match vars asserted == their original hex).
+    Migrated **319 hex literals → `var()`** across the JSX inline region *and* the `<style>` block
+    (a context-aware script for the unambiguous structural ones; targeted edits for the
+    context-dependent semantic/chip helpers — `StageChip` is now className-based off
+    `.stage-chip--{stage}`, `PIPELINE_STAGE_CHIP` deleted). All mobile surfaces + chrome are fully
+    var-based → **mobile light is complete**.
+  - **Known light rough edges (desktop only — for the conformance pass below, NOT mobile):** a
+    handful of bespoke `<style>`-block desktop shades (login glow/gradients `#101926`/`#4a9adf`,
+    scrollbar `#36506a`, desktop table row-hover `#172435`, dashboard KPI green `#10b981`) and the
+    legacy off-palette `.badge-*` family (`#ec407a`/`#ff9800`/etc., partly vestigial old-stage
+    badges) stay dark-tuned in light; **shadows** also stay dark-tuned (not yet tokenized). A few
+    desktop one-off shades were *consolidated* to the nearest token (small intentional dark deltas:
+    e.g. `#2a3a4d`→border, `#d9a15f`→due-soon, `#f3b2b2`→danger-text) — enumerate when polishing.
+  - **Verified:** render-smoke green (Babel transform + jsdom mount); a throwaway jsdom interaction
+    harness mounted the **authed shell** and exercised the toggle (boot-defaults-dark → click→light
+    +persist+relabel → click→dark, 7/7); theme parity + dark-identity + no-undefined-var checks all
+    green. **No real-phone / real-browser check yet** (same deferral as P1–P5 — verify on a device).
+    **Deploy:** folds into the next s9pk.
+
+- **Phase 7 (NEW) — Full design-conformance pass against the complete Claude Design export
+  (`design/_imports/2026-06-19_zip-file/`).** Per Grant (2026-06-19): make sure *everything we
+  implement* matches what he built in Claude Design — **all buttons, colors, spacing, and
+  functionality**, across all four surfaces + the light theme. Concretely: (1) run `design-checker`
+  now that the surfaces exist; (2) reconcile remaining drift (the P6 desktop light rough edges
+  above — bespoke `<style>` shades, the legacy `.badge-*` family, themed shadows; plus any
+  button/interaction deltas vs the `*App.dc.html` comps + the ~25 `screenshots/`); (3) re-pull
+  anything still cloud-only. This is the "conform to the design" sweep that the per-surface builds
+  (P2–P6) deferred.
+
+**Note on `design-checker`:** not run for this round-trip — it audits *existing* UI conformance,
+and the desktop UI still conforms to §1–7 (unchanged). The mobile gap is greenfield
+implementation (captured here), not conformance drift, so there's nothing for it to flag yet; run
+it after the mobile surfaces exist.
+
 ## Definition of done for "Airtable substitute" v1
 - Team can manage all investors in one master table
 - Saved views replicate current Airtable workflows

backend/digest_builder.py

@@ -0,0 +1,483 @@
+"""Daily activity digest — content builder (Phase B).
+
+Assembles the per-user -> per-investor email-activity digest and summarizes each
+team member's day with ONE narrative paragraph from the LOCAL Spark model
+(ingest/llm.py via Spark Control). NEVER Claude: the digest is deliberately
+un-anonymized (real LP names + email substance), so every summarization stays on
+Ten31 infra. Keeping the substance local is the whole point — this is the one
+path that intentionally bypasses the scrub -> Claude -> re-hydrate boundary.
+
+This is an internal ops email to the team's own admins, so it is exempt from the
+"agents draft, humans send" rule — that rule governs outward LP/prospect contact,
+not an internal digest to the fund's own inboxes. Never extend it to send to LPs.
+
+Soft-delete: every read here filters the relevant tombstones —
+`email_account_messages.deleted_at IS NULL`, `users.is_active = 1`, and the
+org/contact name joins drop soft-deleted rows (falling back to the raw address).
+
+Stdlib only; the local-LLM client is imported lazily so this module stays
+importable (and testable with an injected chat fn) without Spark configured.
+"""
+import json
+import os
+import sqlite3
+from datetime import datetime, timedelta, timezone
+
+
+# One row per (account-sighting x investor-link) in the window. Grouped into
+# per-user buckets in Python. Investor display name resolves fundraising grid ->
+# organization -> contact -> the raw matched address, skipping soft-deleted
+# org/contact rows (fundraising_investors has no soft-delete column — it is a
+# rebuilt projection of the grid).
+_ACTIVITY_SQL = """
+SELECT
+    ea.user_id        AS user_id,
+    u.username        AS username,
+    u.full_name       AS full_name,
+    ea.email_address  AS account_email,
+    eam.is_sent       AS is_sent,
+    e.id              AS email_id,
+    e.from_email      AS from_email,
+    e.subject         AS subject,
+    e.body_text       AS body_text,
+    e.snippet         AS snippet,
+    e.sent_at         AS sent_at,
+    COALESCE(
+        NULLIF(TRIM(fi.investor_name), ''),
+        NULLIF(TRIM(o.name), ''),
+        NULLIF(TRIM(COALESCE(c.first_name, '') || ' ' || COALESCE(c.last_name, '')), ''),
+        eil.matched_address
+    )                 AS investor_name
+FROM email_account_messages eam
+JOIN email_accounts ea          ON ea.id = eam.account_id
+JOIN users u                    ON u.id  = ea.user_id
+JOIN emails e                   ON e.id  = eam.email_id
+JOIN email_investor_links eil   ON eil.email_id = e.id
+LEFT JOIN fundraising_investors fi ON fi.id = eil.fundraising_investor_id
+LEFT JOIN organizations o       ON o.id = eil.organization_id AND o.deleted_at IS NULL
+LEFT JOIN contacts c            ON c.id = eil.contact_id      AND c.deleted_at IS NULL
+WHERE eam.deleted_at IS NULL
+  AND u.is_active = 1
+  AND e.is_matched = 1
+  AND e.sent_at >= ? AND e.sent_at < ?
+ORDER BY u.full_name, u.username, e.sent_at ASC
+"""
+
+_RULE = "─" * 52
+_FOOTER = ("— Internal Ten31 CRM digest. Narratives are generated locally (Spark), "
+           "never Claude. This mailbox is unmonitored.")
+
+_SYSTEM = (
+    "You write a brief internal activity digest for a venture fund's partners. "
+    "Given one team member's emails with investors over the last day, write 2-4 "
+    "sentences summarizing what they did: which investors they engaged and the "
+    "gist of each thread. Name the investors. Past tense, plain prose, no "
+    "greeting, no bullet points, no sign-off."
+)
+
+# Reminders due is a current-state addendum (what needs action now), NOT bound to the
+# email-activity window — a 6 PM digest should surface what's overdue / due today.
+# status='open' only: a 'snoozed' reminder is an explicit mute, so it stays out of the
+# digest by design (the quick-snooze UI keeps a reminder 'open' with a pushed-out date).
+_REMINDERS_SQL = """
+SELECT r.title         AS title,
+       r.due_date      AS due_date,
+       r.investor_name AS investor_name,
+       COALESCE(NULLIF(TRIM(u.full_name), ''), u.username) AS assignee
+FROM reminders r
+LEFT JOIN users u ON u.id = r.assignee_id
+WHERE r.deleted_at IS NULL
+  AND r.status = 'open'
+  AND r.due_date IS NOT NULL AND TRIM(r.due_date) != ''
+  AND substr(r.due_date, 1, 10) <= ?
+ORDER BY r.due_date ASC
+"""
+
+
+# ------------------------------------------------------------------ collection
+
+def _fetch_activity_rows(conn, since_iso, until_iso):
+    """Raw (sighting x investor-link) rows for the window. [] if email tables absent."""
+    try:
+        return conn.execute(_ACTIVITY_SQL, (since_iso, until_iso)).fetchall()
+    except sqlite3.OperationalError:
+        return []  # email tables not present (integration disabled) — nothing to report
+
+
+def _own_addresses(conn):
+    """Lower-cased set of enrolled mailbox addresses — used to decide whether an
+    email is outbound (from us) or inbound (from the investor) at the email level."""
+    try:
+        return {(r[0] or "").lower().strip()
+                for r in conn.execute("SELECT email_address FROM email_accounts")}
+    except sqlite3.OperationalError:
+        return set()
+
+
+def collect_user_activity(conn, since_iso, until_iso):
+    """Return per-user activity buckets for emails in [since_iso, until_iso).
+
+    Each bucket: {user_id, username, full_name, account_email, emails[], investors[],
+    sent, received, total}. Empty list if the email tables are absent. Only users
+    who had activity appear. Direction here is per-mailbox (eam.is_sent): did THIS
+    user send the message."""
+    rows = _fetch_activity_rows(conn, since_iso, until_iso)
+    groups = {}
+    for r in rows:
+        uid = r["user_id"]
+        g = groups.get(uid)
+        if g is None:
+            g = {"user_id": uid, "username": r["username"], "full_name": r["full_name"],
+                 "account_email": r["account_email"], "_emails": {}, "_inv": set()}
+            groups[uid] = g
+        eid = r["email_id"]
+        em = g["_emails"].get(eid)
+        if em is None:
+            em = {"email_id": eid,
+                  "direction": "sent" if r["is_sent"] else "received",
+                  "subject": r["subject"], "sent_at": r["sent_at"],
+                  "text": r["body_text"] or r["snippet"] or "", "investors": []}
+            g["_emails"][eid] = em
+        inv = (r["investor_name"] or "").strip()
+        if inv:
+            if inv not in em["investors"]:
+                em["investors"].append(inv)
+            g["_inv"].add(inv)
+
+    out = []
+    for g in groups.values():
+        emails = sorted(g["_emails"].values(), key=lambda e: e["sent_at"] or "")
+        sent = sum(1 for e in emails if e["direction"] == "sent")
+        out.append({
+            "user_id": g["user_id"], "username": g["username"],
+            "full_name": g["full_name"], "account_email": g["account_email"],
+            "emails": emails, "investors": sorted(g["_inv"]),
+            "sent": sent, "received": len(emails) - sent, "total": len(emails),
+        })
+    out.sort(key=lambda x: (x["full_name"] or x["username"] or "").lower())
+    return out
+
+
+def collect_investor_activity(conn, since_iso, until_iso):
+    """Re-pivot the same window by investor (across the whole team), deduping each
+    email so a reply to several team members counts once. Direction is decided at
+    the EMAIL level: outbound if the sender is one of our mailboxes, else inbound.
+
+    Each bucket: {name, emails[{email_id, direction in/out, subject, sent_at,
+    members[]}], inbound, outbound, total}. Sorted most-active first."""
+    rows = _fetch_activity_rows(conn, since_iso, until_iso)
+    own = _own_addresses(conn)
+    groups = {}
+    for r in rows:
+        name = (r["investor_name"] or "").strip() or "(unmatched)"
+        g = groups.get(name)
+        if g is None:
+            g = {"name": name, "_emails": {}}
+            groups[name] = g
+        eid = r["email_id"]
+        em = g["_emails"].get(eid)
+        if em is None:
+            outbound = (r["from_email"] or "").lower().strip() in own
+            em = {"email_id": eid, "direction": "out" if outbound else "in",
+                  "subject": r["subject"], "sent_at": r["sent_at"], "members": []}
+            g["_emails"][eid] = em
+        # Attribute the sending team member on outbound mail (the sighting with
+        # is_sent=1); inbound is "from them", so no member shown.
+        if em["direction"] == "out" and r["is_sent"]:
+            who = (r["full_name"] or r["username"] or "").strip()
+            if who and who not in em["members"]:
+                em["members"].append(who)
+
+    out = []
+    for g in groups.values():
+        emails = sorted(g["_emails"].values(), key=lambda e: e["sent_at"] or "")
+        inbound = sum(1 for e in emails if e["direction"] == "in")
+        out.append({"name": g["name"], "emails": emails,
+                    "inbound": inbound, "outbound": len(emails) - inbound, "total": len(emails)})
+    out.sort(key=lambda x: (-x["total"], x["name"].lower()))
+    return out
+
+
+def collect_due_reminders(conn, today_iso):
+    """Open reminders due on or before `today_iso` (overdue + due today), soft-delete
+    filtered. Returns [{title, due_date, investor_name, assignee, overdue}] sorted soonest
+    first. Empty if the reminders table is absent (feature not migrated on this box)."""
+    try:
+        rows = conn.execute(_REMINDERS_SQL, (today_iso,)).fetchall()
+    except sqlite3.OperationalError:
+        return []
+    out = []
+    for r in rows:
+        due = str(r["due_date"] or "")[:10]
+        out.append({
+            "title": (r["title"] or "").strip(),
+            "due_date": due,
+            "investor_name": (r["investor_name"] or "").strip(),
+            "assignee": (r["assignee"] or "").strip(),
+            "overdue": bool(due and due < today_iso),
+        })
+    return out
+
+
+# ------------------------------------------------------------------ policy
+
+DIGEST_POLICY_KEY = "digest_policy"
+DEFAULT_DIGEST_POLICY = {"enabled": False, "send_hour": 18}
+
+
+def load_digest_policy(conn):
+    """Resolve the live digest policy. Precedence: the app_settings DB row (the
+    admin-panel control) wins; absent that, the CRM_DIGEST_ENABLED/SEND_HOUR env
+    vars seed a first-boot default; absent those, DEFAULT_DIGEST_POLICY. Returns
+    {enabled: bool, send_hour: int 0-23}. Shared by the server (API) and the
+    scheduler so both read one source of truth."""
+    pol = dict(DEFAULT_DIGEST_POLICY)
+
+    env_enabled = os.environ.get("CRM_DIGEST_ENABLED")
+    if env_enabled is not None:
+        pol["enabled"] = env_enabled.lower() in ("1", "true", "yes", "on")
+    env_hour = os.environ.get("CRM_DIGEST_SEND_HOUR")
+    if env_hour:
+        try:
+            pol["send_hour"] = min(23, max(0, int(env_hour)))
+        except ValueError:
+            pass
+
+    try:
+        row = conn.execute(
+            "SELECT value_json FROM app_settings WHERE key = ?", (DIGEST_POLICY_KEY,)).fetchone()
+    except sqlite3.OperationalError:
+        row = None
+    if row:
+        try:
+            saved = json.loads(row["value_json"])
+        except Exception:
+            saved = None
+        if isinstance(saved, dict):
+            if "enabled" in saved:
+                pol["enabled"] = bool(saved["enabled"])
+            if "send_hour" in saved:
+                try:
+                    pol["send_hour"] = min(23, max(0, int(saved["send_hour"])))
+                except (ValueError, TypeError):
+                    pass
+    return pol
+
+
+# ------------------------------------------------------------------ window
+
+# Cap a manual/preview window so an admin can't accidentally fire a build over
+# years of history — each active user in the window costs one Spark call. ~3
+# months covers any realistic "since last quarter" preview.
+MAX_WINDOW_DAYS = 92
+
+
+def resolve_digest_window(*, hours=None, since=None, now_local=None, now_utc=None):
+    """Resolve a digest content window to (since_iso, until_iso) as UTC ISO-8601.
+
+    `until` is always now. The start is driven by exactly one of:
+      - since: a local calendar date 'YYYY-MM-DD' -> that day's local midnight
+      - hours: a positive integer lookback (the default path; 24 when nothing given)
+    `since` wins if both are supplied. The span is clamped to MAX_WINDOW_DAYS and
+    the start must be strictly before now. Raises ValueError on malformed input so
+    the caller can return a clean 400. Pure (now_* injectable) for testing.
+
+    Used by the admin-panel preview and manual-send — neither advances the daily
+    cursor, so a wide window here never suppresses the scheduled digest."""
+    nu = (now_utc or datetime.now(timezone.utc)).astimezone(timezone.utc)
+    nl = now_local or datetime.now().astimezone()
+    floor = nu - timedelta(days=MAX_WINDOW_DAYS)
+
+    if since not in (None, ""):
+        try:
+            d = datetime.strptime(str(since).strip()[:10], "%Y-%m-%d")
+        except ValueError:
+            raise ValueError("since must be a date in YYYY-MM-DD form")
+        start = d.replace(tzinfo=nl.tzinfo or timezone.utc).astimezone(timezone.utc)
+    else:
+        h = 24 if hours in (None, "") else hours
+        try:
+            h = int(h)
+        except (ValueError, TypeError):
+            raise ValueError("hours must be an integer")
+        if h < 1:
+            raise ValueError("hours must be a positive integer")
+        start = nu - timedelta(hours=h)
+
+    if start >= nu:
+        raise ValueError("window start must be before now")
+    if start < floor:
+        start = floor  # clamp to the max span (the response echoes the real window)
+
+    fmt = "%Y-%m-%dT%H:%M:%SZ"
+    return start.strftime(fmt), nu.strftime(fmt)
+
+
+# ------------------------------------------------------------------ summarization
+
+def _default_chat(prompt, system=None, max_tokens=220):
+    """Lazily reach the local Qwen chat via Spark Control (ingest/llm.py)."""
+    import os
+    import sys
+    sys.path.insert(0, os.path.join(os.path.dirname(os.path.abspath(__file__)), "ingest"))
+    import llm  # noqa: E402
+    return llm.chat(prompt, system=system, max_tokens=max_tokens)
+
+
+def _user_email_block(group, max_emails=20, body_chars=500):
+    lines = []
+    for em in group["emails"][:max_emails]:
+        invs = ", ".join(em["investors"]) or "(unmatched)"
+        body = " ".join((em.get("text") or "").split())[:body_chars]
+        line = f"- [{em['direction']}] {invs} | subject: {em.get('subject') or '(none)'}"
+        if body:
+            line += f" | {body}"
+        lines.append(line)
+    return "\n".join(lines)
+
+
+def _fallback_narrative(group):
+    """Deterministic summary when the local model is unavailable — the digest
+    must still send (always-send) with real counts rather than fail."""
+    name = group.get("full_name") or group.get("username") or "Team member"
+    invs = ", ".join(group["investors"]) or "no matched investors"
+    return (f"{name} had {group['total']} email(s) "
+            f"({group['sent']} sent, {group['received']} received) with {invs}. "
+            "(Local summary unavailable.)")
+
+
+def summarize_user_day(group, chat_fn=None):
+    """One narrative paragraph for a user's day, from the local model. Falls back
+    to a deterministic count summary on any error or empty reply."""
+    fn = chat_fn or _default_chat
+    name = group.get("full_name") or group.get("username") or "The team member"
+    prompt = f"Team member: {name}\nEmails:\n{_user_email_block(group)}"
+    try:
+        out = fn(prompt, system=_SYSTEM, max_tokens=220)
+        out = " ".join((out or "").split()).strip()
+        if out:
+            return out
+    except Exception:
+        pass
+    return _fallback_narrative(group)
+
+
+# ------------------------------------------------------------------ composition
+
+def _parse_iso(iso):
+    if not iso:
+        return None
+    s = str(iso).strip()
+    for fmt in ("%Y-%m-%dT%H:%M:%S.%fZ", "%Y-%m-%dT%H:%M:%SZ", "%Y-%m-%dT%H:%M:%S"):
+        try:
+            return datetime.strptime(s, fmt).replace(tzinfo=timezone.utc)
+        except ValueError:
+            continue
+    return None
+
+
+def _fmt_local(iso):
+    """UTC ISO -> human local time, e.g. 'Jun 17 2:14 PM'. Manual 12h formatting
+    to stay portable (no platform-specific %-I)."""
+    dt = _parse_iso(iso)
+    if dt is None:
+        return str(iso)
+    dt = dt.astimezone()
+    hour12 = dt.hour % 12 or 12
+    ampm = "AM" if dt.hour < 12 else "PM"
+    return f"{dt.strftime('%b')} {dt.day} {hour12}:{dt.minute:02d} {ampm}"
+
+
+def _reminders_section(due_reminders):
+    """Render the 'reminders due' block (overdue + due today). An empty list renders
+    nothing, so a clear deck adds no noise to the digest."""
+    if not due_reminders:
+        return []
+    overdue = [r for r in due_reminders if r["overdue"]]
+    due_today = [r for r in due_reminders if not r["overdue"]]
+
+    def _line(r):
+        inv = f"{r['investor_name']} — " if r["investor_name"] else ""
+        who = f"  [{r['assignee']}]" if r["assignee"] else ""
+        return f"  • {inv}{r['title']} (due {r['due_date']}){who}"
+
+    L = ["", _RULE, f"REMINDERS DUE ({len(due_reminders)})", _RULE]
+    if overdue:
+        L += ["", f"Overdue ({len(overdue)}):"] + [_line(r) for r in overdue]
+    if due_today:
+        L += ["", f"Due today ({len(due_today)}):"] + [_line(r) for r in due_today]
+    return L
+
+
+def _compose_body(user_groups, investor_groups, narratives, since_iso, until_iso, due_reminders=None):
+    title_date = datetime.now().astimezone().strftime("%A, %b %d %Y")
+    window = f"{_fmt_local(since_iso)} – {_fmt_local(until_iso)}"
+    L = ["Ten31 CRM — Daily Activity Digest", title_date, f"Window: {window}", ""]
+
+    if not user_groups:
+        L.append("No tracked email activity from any user in this window.")
+    else:
+        total_emails = sum(g["total"] for g in user_groups)
+        total_invs = len({i for g in user_groups for i in g["investors"]})
+        L.append(f"{len(user_groups)} team member(s) active · {total_emails} email(s) "
+                 f"· {total_invs} investor(s)")
+
+        # ── Section 1: by team member (who did what; per-user Spark narrative) ──
+        L += ["", _RULE, "BY TEAM MEMBER", _RULE]
+        for g in user_groups:
+            invs = ", ".join(g["investors"]) or "(no matched investor)"
+            L += ["",
+                  f"{g['full_name'] or g['username']}  ·  {g['account_email']}",
+                  f"{g['total']} email(s) ({g['sent']} sent, {g['received']} received) "
+                  f"· {invs}", "",
+                  narratives.get(g["user_id"], ""), ""]
+            for em in g["emails"]:
+                arrow = "→ Sent" if em["direction"] == "sent" else "← Received"
+                invs_e = ", ".join(em["investors"]) or "(unmatched)"
+                subj = em.get("subject") or "(no subject)"
+                L.append(f"  {arrow} · {invs_e} · \"{subj}\"  ({_fmt_local(em['sent_at'])})")
+
+        # ── Section 2: by investor (team-wide; both directions, structured) ──
+        L += ["", _RULE, "BY INVESTOR", _RULE]
+        for inv in investor_groups:
+            L += ["",
+                  f"{inv['name']}  ·  {inv['total']} email(s) "
+                  f"({inv['inbound']} in, {inv['outbound']} out)"]
+            for em in inv["emails"]:
+                subj = em.get("subject") or "(no subject)"
+                when = _fmt_local(em["sent_at"])
+                if em["direction"] == "out":
+                    who = ", ".join(em["members"]) or "team"
+                    L.append(f"  → Sent by {who} · \"{subj}\"  ({when})")
+                else:
+                    L.append(f"  ← Received · \"{subj}\"  ({when})")
+
+    # ── Reminders due (current state — independent of the activity window) ──
+    L += _reminders_section(due_reminders or [])
+
+    L += ["", _RULE, _FOOTER]
+    return "\n".join(L)
+
+
+def build_digest(conn, since_iso, until_iso, chat_fn=None):
+    """Build the daily digest for [since_iso, until_iso). Returns
+    {subject, body, has_activity, user_count, email_count, investor_count,
+    reminder_count}. Always returns a body (empty windows get a 'no activity' note —
+    the team chose always-send). Sections: by team member (per-user Spark narrative),
+    by investor (structured), and reminders due (overdue + due today, current-state)."""
+    user_groups = collect_user_activity(conn, since_iso, until_iso)
+    investor_groups = collect_investor_activity(conn, since_iso, until_iso)
+    narratives = {g["user_id"]: summarize_user_day(g, chat_fn) for g in user_groups}
+    today_iso = datetime.now().astimezone().strftime("%Y-%m-%d")
+    due_reminders = collect_due_reminders(conn, today_iso)
+    body = _compose_body(user_groups, investor_groups, narratives, since_iso, until_iso, due_reminders)
+    stamp = datetime.now().astimezone().strftime("%b %d")
+    return {
+        "subject": f"Ten31 CRM — Daily Activity Digest · {stamp}",
+        "body": body,
+        "has_activity": bool(user_groups),
+        "user_count": len(user_groups),
+        "email_count": sum(g["total"] for g in user_groups),
+        "investor_count": len(investor_groups),
+        "reminder_count": len(due_reminders),
+    }

backend/digest_mailer.py

@@ -0,0 +1,69 @@
+"""Transport selection for CRM-originated email (daily digest, admin test sends).
+
+Prefers Gmail-over-DWD — it reuses the service account that already powers email
+capture (the grant includes gmail.compose, which can send), so there's no extra
+credential to manage — and falls back to SMTP (`smtp_send`) when DWD isn't
+available. One entry point so the digest and the admin test endpoint share the
+same routing. Stdlib only.
+"""
+import os
+
+
+class NoTransport(Exception):
+    """Neither Gmail DWD nor SMTP is configured."""
+
+
+def transport():
+    """Return the active transport: 'gmail-dwd', 'smtp', or None."""
+    try:
+        from email_integration import gmail_send
+        if gmail_send.gmail_available():
+            return "gmail-dwd"
+    except Exception:
+        pass
+    try:
+        import smtp_send
+        if smtp_send.smtp_configured():
+            return "smtp"
+    except Exception:
+        pass
+    return None
+
+
+def default_sender(conn):
+    """Domain user to send as for the DWD path. `CRM_DIGEST_SENDER` if set, else
+    the first active admin's email."""
+    s = os.environ.get("CRM_DIGEST_SENDER", "").strip()
+    if s:
+        return s
+    if conn is None:
+        return None
+    row = conn.execute(
+        "SELECT email FROM users WHERE role='admin' AND is_active=1 "
+        "AND email IS NOT NULL AND TRIM(email)!='' ORDER BY created_at LIMIT 1"
+    ).fetchone()
+    return row["email"].strip() if row and row["email"] else None
+
+
+def send_digest(conn, to_addrs, subject, body, sender_email=None):
+    """Send via the active transport. Returns the transport's result dict with a
+    'transport' key added; raises NoTransport if neither is configured."""
+    t = transport()
+    if t == "gmail-dwd":
+        from email_integration import gmail_send
+        sender = sender_email or default_sender(conn)
+        if not sender:
+            # Gmail IS available but we have nobody to send as — a config gap, not a
+            # send failure. Surface it as NoTransport so the caller returns a clear 400.
+            raise NoTransport("Gmail is enabled but no sender address is set: "
+                              "set CRM_DIGEST_SENDER or give an active admin an email.")
+        result = gmail_send.send_via_gmail(sender, to_addrs, subject, body, conn=conn)
+        result["transport"] = "gmail-dwd"
+        return result
+    if t == "smtp":
+        import smtp_send
+        result = smtp_send.send_email(to_addrs, subject, body)
+        result["transport"] = "smtp"
+        return result
+    raise NoTransport("No email transport configured: enable Gmail (DWD) or set "
+                      "SMTP via the 'Configure Digest SMTP' action.")

backend/email_integration/compose.py

@@ -0,0 +1,92 @@
+"""Create a Gmail DRAFT (never send) in the sender's mailbox via domain-wide delegation
+with the gmail.compose scope. Lets an approved Outreach draft land in the user's Gmail
+(and therefore Superhuman) Drafts — as an in-thread reply when there's an active thread,
+or a fresh email otherwise. The human reviews and sends from Gmail (guardrails #4, #6 —
+our code only ever creates a draft, it never sends).
+"""
+import base64
+import email.message
+import json as _json
+import urllib.error
+import urllib.parse
+import urllib.request
+
+from . import config as _cfg
+from . import credentials as _creds
+
+
+def _parse_subject_body(draft_text):
+    """Split a draft of the form 'Subject: ...\\n\\n<body>' into (subject, body)."""
+    text = (draft_text or "").strip()
+    lines = text.split("\n")
+    if lines and lines[0].lower().startswith("subject:"):
+        subject = lines[0].split(":", 1)[1].strip()
+        body = "\n".join(lines[1:]).lstrip("\n")
+        return subject, body
+    return "", text
+
+
+def _reply_target(conn, investor_id):
+    """LP address + active-thread headers for an in-thread reply, from the most recent
+    matched email with this investor. Returns {to, thread_id, in_reply_to} or None."""
+    try:
+        row = conn.execute(
+            "SELECT e.rfc_message_id, e.gmail_thread_id, l.matched_address "
+            "FROM emails e JOIN email_investor_links l ON l.email_id = e.id "
+            "WHERE l.fundraising_investor_id = ? AND e.is_matched = 1 "
+            "ORDER BY e.sent_at DESC LIMIT 1", (investor_id,)).fetchone()
+    except Exception:
+        return None
+    if not row or not row["matched_address"]:
+        return None
+    return {"to": row["matched_address"], "thread_id": row["gmail_thread_id"],
+            "in_reply_to": row["rfc_message_id"]}
+
+
+def _build_raw(from_addr, to_addr, subject, body, in_reply_to=None):
+    msg = email.message.EmailMessage()
+    msg["From"] = from_addr
+    msg["To"] = to_addr
+    msg["Subject"] = subject or "(no subject)"
+    if in_reply_to:
+        msg["In-Reply-To"] = in_reply_to
+        msg["References"] = in_reply_to
+    msg.set_content(body or "")
+    return base64.urlsafe_b64encode(msg.as_bytes()).decode("ascii")
+
+
+def create_outreach_draft(conn, sender_email, investor_id, draft_text):
+    """Create a Gmail draft in `sender_email`'s mailbox addressed to the investor.
+    Returns {status, ...}. Never sends."""
+    if not sender_email:
+        return {"status": "no_sender"}
+    if not _cfg.CONFIG.enabled:
+        return {"status": "integration_disabled"}
+    subject, body = _parse_subject_body(draft_text)
+    if not body.strip():
+        return {"status": "empty"}
+    target = _reply_target(conn, investor_id)
+    if not target:
+        return {"status": "no_recipient"}  # no email history -> no LP address to draft to
+    try:
+        provider = _creds.build_provider(lambda: conn)
+        token = provider.access_token_for(sender_email, _creds.GMAIL_COMPOSE_SCOPE).token
+    except Exception as exc:
+        return {"status": "auth_error", "reason": str(exc)}
+    raw = _build_raw(sender_email, target["to"], subject, body, target.get("in_reply_to"))
+    payload = {"message": {"raw": raw}}
+    if target.get("thread_id"):
+        payload["message"]["threadId"] = target["thread_id"]
+    url = f"https://gmail.googleapis.com/gmail/v1/users/{urllib.parse.quote(sender_email)}/drafts"
+    req = urllib.request.Request(url, data=_json.dumps(payload).encode("utf-8"), method="POST",
+                                 headers={"Authorization": f"Bearer {token}",
+                                          "Content-Type": "application/json"})
+    try:
+        with urllib.request.urlopen(req, timeout=20) as resp:
+            _json.loads(resp.read())
+    except urllib.error.HTTPError as e:
+        return {"status": "gmail_error", "reason": e.read().decode("utf-8", "replace")[:300]}
+    except Exception as exc:
+        return {"status": "gmail_error", "reason": str(exc)}
+    return {"status": "ok", "to": target["to"], "threaded": bool(target.get("thread_id")),
+            "gmail_url": "https://mail.google.com/mail/u/0/#drafts"}

backend/email_integration/credentials.py

@@ -32,6 +32,11 @@ from . import errors
 
 
 GMAIL_READONLY_SCOPE = "https://www.googleapis.com/auth/gmail.readonly"
+# Compose scope (authorized in Workspace DWD). Two consumers: outreach (compose.py)
+# only CREATES drafts — the human sends from Gmail; the daily-digest mailer
+# (gmail_send.py) uses this same scope to SEND, since gmail.compose authorizes
+# users.messages.send. (The narrow gmail.send scope is NOT on the DWD grant.)
+GMAIL_COMPOSE_SCOPE = "https://www.googleapis.com/auth/gmail.compose"
 GOOGLE_TOKEN_URL = "https://oauth2.googleapis.com/token"
 
 
@@ -61,13 +66,14 @@ class DWDCredentialProvider:
         self._cache: dict[str, AccessToken] = {}
         self._lock = threading.Lock()
 
-    def access_token_for(self, email_address: str) -> AccessToken:
+    def access_token_for(self, email_address: str, scope: str = GMAIL_READONLY_SCOPE) -> AccessToken:
+        key = f"{email_address}|{scope}"
         with self._lock:
-            cached = self._cache.get(email_address)
+            cached = self._cache.get(key)
             if cached and cached.expires_at - time.time() > 60:
                 return cached
-            token = self._mint(email_address)
+            token = self._mint(email_address, scope)
-            self._cache[email_address] = token
+            self._cache[key] = token
             return token
 
     def revoke(self, email_address: str) -> None:
@@ -78,7 +84,7 @@ class DWDCredentialProvider:
 
     # ------------------------------------------------------------------ helpers
 
-    def _mint(self, subject_email: str) -> AccessToken:
+    def _mint(self, subject_email: str, scope: str = GMAIL_READONLY_SCOPE) -> AccessToken:
         try:
             from cryptography.hazmat.primitives import hashes, serialization  # type: ignore
             from cryptography.hazmat.primitives.asymmetric import padding  # type: ignore
@@ -92,7 +98,7 @@ class DWDCredentialProvider:
         claim = {
             "iss": self._client_email,
             "sub": subject_email,
-            "scope": GMAIL_READONLY_SCOPE,
+            "scope": scope,
             "aud": GOOGLE_TOKEN_URL,
             "iat": now,
             "exp": now + 3600,

backend/email_integration/db.py

@@ -57,10 +57,20 @@ def _json(v) -> str:
 # ------------------------------------------------------------------ email_accounts
 
 def list_sync_ready_accounts(conn: sqlite3.Connection) -> list[sqlite3.Row]:
+    # Ready = healthy ('pending'/'active') + transient-failing ('retrying', retried every
+    # cycle for fast recovery) + 'error' accounts whose last attempt was over an hour ago.
+    # The hour-backoff on 'error' means a terminal failure (auth/permanent) self-heals once
+    # the operator fixes it WITHOUT hammering Google, and un-sticks any mailbox parked by the
+    # pre-v0.1.0:104 bug where one timeout dark-listed it forever. (last_synced_at is stamped
+    # on every attempt, success or fail, so it doubles as the last-attempt clock here.)
     cur = conn.cursor()
     cur.execute(
         "SELECT * FROM email_accounts "
-        "WHERE sync_enabled = 1 AND sync_status IN ('pending','active') "
+        "WHERE sync_enabled = 1 AND ("
+        "  sync_status IN ('pending','active','retrying') "
+        "  OR (sync_status = 'error' AND (last_synced_at IS NULL "
+        "      OR last_synced_at < datetime('now','-1 hour')))"
+        ") "
         "ORDER BY last_synced_at IS NOT NULL, last_synced_at"
     )
     return cur.fetchall()
@@ -398,6 +408,275 @@ def start_sync_run(conn: sqlite3.Connection, *, account_id: str, kind: str) -> s
     return run_id
 
 
+def _resolve_entity(row) -> tuple:
+    """Reduce one email_investor_links hydration row to a (key, name) identity for
+    the matched investor, with the same precedence the digest uses:
+    grid investor -> organization -> contact -> raw matched address. The key is
+    *typed* (`fund:`/`org:`/`contact:`/`addr:`) so the Communications filter can
+    target the right column. Soft-deleted org/contact rows arrive as NULL (filtered
+    in the join) and fall through to the next tier."""
+    if row["fund_id"] and (row["fund_name"] or "").strip():
+        return f"fund:{row['fund_id']}", row["fund_name"].strip()
+    if row["org_id"] and (row["org_name"] or "").strip():
+        return f"org:{row['org_id']}", row["org_name"].strip()
+    if row["contact_id"] and (row["contact_name"] or "").strip():
+        return f"contact:{row['contact_id']}", row["contact_name"].strip()
+    addr = (row["addr"] or "").strip()
+    return (f"addr:{addr.lower()}", addr) if addr else (None, None)
+
+
+# Hydration of an email_investor_links row up to the resolvable investor identity,
+# shared by the per-email tags and the facet dropdown. Soft-deleted org/contact
+# rows are dropped in the join so they fall through to the next identity tier.
+_LINK_IDENTITY_JOINS = """
+    LEFT JOIN fundraising_investors fi  ON fi.id = l.fundraising_investor_id
+    LEFT JOIN fundraising_contacts fic  ON fic.id = l.fundraising_contact_id
+    LEFT JOIN fundraising_investors fic_inv ON fic_inv.id = fic.investor_id
+    LEFT JOIN organizations o ON o.id = l.organization_id AND o.deleted_at IS NULL
+    LEFT JOIN contacts c      ON c.id = l.contact_id      AND c.deleted_at IS NULL
+"""
+_LINK_IDENTITY_COLS = """
+    l.matched_address AS addr,
+    COALESCE(fi.id, fic_inv.id)                       AS fund_id,
+    COALESCE(fi.investor_name, fic_inv.investor_name) AS fund_name,
+    COALESCE(fi.graveyard, fic_inv.graveyard)         AS fund_graveyard,
+    o.id AS org_id, o.name AS org_name,
+    c.id AS contact_id,
+    NULLIF(TRIM(COALESCE(c.first_name,'') || ' ' || COALESCE(c.last_name,'')), '') AS contact_name
+"""
+
+
+def query_email_activity(conn: sqlite3.Connection, *, investor_id: Optional[str] = None,
+                         account_id: Optional[str] = None, search: Optional[str] = None,
+                         direction: Optional[str] = None, since: Optional[str] = None,
+                         until: Optional[str] = None, limit: int = 100) -> dict:
+    """Captured-Gmail activity for the admin Communications panel, filterable by
+    matched investor entity, mailbox, direction and date range, with free-text
+    search over subject/snippet/sender. Returns the email rows plus the filter facets.
+
+    Matched-only: the panel shows ONLY email that links to a known investor/contact
+    (an `email_investor_links` row exists). Unmatched cold/unknown-sender email is
+    still captured for completeness but is never surfaced here.
+
+    Investor identity is *typed* (`fund:`/`org:`/`contact:`/`addr:`) and resolved with
+    the digest's precedence (grid investor -> organization -> contact -> raw address),
+    so an email matched only to a classic contact or an org domain — not yet wired to a
+    grid investor — still shows a real name and is selectable in the dropdown, instead
+    of the facet coming back empty. `investor_id` accepts a typed key (a bare id is
+    treated as `fund:` for backward compatibility).
+
+    Soft-delete: an email is live only if it still has a non-tombstoned per-mailbox
+    sighting (`email_account_messages.deleted_at IS NULL`) — the `emails` row itself
+    carries no deleted_at, so deletion lives on the sighting. Direction is decided at
+    the email level (outbound if the sender is one of our mailboxes), mirroring the
+    digest builder, so a thread reads consistently regardless of which mailbox saw it.
+    """
+    limit = max(1, min(int(limit or 100), 500))
+    cur = conn.cursor()
+    own = {(r["email_address"] or "").lower().strip()
+           for r in cur.execute("SELECT email_address FROM email_accounts")}
+    own.discard("")
+
+    where = ["EXISTS (SELECT 1 FROM email_account_messages eam "
+             "WHERE eam.email_id = e.id AND eam.deleted_at IS NULL)",
+             # Matched-only: surface email that links to a known investor/contact.
+             # Unmatched (unknown-sender) email is captured but never shown here.
+             "EXISTS (SELECT 1 FROM email_investor_links l WHERE l.email_id = e.id)"]
+    params: list = []
+    if account_id:
+        where.append("EXISTS (SELECT 1 FROM email_account_messages eam "
+                     "WHERE eam.email_id = e.id AND eam.account_id = ? "
+                     "AND eam.deleted_at IS NULL)")
+        params.append(account_id)
+    if investor_id:
+        kind, _, val = str(investor_id).partition(":")
+        if not val:                     # bare id (legacy) -> grid investor
+            kind, val = "fund", str(investor_id)
+        if kind == "fund":
+            where.append("EXISTS (SELECT 1 FROM email_investor_links l WHERE l.email_id = e.id "
+                         "AND (l.fundraising_investor_id = ? OR l.fundraising_contact_id IN "
+                         "(SELECT id FROM fundraising_contacts WHERE investor_id = ?)))")
+            params.extend([val, val])
+        elif kind == "org":
+            where.append("EXISTS (SELECT 1 FROM email_investor_links l "
+                         "WHERE l.email_id = e.id AND l.organization_id = ?)")
+            params.append(val)
+        elif kind == "contact":
+            where.append("EXISTS (SELECT 1 FROM email_investor_links l "
+                         "WHERE l.email_id = e.id AND l.contact_id = ?)")
+            params.append(val)
+        elif kind == "addr":
+            where.append("EXISTS (SELECT 1 FROM email_investor_links l "
+                         "WHERE l.email_id = e.id AND LOWER(l.matched_address) = ?)")
+            params.append(val.lower())
+        else:
+            # Unknown key prefix (malformed input) -> match nothing, never silently
+            # fall through to an unfiltered list.
+            where.append("1 = 0")
+    if search:
+        like = f"%{search.strip()}%"
+        where.append("(e.subject LIKE ? OR e.snippet LIKE ? "
+                     "OR e.from_email LIKE ? OR e.from_name LIKE ?)")
+        params.extend([like, like, like, like])
+    # Date range over the send time (ISO-8601 strings sort lexically). [since, until).
+    if since:
+        where.append("e.sent_at >= ?")
+        params.append(since)
+    if until:
+        where.append("e.sent_at < ?")
+        params.append(until)
+    direction = (direction or "").strip().lower()
+    if direction in ("inbound", "outbound") and own:
+        marks = ",".join("?" for _ in own)
+        op = "IN" if direction == "outbound" else "NOT IN"
+        where.append(f"LOWER(e.from_email) {op} ({marks})")
+        params.extend(sorted(own))
+
+    sql = ("SELECT e.id, e.subject, e.from_name, e.from_email, e.sent_at, e.snippet, "
+           "e.has_attachments, e.is_matched, e.match_status FROM emails e WHERE "
+           + " AND ".join(where) + " ORDER BY e.sent_at DESC LIMIT ?")
+    rows = [dict(r) for r in cur.execute(sql, params + [limit + 1])]
+    truncated = len(rows) > limit
+    rows = rows[:limit]
+    by_id = {r["id"]: r for r in rows}
+    for r in rows:
+        r["direction"] = "outbound" if (r["from_email"] or "").lower().strip() in own else "inbound"
+        r["mailboxes"] = []
+        r["investors"] = []      # [{id: typed-key, name}] — resolved identities
+
+    ids = list(by_id)
+    if ids:
+        marks = ",".join("?" for _ in ids)
+        for s in cur.execute(
+                "SELECT eam.email_id AS eid, ea.email_address AS addr "
+                "FROM email_account_messages eam JOIN email_accounts ea ON ea.id = eam.account_id "
+                f"WHERE eam.deleted_at IS NULL AND eam.email_id IN ({marks}) "
+                "ORDER BY ea.email_address", ids):
+            mb = by_id[s["eid"]]["mailboxes"]
+            if s["addr"] and s["addr"] not in mb:
+                mb.append(s["addr"])
+        for lnk in cur.execute(
+                f"SELECT l.email_id AS eid, {_LINK_IDENTITY_COLS} "
+                f"FROM email_investor_links l {_LINK_IDENTITY_JOINS} "
+                f"WHERE l.email_id IN ({marks})", ids):
+            # No graveyard filter here on purpose: a graveyarded investor's *email*
+            # still shows in the list with its chip (audit completeness, direct or
+            # via-contact); only the facet dropdown below hides graveyard from the picker.
+            key, name = _resolve_entity(lnk)
+            if not key:
+                continue
+            invs = by_id[lnk["eid"]]["investors"]
+            if not any(iv["id"] == key for iv in invs):
+                invs.append({"id": key, "name": name})
+
+    accounts = [dict(r) for r in cur.execute(
+        "SELECT id, email_address FROM email_accounts ORDER BY email_address")]
+
+    # Facet dropdown mirrors what the list resolves: one entry per distinct matched
+    # entity (grid investor / org / contact), across all live matched email — not just
+    # the current page — so the picker is stable under filtering. Excluded from the
+    # picker: graveyarded grid investors (CRM-wide convention) and raw-address-only
+    # matches (too many, too noisy). Both still appear in the list and remain findable
+    # by free-text search — this is an audit surface, so history is never hidden, only
+    # the picker is.
+    facet: dict[str, str] = {}
+    for r in cur.execute(
+            f"SELECT DISTINCT {_LINK_IDENTITY_COLS} "
+            f"FROM email_investor_links l {_LINK_IDENTITY_JOINS} "
+            "WHERE EXISTS (SELECT 1 FROM email_account_messages eam "
+            "WHERE eam.email_id = l.email_id AND eam.deleted_at IS NULL)"):
+        key, name = _resolve_entity(r)
+        if not key or key.startswith("addr:"):
+            continue
+        if key.startswith("fund:") and (r["fund_graveyard"] or 0):
+            continue
+        facet.setdefault(key, name)
+    investors = [{"id": k, "name": v}
+                 for k, v in sorted(facet.items(), key=lambda kv: kv[1].lower())]
+
+    return {"emails": rows, "accounts": accounts, "investors": investors,
+            "count": len(rows), "truncated": truncated}
+
+
+def search_hit_emails(conn: sqlite3.Connection, email_ids) -> dict:
+    """Display fields for the given email ids that are still live (have a
+    non-tombstoned sighting), keyed by id, with email-level direction.
+
+    Used to hydrate + soft-delete-filter semantic-search hits: the Qdrant index can
+    lag a deletion, and SQLite is canonical (never trust the derived index), so a hit
+    whose email no longer has a live sighting is dropped here rather than shown."""
+    ids = [i for i in dict.fromkeys(email_ids) if i]
+    if not ids:
+        return {}
+    cur = conn.cursor()
+    own = {(r["email_address"] or "").lower().strip()
+           for r in cur.execute("SELECT email_address FROM email_accounts")}
+    own.discard("")
+    marks = ",".join("?" for _ in ids)
+    out: dict = {}
+    for e in cur.execute(
+            "SELECT e.id, e.subject, e.from_name, e.from_email, e.sent_at, e.has_attachments "
+            f"FROM emails e WHERE e.id IN ({marks}) AND EXISTS (SELECT 1 FROM email_account_messages eam "
+            "WHERE eam.email_id = e.id AND eam.deleted_at IS NULL)", ids):
+        d = dict(e)
+        d["direction"] = "outbound" if (d["from_email"] or "").lower().strip() in own else "inbound"
+        out[d["id"]] = d
+    return out
+
+
+def query_email_detail(conn: sqlite3.Connection, email_id: str) -> Optional[dict]:
+    """Full record for one captured email — the Communications detail view (full
+    body + recipients + matched investor identities + mailboxes + attachments).
+
+    Returns None if the email doesn't exist or has no live (non-tombstoned) sighting:
+    soft-delete lives on the per-mailbox `email_account_messages` row, not on `emails`,
+    so an email is only "live" while at least one sighting survives. Direction is set
+    at the email level (outbound if the sender is one of our mailboxes), matching the
+    list. The raw remote `body_html` is NOT returned (XSS); the response carries the
+    plain-text `body_text` plus a `has_html` flag so the UI can note an HTML-only email."""
+    cur = conn.cursor()
+    e = cur.execute(
+        "SELECT e.id, e.subject, e.from_name, e.from_email, e.sent_at, e.snippet, "
+        "e.body_text, e.body_html, e.has_attachments, e.match_status, e.gmail_thread_id "
+        "FROM emails e WHERE e.id = ? AND EXISTS (SELECT 1 FROM email_account_messages eam "
+        "WHERE eam.email_id = e.id AND eam.deleted_at IS NULL)", (email_id,)).fetchone()
+    if not e:
+        return None
+    row = dict(e)
+    # Don't ship the raw remote HTML to the client (XSS if any consumer ever renders
+    # it); the UI shows the plain-text body and only needs to know HTML exists.
+    row["has_html"] = bool((row.pop("body_html", None) or "").strip())
+
+    own = {(r["email_address"] or "").lower().strip()
+           for r in cur.execute("SELECT email_address FROM email_accounts")}
+    own.discard("")
+    row["direction"] = "outbound" if (row["from_email"] or "").lower().strip() in own else "inbound"
+
+    row["mailboxes"] = [r["addr"] for r in cur.execute(
+        "SELECT DISTINCT ea.email_address AS addr FROM email_account_messages eam "
+        "JOIN email_accounts ea ON ea.id = eam.account_id "
+        "WHERE eam.email_id = ? AND eam.deleted_at IS NULL ORDER BY ea.email_address", (email_id,))]
+
+    row["recipients"] = [dict(r) for r in cur.execute(
+        "SELECT address, display_name, kind FROM email_recipients "
+        "WHERE email_id = ? AND kind IN ('to','cc') "
+        "ORDER BY CASE kind WHEN 'to' THEN 0 ELSE 1 END, address", (email_id,))]
+
+    row["attachments"] = [dict(r) for r in cur.execute(
+        "SELECT filename, mime_type, size_bytes, download_status FROM email_attachments "
+        "WHERE email_id = ? ORDER BY filename", (email_id,))]
+
+    investors: dict[str, str] = {}
+    for lnk in cur.execute(
+            f"SELECT {_LINK_IDENTITY_COLS} FROM email_investor_links l {_LINK_IDENTITY_JOINS} "
+            "WHERE l.email_id = ?", (email_id,)):
+        key, name = _resolve_entity(lnk)
+        if key:
+            investors.setdefault(key, name)
+    row["investors"] = [{"id": k, "name": v} for k, v in investors.items()]
+    return row
+
+
 def finish_sync_run(conn: sqlite3.Connection, run_id: str, *, status: str,
                     stats: Optional[dict] = None, error: Optional[str] = None) -> None:
     stats = stats or {}

backend/email_integration/digest_scheduler.py

@@ -0,0 +1,206 @@
+"""Daily activity-digest scheduler (Phase B).
+
+Co-located with the Gmail sync scheduler (it shares the same conn-factory and
+daemon-thread idiom). One daemon thread wakes every 60s and fires the daily
+activity digest once per local day, at/after the configured send hour.
+
+Control lives in the DB, set from Settings -> Admin (digest_builder.load_digest_policy
+-> app_settings 'digest_policy'): {enabled, send_hour}. The thread always runs and
+re-reads the policy each cycle, so toggling the digest on/off or changing the time
+takes effect on the next loop — no restart. CRM_DIGEST_ENABLED/SEND_HOUR only seed
+the first-boot default before an admin sets the policy.
+
+The send is an internal ops email to the team's own admins — exempt from the
+"agents draft, humans send" rule (which governs outward LP/prospect contact).
+Digest content is summarized on Spark (local), never Claude — see digest_builder.
+
+Window: the content covers (last successful send, now]. Tracked in app_settings
+so a missed day's activity rolls into the next digest rather than being dropped;
+the first-ever run covers the prior 24h. The once-per-day guard is a separate
+local-date key. The transport (Gmail-DWD -> SMTP) is digest_mailer's job.
+"""
+import json
+import logging
+import sqlite3
+import threading
+from datetime import datetime, timedelta, timezone
+
+from .scheduler import _conn_factory_from_env
+
+
+log = logging.getLogger("email_integration.digest_scheduler")
+
+_LAST_DATE_KEY = "digest_last_sent_date"   # local YYYY-MM-DD — once-per-day guard
+_LAST_AT_KEY = "digest_last_sent_at"       # UTC ISO — content-window cursor
+
+_state: dict[str, object] = {"thread": None, "stop": threading.Event()}
+
+
+def _utc_iso(dt: datetime) -> str:
+    return dt.astimezone(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
+
+
+# app_settings access kept local (no server.py import — avoid the startup cycle);
+# the value_json/JSON encoding matches server.get/set_app_setting exactly.
+def _get_setting(conn, key):
+    try:
+        row = conn.execute("SELECT value_json FROM app_settings WHERE key = ?", (key,)).fetchone()
+    except sqlite3.OperationalError:
+        return None
+    if not row:
+        return None
+    try:
+        return json.loads(row["value_json"])
+    except Exception:
+        return None
+
+
+def _set_setting(conn, key, value) -> None:
+    conn.execute(
+        "INSERT INTO app_settings (key, value_json, updated_at) VALUES (?, ?, ?) "
+        "ON CONFLICT(key) DO UPDATE SET value_json = excluded.value_json, "
+        "updated_at = excluded.updated_at",
+        (key, json.dumps(value), _utc_iso(datetime.now(timezone.utc))),
+    )
+
+
+def _admin_recipients(conn) -> list[str]:
+    rows = conn.execute(
+        "SELECT email FROM users WHERE role = 'admin' AND is_active = 1 "
+        "AND email IS NOT NULL AND TRIM(email) != ''"
+    ).fetchall()
+    return [str(r["email"]).strip() for r in rows if str(r["email"]).strip()]
+
+
+def _build_and_send(conn, since_iso, until_iso, *, build_fn=None, send_fn=None):
+    """Build the digest and hand it to the transport. Raises digest_mailer.NoTransport
+    (no transport / no recipient) — callers map that to a clear 400; the daily loop
+    logs it. build_fn/send_fn are injectable for tests."""
+    import digest_builder
+    import digest_mailer
+
+    bf = build_fn or digest_builder.build_digest
+    sf = send_fn or digest_mailer.send_digest
+
+    recipients = _admin_recipients(conn)
+    if not recipients:
+        raise digest_mailer.NoTransport(
+            "No active admin has an email address — give one an address to receive the digest.")
+
+    digest = bf(conn, since_iso, until_iso)
+    result = sf(conn, recipients, digest["subject"], digest["body"])
+    return {
+        "recipients": recipients,
+        "transport": (result or {}).get("transport"),
+        "has_activity": digest["has_activity"],
+        "user_count": digest["user_count"],
+        "email_count": digest["email_count"],
+        "investor_count": digest.get("investor_count"),
+        "window": [since_iso, until_iso],
+    }
+
+
+def send_digest_window(conn_factory=None, *, since_iso, until_iso,
+                       build_fn=None, send_fn=None):
+    """Build the digest for an explicit (since_iso, until_iso] window and send it
+    to the active-admin set now, WITHOUT advancing the daily cursor — a manual or
+    preview send must never suppress the scheduled daily digest. Same transport +
+    recipient rules as the daily path (raises digest_mailer.NoTransport when none
+    is configured / no admin has an address). Backs the admin 'send now' endpoint.
+
+    No DB writes happen here (the cursor is deliberately untouched), so the connection
+    is opened and closed without a commit — don't add one without revisiting that."""
+    factory = conn_factory or _conn_factory_from_env()
+    conn = factory()
+    try:
+        result = _build_and_send(conn, since_iso, until_iso,
+                                 build_fn=build_fn, send_fn=send_fn)
+        return {"status": "sent", **result}
+    finally:
+        conn.close()
+
+
+def maybe_send_digest(conn_factory=None, *, force=False,
+                      now_local=None, now_utc=None, build_fn=None, send_fn=None):
+    """Send the daily digest if it is due (or unconditionally when force=True).
+
+    Daily path (the scheduler loop): skips before the send hour and if already sent
+    today; content window runs from the last send to now and the cursor advances on
+    success. force path: ignores the policy and the guards, uses a fixed last-24h
+    window, and does NOT advance the daily cursor. (The admin 'send now' / preview
+    endpoints now use send_digest_window for an arbitrary window; force is retained
+    for the fixed last-24h case and its tests.)"""
+    import digest_builder
+
+    factory = conn_factory or _conn_factory_from_env()
+    conn = factory()
+    try:
+        policy = digest_builder.load_digest_policy(conn)
+        if not force and not policy["enabled"]:
+            return {"status": "disabled"}
+
+        nl = now_local or datetime.now()
+        nu = now_utc or datetime.now(timezone.utc)
+
+        if not force:
+            today = nl.strftime("%Y-%m-%d")
+            if nl.hour < policy["send_hour"]:
+                return {"status": "before_send_hour", "send_hour": policy["send_hour"]}
+            if _get_setting(conn, _LAST_DATE_KEY) == today:
+                return {"status": "already_sent_today"}
+
+        until_iso = _utc_iso(nu)
+        last_at = None if force else _get_setting(conn, _LAST_AT_KEY)
+        since_iso = last_at or _utc_iso(nu - timedelta(hours=24))
+
+        result = _build_and_send(conn, since_iso, until_iso, build_fn=build_fn, send_fn=send_fn)
+
+        if not force:
+            _set_setting(conn, _LAST_DATE_KEY, nl.strftime("%Y-%m-%d"))
+            _set_setting(conn, _LAST_AT_KEY, until_iso)
+            conn.commit()
+        return {"status": "sent", **result}
+    finally:
+        conn.close()
+
+
+def start_digest_scheduler(conn_factory=None) -> None:
+    """Start the daily digest loop (idempotent). The thread always runs and reads
+    the DB policy each cycle (admin-panel control), so it sends only when the policy
+    is enabled — no env gate, no restart needed to toggle."""
+    if _state["thread"] is not None:
+        return
+
+    factory = conn_factory or _conn_factory_from_env()
+    stop = threading.Event()
+    _state["stop"] = stop
+
+    def _loop():
+        log.info("digest scheduler started (policy-controlled via Settings -> Admin)")
+        if stop.wait(15):   # let server finish startup
+            return
+        while not stop.is_set():
+            try:
+                res = maybe_send_digest(factory)
+                if res.get("status") == "sent":
+                    log.info("daily digest sent: %s", res)
+            except Exception:
+                log.exception("digest send failed; will retry next cycle")
+            if stop.wait(60):
+                return
+
+    t = threading.Thread(target=_loop, name="digest", daemon=True)
+    t.start()
+    _state["thread"] = t
+
+
+def stop_digest_scheduler() -> None:
+    ev: threading.Event = _state["stop"]  # type: ignore
+    ev.set()
+    t = _state.get("thread")
+    if t:
+        try:
+            t.join(timeout=5)  # type: ignore
+        except Exception:
+            pass
+    _state["thread"] = None

backend/email_integration/gmail_send.py

@@ -0,0 +1,69 @@
+"""Send an email via the Gmail API using the same domain-wide delegation that
+powers capture and draft creation.
+
+The DWD grant on this deployment includes the `gmail.compose` scope (verified
+2026-06-15: token mint + a live messages.send both succeed), and `gmail.compose`
+authorizes `users.messages.send`. So CRM-originated mail (the daily digest) can
+send through the existing service account — no SMTP account, no app password, no
+admin change. Sends impersonating `sender_email`, which must be a Workspace user
+in the delegated domain. Mirrors the REST pattern in compose.py; stdlib only.
+"""
+import base64
+import email.message
+import json
+import os
+import urllib.error
+import urllib.parse
+import urllib.request
+
+from . import config as _cfg
+from . import credentials as _creds
+
+
+def gmail_available():
+    """True when DWD send is usable: integration enabled, DWD auth, key present."""
+    cfg = _cfg.CONFIG
+    if not cfg.enabled or cfg.primary_auth != "dwd":
+        return False
+    return bool(cfg.dwd_key_path) and os.path.exists(cfg.dwd_key_path)
+
+
+def _build_raw(from_addr, to_addrs, subject, body):
+    msg = email.message.EmailMessage()
+    msg["From"] = from_addr
+    msg["To"] = ", ".join(to_addrs)
+    msg["Subject"] = subject or "(no subject)"
+    msg.set_content(body or "")
+    return base64.urlsafe_b64encode(msg.as_bytes()).decode("ascii")
+
+
+def send_via_gmail(sender_email, to_addrs, subject, body, conn=None):
+    """Send one message as `sender_email` to `to_addrs` via the Gmail API (DWD).
+    Returns {'sent_to', 'from', 'message_id'}; raises on failure."""
+    if isinstance(to_addrs, str):
+        to_addrs = [to_addrs]
+    to_addrs = [a for a in (str(x).strip() for x in to_addrs) if a]
+    if not to_addrs:
+        raise ValueError("no recipients")
+    if not sender_email:
+        raise ValueError("no sender_email (DWD impersonation needs a domain user)")
+
+    # conn is only consulted by the OAuth provider path; the DWD provider (the one
+    # used here) reads the service-account key from disk and ignores it.
+    provider = _creds.build_provider(lambda: conn)
+    token = provider.access_token_for(sender_email, _creds.GMAIL_COMPOSE_SCOPE).token
+    raw = _build_raw(sender_email, to_addrs, subject, body)
+    url = ("https://gmail.googleapis.com/gmail/v1/users/"
+           f"{urllib.parse.quote(sender_email)}/messages/send")
+    req = urllib.request.Request(
+        url, data=json.dumps({"raw": raw}).encode("utf-8"), method="POST",
+        headers={"Authorization": f"Bearer {token}", "Content-Type": "application/json"})
+    try:
+        with urllib.request.urlopen(req, timeout=20) as resp:
+            result = json.loads(resp.read())
+    except urllib.error.HTTPError as e:
+        detail = e.read().decode("utf-8", "replace")[:300]
+        raise RuntimeError(f"Gmail API send failed for {sender_email} (HTTP {e.code}): {detail}")
+    except OSError as e:  # URLError/timeout/DNS (URLError subclasses OSError)
+        raise RuntimeError(f"Gmail API unreachable: {e}")
+    return {"sent_to": to_addrs, "from": sender_email, "message_id": result.get("id")}

backend/email_integration/migrations/0003_email_proposal_matrix.sql

@@ -0,0 +1,30 @@
+-- ============================================================================
+-- email_proposal_matrix — Matrix-review state for an email_activity_proposal,
+-- kept 1:1 with the proposal (proposal_id PK). The CRM runs on the box and has
+-- no matrix-nio, so it cannot post to Matrix itself: the intake bot (on the Spark)
+-- PULLS pending proposals, posts a review card to the dedicated Matrix review room,
+-- and writes the thread-root event_id back here. Persisting it CRM-side (not just in
+-- the bot's memory) keeps both surfaces in sync and survives a bot restart.
+--
+-- A SIDE TABLE rather than new columns on email_activity_proposals because the
+-- email-integration migration runner (email_integration/db.py:apply_migrations)
+-- re-runs every .sql file on every boot via executescript with no ledger — so
+-- CREATE TABLE IF NOT EXISTS is idempotent, whereas ALTER ... ADD COLUMN would throw
+-- "duplicate column" on the second boot and abort startup. Reversal: DROP TABLE
+-- (this runner has no .down.sql convention; cf. 0001/0002).
+--
+--   posted_at — set once the bot has posted the review card (event_id = thread root).
+--   closed_at — set when the thread is resolved: either the bot decided in-thread, OR
+--               the bot announced a web-side decision. A posted+decided proposal with
+--               closed_at NULL is exactly the bot's signal to post "decided on the web"
+--               into the thread and then close it.
+-- ============================================================================
+CREATE TABLE IF NOT EXISTS email_proposal_matrix (
+    proposal_id  TEXT PRIMARY KEY,
+    event_id     TEXT,                  -- Matrix thread-root event id of the posted review card
+    posted_at    TEXT,
+    closed_at    TEXT,
+    created_at   TEXT DEFAULT (datetime('now')),
+    FOREIGN KEY(proposal_id) REFERENCES email_activity_proposals(id) ON DELETE CASCADE
+);
+CREATE INDEX IF NOT EXISTS idx_email_proposal_matrix_event ON email_proposal_matrix(event_id);

backend/email_integration/routes.py

@@ -33,6 +33,9 @@ from . import scheduler as _sched
 _GET_ROUTES = {
     "/api/email/status": "status",
     "/api/email/accounts": "list_accounts",
+    "/api/email/activity": "activity",
+    "/api/email/detail": "detail",
+    "/api/email/search": "search",
     "/api/email/threads": "list_threads",
     "/api/email/oauth/start": "oauth_start",
     "/api/email/oauth/callback": "oauth_callback",
@@ -115,7 +118,9 @@ def _require_admin(handler) -> Optional[dict]:
 # ---------------------------------------------------------------------------- GET handlers
 
 def _h_status(handler):
-    user = _require_auth(handler)
+    # Email Capture is an admin-only surface (nav-hidden from members); these read
+    # endpoints expose mailbox/sync metadata, so enforce admin server-side too.
+    user = _require_admin(handler)
     if not user:
         return
     snap = _sched.status_snapshot()
@@ -150,7 +155,9 @@ def _h_status(handler):
 
 
 def _h_list_accounts(handler):
-    user = _require_auth(handler)
+    # Admin-only: the mailbox list (addresses, sync state, errors) belongs to the
+    # admin-only Email Capture surface. Enforced server-side, not just nav-hidden.
+    user = _require_admin(handler)
     if not user:
         return
     conn = _conn()
@@ -162,14 +169,147 @@ def _h_list_accounts(handler):
             "FROM email_accounts ORDER BY email_address"
         )
         rows = [dict(r) for r in cur.fetchall()]
+        # Per-mailbox counts: emails are de-duplicated globally, so "captured per
+        # mailbox" comes from the per-account sighting table; "matched" joins to emails.
+        captured, matched = {}, {}
+        try:
+            captured = {r["account_id"]: r["n"] for r in cur.execute(
+                "SELECT account_id, COUNT(*) AS n FROM email_account_messages "
+                "WHERE deleted_at IS NULL GROUP BY account_id")}
+            matched = {r["account_id"]: r["n"] for r in cur.execute(
+                "SELECT eam.account_id AS account_id, COUNT(*) AS n FROM email_account_messages eam "
+                "JOIN emails e ON e.id = eam.email_id "
+                "WHERE eam.deleted_at IS NULL AND e.is_matched = 1 GROUP BY eam.account_id")}
+        except sqlite3.OperationalError:
+            pass
+        for r in rows:
+            r["captured"] = captured.get(r["id"], 0)
+            r["matched"] = matched.get(r["id"], 0)
     finally:
         conn.close()
-    # Non-admins only see their own row
-    if user.get("role") != "admin":
-        rows = [r for r in rows if r["user_id"] == user["user_id"]]
     handler.send_json({"accounts": rows})
 
 
+def _h_activity(handler):
+    # Admin-only: the Communications page renders captured-Gmail activity (the classic
+    # manual-log surface was retired). Mailbox/investor substance is admin-scoped, so
+    # enforce admin server-side, not just nav-hide.
+    user = _require_admin(handler)
+    if not user:
+        return
+    q = handler.get_query_params()
+    try:
+        limit = int(q.get("limit", 100))
+    except (TypeError, ValueError):
+        limit = 100
+    conn = _conn()
+    try:
+        result = _db.query_email_activity(
+            conn,
+            investor_id=(q.get("investor_id") or "").strip() or None,
+            account_id=(q.get("account_id") or "").strip() or None,
+            search=(q.get("q") or q.get("search") or "").strip() or None,
+            direction=(q.get("direction") or "").strip() or None,
+            since=(q.get("since") or "").strip() or None,
+            until=(q.get("until") or "").strip() or None,
+            limit=limit,
+        )
+    finally:
+        conn.close()
+    handler.send_json(result)
+
+
+def _h_detail(handler):
+    # Admin-only: the full body + recipients of a captured email is admin-scoped
+    # substance, same as the activity list it expands from.
+    user = _require_admin(handler)
+    if not user:
+        return
+    email_id = (handler.get_query_params().get("id") or "").strip()
+    if not email_id:
+        return handler.send_error_json("id required", 400)
+    conn = _conn()
+    try:
+        detail = _db.query_email_detail(conn, email_id)
+    finally:
+        conn.close()
+    if detail is None:
+        return handler.send_error_json("Not found", 404)
+    handler.send_json(detail)
+
+
+def _semantic_email_search(query: str, top_k: int) -> list:
+    """Hybrid (dense + BM25, reranked) retrieval over the email bodies indexed in
+    Qdrant, pre-filtered to doc_type='email'. Returns raw ranked hits (payload carries
+    source_id=email_id, lp_name, date_ts, text). The ingest stack (Spark Control +
+    Qdrant + the sparse encoder) lives in the Docker image, so it's imported lazily —
+    a bare CRM without it raises, and the caller maps that to a 503."""
+    import os
+    import sys
+    ingest_dir = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "ingest")
+    if ingest_dir not in sys.path:
+        sys.path.insert(0, ingest_dir)
+    import search as _ingest_search  # ingest/search.py
+    filt = {"must": [{"key": "doc_type", "match": {"value": "email"}}]}
+    return _ingest_search.hybrid_search(query, top_k=top_k, rerank=True, filt=filt)
+
+
+def _h_search(handler):
+    # Admin-only semantic search over captured email *content* (bodies), distinct from
+    # the structured subject/sender filters in _h_activity. Matched email bodies are the
+    # only email indexed in Qdrant (see ingest/chunking). Soft-delete-filtered + hydrated
+    # against SQLite (canonical) so a deleted email never surfaces from the stale index.
+    user = _require_admin(handler)
+    if not user:
+        return
+    q = handler.get_query_params()
+    query = (q.get("q") or q.get("query") or "").strip()
+    if not query:
+        return handler.send_json({"query": "", "results": []})
+    try:
+        top_k = min(50, max(1, int(q.get("top_k", 25))))
+    except (TypeError, ValueError):
+        top_k = 25
+
+    try:
+        hits = _semantic_email_search(query, top_k)
+    except Exception as e:
+        # Spark Control / Qdrant unreachable, or the ingest stack isn't installed.
+        # Log server-side (an error can carry a URL/host); give the UI a clean 503.
+        import sys
+        print(f"[email-search] retrieval failed: {type(e).__name__}: {e}", file=sys.stderr)
+        return handler.send_error_json("Content search is unavailable (Spark/Qdrant not reachable).", 503)
+
+    # Hydrate + soft-delete-filter against SQLite (canonical), preserving rank order.
+    payloads = [(h.get("payload", {}) or {}, h) for h in hits]
+    ids = [p.get("source_id") for p, _ in payloads]
+    conn = _conn()
+    try:
+        live = _db.search_hit_emails(conn, ids)
+    finally:
+        conn.close()
+
+    results = []
+    for p, h in payloads:
+        eid = p.get("source_id")
+        e = live.get(eid)
+        if not e:
+            continue  # deleted since indexing, or not matched-resolvable -> drop
+        results.append({
+            "email_id": eid,
+            "subject": e["subject"],
+            "from_name": e["from_name"],
+            "from_email": e["from_email"],
+            "sent_at": e["sent_at"],
+            "direction": e["direction"],
+            "has_attachments": e["has_attachments"],
+            "lp_name": p.get("lp_name"),
+            "score": h.get("score"),
+            "excerpt": (h.get("text") or p.get("text") or "").replace("\n", " ").strip()[:300],
+        })
+    handler.send_json({"query": query, "results": results, "count": len(results)})
+
+
 def _h_list_threads(handler):
     user = _require_auth(handler)
     if not user:

backend/email_integration/sync.py

@@ -23,6 +23,7 @@ import logging
 import sqlite3
 import traceback
 from typing import Optional
+from urllib.error import URLError
 
 from . import attachments as _attach
 from . import config as _cfg
@@ -112,6 +113,15 @@ def sync_account(conn_factory, credential_provider, account,
         error_str = "history expired; fallback to date backfill"
         status = "partial"
         _fallback_date_backfill(conn_factory, client, account, index, run_stats)
+    except (_errors.RateLimitError, _errors.TransientError, URLError, TimeoutError) as e:
+        # A network / 5xx / rate-limit error that outlived the in-pass retry loop.
+        # This is TRANSIENT, not terminal: park it as 'retrying' (which the scheduler
+        # still picks up every cycle) instead of 'error' (which it excludes). Fixes the
+        # v<=0.1.0:103 bug where a single timeout dark-listed a mailbox until a manual
+        # kick. Terminal causes (auth, permanent, unexpected) still fall through to 'error'.
+        error_str = f"transient: {type(e).__name__}: {e}"
+        status = "retrying"
+        log.warning("transient error during sync of %s: %s", email_addr, e)
     except Exception as e:
         error_str = f"unexpected: {type(e).__name__}: {e}"
         status = "error"

backend/email_integration/test_compose.py

@@ -0,0 +1,62 @@
+#!/usr/bin/env python3
+"""Test the Gmail-draft message construction (the part that doesn't need live Gmail):
+subject/body parsing, reply-target resolution, and the RFC822 build incl. threading
+headers. The actual drafts.create call is exercised on the box. Synthetic data only.
+Run: cd backend && python3 email_integration/test_compose.py
+"""
+import base64
+import os
+import sqlite3
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+from email_integration import compose as cp  # noqa: E402
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+def main():
+    s, b = cp._parse_subject_body("Subject: Following up\n\nHi Sarah,\n\nthanks for the call.")
+    check(s == "Following up" and b.startswith("Hi Sarah"), "parses 'Subject:' line + body")
+    s2, b2 = cp._parse_subject_body("No subject prefix here")
+    check(s2 == "" and b2 == "No subject prefix here", "no subject line -> empty subject, full body")
+
+    c = sqlite3.connect(":memory:")
+    c.row_factory = sqlite3.Row
+    c.executescript("""
+        CREATE TABLE emails(id TEXT, rfc_message_id TEXT, gmail_thread_id TEXT, sent_at TEXT, is_matched INT);
+        CREATE TABLE email_investor_links(email_id TEXT, fundraising_investor_id TEXT, matched_address TEXT);
+    """)
+    c.execute("INSERT INTO emails VALUES('e1','<m1@x>','t1','2026-06-01',1)")
+    c.execute("INSERT INTO email_investor_links VALUES('e1','inv1','lp@harborvine.example')")
+    c.commit()
+    t = cp._reply_target(c, "inv1")
+    check(t and t["to"] == "lp@harborvine.example" and t["thread_id"] == "t1" and t["in_reply_to"] == "<m1@x>",
+          "reply target resolves LP address + thread + in-reply-to")
+    check(cp._reply_target(c, "nope") is None, "no history -> no reply target")
+
+    raw = cp._build_raw("grant@ten31.xyz", "lp@x.example", "Hi", "Body text here", "<m1@x>")
+    dec = base64.urlsafe_b64decode(raw).decode("utf-8", "replace")
+    check("From: grant@ten31.xyz" in dec and "To: lp@x.example" in dec, "RFC822 has From + To")
+    check("Subject: Hi" in dec and "Body text here" in dec, "RFC822 has Subject + body")
+    check("In-Reply-To: <m1@x>" in dec and "References: <m1@x>" in dec, "threading headers set for replies")
+    raw2 = cp._build_raw("a@b.co", "c@d.co", "", "body", None)
+    dec2 = base64.urlsafe_b64decode(raw2).decode("utf-8", "replace")
+    check("Subject: (no subject)" in dec2 and "In-Reply-To" not in dec2, "no subject / no thread -> fresh email")
+
+    if FAILS:
+        print(f"\nFAILED ({len(FAILS)})")
+        for f in FAILS:
+            print("  - " + f)
+        sys.exit(1)
+    print("\nALL PASS (gmail compose message construction)")
+
+
+if __name__ == "__main__":
+    main()

backend/email_integration/test_email_activity_panel.py

@@ -0,0 +1,335 @@
+#!/usr/bin/env python3
+"""Test the admin-only email-activity panel (Communications tab, v0.1.0:80).
+
+Covers the pure query (`db.query_email_activity`): matched-only scope (unmatched
+cold/unknown-sender email is never surfaced), investor/mailbox/search/direction/
+date-range filters, per-sighting soft-delete, direction at the email level, mailbox
+roll-ups, and the *typed* investor facet (grid investor / org / contact), including
+the v83 fix where an email matched only to a classic contact or org domain — not yet
+wired to a grid investor — still resolves to a real name and appears in the dropdown
+(previously the facet came back empty). Also asserts the route handler enforces admin
+server-side. Synthetic data only.
+
+Run: cd backend && python3 email_integration/test_email_activity_panel.py
+"""
+import os
+import sqlite3
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+from email_integration import db as _db  # noqa: E402
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+def make_db():
+    conn = sqlite3.connect(":memory:")
+    conn.row_factory = sqlite3.Row
+    conn.executescript("""
+        CREATE TABLE email_accounts (id TEXT PRIMARY KEY, email_address TEXT);
+        CREATE TABLE emails (id TEXT PRIMARY KEY, subject TEXT, from_name TEXT, from_email TEXT,
+            sent_at TEXT, snippet TEXT, body_text TEXT, body_html TEXT, gmail_thread_id TEXT,
+            has_attachments INT DEFAULT 0, is_matched INT DEFAULT 0,
+            match_status TEXT DEFAULT 'unmatched');
+        CREATE TABLE email_account_messages (id TEXT PRIMARY KEY, email_id TEXT, account_id TEXT,
+            is_sent INT DEFAULT 0, deleted_at TEXT);
+        CREATE TABLE email_recipients (id TEXT PRIMARY KEY, email_id TEXT, address TEXT,
+            display_name TEXT, kind TEXT);
+        CREATE TABLE email_attachments (id TEXT PRIMARY KEY, email_id TEXT, filename TEXT,
+            mime_type TEXT, size_bytes INTEGER, download_status TEXT);
+        CREATE TABLE email_investor_links (id TEXT PRIMARY KEY, email_id TEXT,
+            fundraising_investor_id TEXT, fundraising_contact_id TEXT,
+            organization_id TEXT, contact_id TEXT, matched_address TEXT);
+        CREATE TABLE fundraising_investors (id TEXT PRIMARY KEY, investor_name TEXT, graveyard INTEGER DEFAULT 0);
+        CREATE TABLE fundraising_contacts (id TEXT PRIMARY KEY, investor_id TEXT, full_name TEXT);
+        CREATE TABLE organizations (id TEXT PRIMARY KEY, name TEXT, deleted_at TEXT);
+        CREATE TABLE contacts (id TEXT PRIMARY KEY, first_name TEXT, last_name TEXT,
+            organization_id TEXT, deleted_at TEXT);
+    """)
+    # Two mailboxes (us); investors reached different ways: a grid investor directly,
+    # a grid investor only via a contact link, a graveyarded grid investor, an org-only
+    # (domain) match, and a classic-contact-only match (the case that left the dropdown
+    # empty before v83 — neither carries a grid id).
+    conn.executemany("INSERT INTO email_accounts VALUES (?,?)", [
+        ("acc-grant", "grant@ten31.xyz"),
+        ("acc-jon", "jonathan@ten31.xyz"),
+    ])
+    conn.executemany("INSERT INTO fundraising_investors VALUES (?,?,?)", [
+        ("inv-harbor", "Harbor & Vine", 0),
+        ("inv-pacific", "Pacific Capital", 0),
+        ("inv-dead", "Dead Deal LP", 1),
+    ])
+    conn.execute("INSERT INTO fundraising_contacts VALUES ('fc-1','inv-pacific','Sarah Williams')")
+    conn.execute("INSERT INTO organizations VALUES ('org-bridge','Bridgewater',NULL)")
+    conn.execute("INSERT INTO contacts VALUES ('c-solo','Nina','Park',NULL,NULL)")
+    # Emails:
+    #  e1 outbound -> Harbor (grid), seen by grant
+    #  e2 inbound  -> Harbor (grid), seen by grant + jonathan
+    #  e3 inbound  -> Pacific via grid contact link, seen by jonathan
+    #  e4 inbound, UNMATCHED -> excluded (matched-only)
+    #  e5 inbound, only sighting tombstoned -> excluded
+    #  e6 inbound  -> Dead Deal LP (graveyard grid investor)
+    #  e7 inbound  -> Bridgewater via ORG-domain match (no grid id)
+    #  e8 inbound  -> Nina Park via CLASSIC-contact match (no grid id, no org)
+    conn.executemany(
+        "INSERT INTO emails (id,subject,from_name,from_email,sent_at,snippet,has_attachments,is_matched,match_status) VALUES (?,?,?,?,?,?,?,?,?)",
+        [
+            ("e1", "Fund III update", "Grant", "grant@ten31.xyz", "2026-06-05T10:00:00", "here is the deck", 1, 1, "matched"),
+            ("e2", "Re: Fund III update", "LP Harbor", "lp@harborvine.example", "2026-06-06T09:00:00", "thanks, one question", 0, 1, "matched"),
+            ("e3", "Intro", "Sarah Williams", "sarah@pacificcap.example", "2026-06-07T08:00:00", "would love to chat", 0, 1, "matched"),
+            ("e4", "Cold inbound", "Random", "noreply@spam.example", "2026-06-08T08:00:00", "buy now", 0, 0, "unmatched"),
+            ("e5", "Deleted thread", "Ghost", "ghost@x.example", "2026-06-09T08:00:00", "gone", 0, 1, "matched"),
+            ("e6", "Old dead-deal thread", "Dead LP", "lp@deaddeal.example", "2026-06-01T00:00:00", "we passed", 0, 1, "matched"),
+            ("e7", "Macro view", "Ray", "ray@bridgewater.example", "2026-06-10T08:00:00", "rates outlook", 0, 1, "matched"),
+            ("e8", "Coffee?", "Nina Park", "nina@solo.example", "2026-06-11T08:00:00", "in town next week", 0, 1, "matched"),
+        ])
+    conn.executemany(
+        "INSERT INTO email_account_messages (id,email_id,account_id,is_sent,deleted_at) VALUES (?,?,?,?,?)",
+        [
+            ("m1", "e1", "acc-grant", 1, None),
+            ("m2", "e2", "acc-grant", 0, None),
+            ("m3", "e2", "acc-jon", 0, None),
+            ("m4", "e3", "acc-jon", 0, None),
+            ("m5", "e4", "acc-grant", 0, None),
+            ("m6", "e5", "acc-grant", 0, "2026-06-10T00:00:00"),  # tombstoned
+            ("m7", "e6", "acc-grant", 0, None),
+            ("m8", "e7", "acc-grant", 0, None),
+            ("m9", "e8", "acc-jon", 0, None),
+        ])
+    conn.executemany(
+        "INSERT INTO email_investor_links (id,email_id,fundraising_investor_id,fundraising_contact_id,organization_id,contact_id,matched_address) VALUES (?,?,?,?,?,?,?)",
+        [
+            ("l1", "e1", "inv-harbor", None, None, None, "lp@harborvine.example"),
+            ("l2", "e2", "inv-harbor", None, None, None, "lp@harborvine.example"),
+            ("l3", "e3", None, "fc-1", None, None, "sarah@pacificcap.example"),
+            ("l5", "e5", "inv-harbor", None, None, None, "lp@harborvine.example"),
+            ("l6", "e6", "inv-dead", None, None, None, "lp@deaddeal.example"),
+            ("l7", "e7", None, None, "org-bridge", None, "ray@bridgewater.example"),
+            ("l8", "e8", None, None, None, "c-solo", "nina@solo.example"),
+        ])
+    # Full body + recipients + an attachment on e2, for the detail view.
+    conn.execute("UPDATE emails SET body_text = ?, gmail_thread_id = ?, has_attachments = 1 WHERE id = 'e2'",
+                 ("Thanks for the deck — one question on the carry.", "thr-harbor"))
+    conn.executemany(
+        "INSERT INTO email_recipients (id,email_id,address,display_name,kind) VALUES (?,?,?,?,?)",
+        [
+            ("r1", "e2", "grant@ten31.xyz", "Grant", "to"),
+            ("r2", "e2", "jonathan@ten31.xyz", "Jonathan", "cc"),
+            ("r3", "e2", "lp@harborvine.example", "LP Harbor", "from"),  # from -> not surfaced
+        ])
+    conn.execute("INSERT INTO email_attachments (id,email_id,filename,mime_type,size_bytes,download_status) "
+                 "VALUES ('a1','e2','term_sheet.pdf','application/pdf',20480,'downloaded')")
+    conn.commit()
+    return conn
+
+
+def ids(res):
+    return [e["id"] for e in res["emails"]]
+
+
+def main():
+    conn = make_db()
+
+    # --- baseline: matched live emails only, newest first, tombstoned excluded ---
+    res = _db.query_email_activity(conn)
+    check(ids(res) == ["e8", "e7", "e3", "e2", "e1", "e6"],
+          f"matched live emails newest-first; e5 (tombstoned) + e4 (unmatched) excluded; got {ids(res)}")
+    check(res["count"] == 6 and res["truncated"] is False, "count + not truncated")
+    check("e4" not in ids(res), "unmatched email (no investor link) never surfaces in the panel")
+
+    # --- direction at the email level ---
+    e1 = next(e for e in res["emails"] if e["id"] == "e1")
+    e2 = next(e for e in res["emails"] if e["id"] == "e2")
+    check(e1["direction"] == "outbound", "e1 from our mailbox -> outbound")
+    check(e2["direction"] == "inbound", "e2 from LP -> inbound")
+    check(_db.query_email_activity(conn, direction="outbound")["emails"][0]["id"] == "e1"
+          and len(_db.query_email_activity(conn, direction="outbound")["emails"]) == 1,
+          "direction=outbound returns only e1")
+    check(ids(_db.query_email_activity(conn, direction="inbound")) == ["e8", "e7", "e3", "e2", "e6"],
+          "direction=inbound excludes the outbound e1 (and unmatched e4)")
+
+    # --- mailbox roll-up + per-account filter ---
+    check(set(e2["mailboxes"]) == {"grant@ten31.xyz", "jonathan@ten31.xyz"}, "e2 seen by both mailboxes")
+    check(ids(_db.query_email_activity(conn, account_id="acc-jon")) == ["e8", "e3", "e2"],
+          "account_id=acc-jon returns only emails that mailbox saw")
+
+    # --- date-range filter [since, until) over sent_at ---
+    check(ids(_db.query_email_activity(conn, since="2026-06-07T00:00:00", until="2026-06-11T00:00:00")) == ["e7", "e3"],
+          "date range [06-07, 06-11) -> e7,e3 (excludes 06-11 e8 and earlier e2/e1/e6)")
+    check(ids(_db.query_email_activity(conn, since="2026-06-10T00:00:00")) == ["e8", "e7"],
+          "since=06-10 -> e8,e7 only")
+
+    # --- investor filter: typed keys + legacy bare-id back-compat ---
+    check(set(ids(_db.query_email_activity(conn, investor_id="fund:inv-harbor"))) == {"e2", "e1"},
+          "investor_id=fund:inv-harbor -> e1,e2")
+    check(set(ids(_db.query_email_activity(conn, investor_id="inv-harbor"))) == {"e2", "e1"},
+          "legacy bare id treated as fund: -> e1,e2")
+    check(ids(_db.query_email_activity(conn, investor_id="fund:inv-pacific")) == ["e3"],
+          "fund:inv-pacific resolved through fundraising_contacts -> e3")
+    check(ids(_db.query_email_activity(conn, investor_id="org:org-bridge")) == ["e7"],
+          "org:org-bridge -> e7 (org-domain match)")
+    check(ids(_db.query_email_activity(conn, investor_id="contact:c-solo")) == ["e8"],
+          "contact:c-solo -> e8 (classic-contact match)")
+    check(_db.query_email_activity(conn, investor_id="bogus:x")["emails"] == [],
+          "unknown investor_id key prefix -> match nothing (never silently unfiltered)")
+
+    # --- investor identity roll-up, typed + resolved name ---
+    check(e1["investors"] == [{"id": "fund:inv-harbor", "name": "Harbor & Vine"}], "e1 grid investor resolved")
+    e3 = next(e for e in res["emails"] if e["id"] == "e3")
+    check(e3["investors"] == [{"id": "fund:inv-pacific", "name": "Pacific Capital"}], "e3 resolved via grid contact")
+    e7 = next(e for e in res["emails"] if e["id"] == "e7")
+    check(e7["investors"] == [{"id": "org:org-bridge", "name": "Bridgewater"}],
+          "e7 org-domain match resolves to the org name (not a raw address)")
+    e8 = next(e for e in res["emails"] if e["id"] == "e8")
+    check(e8["investors"] == [{"id": "contact:c-solo", "name": "Nina Park"}],
+          "e8 classic-contact match resolves to the contact name")
+
+    # --- free-text search over subject / snippet / sender ---
+    check(set(ids(_db.query_email_activity(conn, search="Fund III"))) == {"e1", "e2"}, "search subject")
+    check(ids(_db.query_email_activity(conn, search="pacificcap")) == ["e3"], "search sender address")
+    check(ids(_db.query_email_activity(conn, search="deck")) == ["e1"], "search snippet (matched email)")
+    check(ids(_db.query_email_activity(conn, search="buy now")) == [],
+          "unmatched email never surfaces, even by free-text search")
+
+    # --- facets: typed entries spanning grid / org / contact matches ---
+    check([a["email_address"] for a in res["accounts"]] == ["grant@ten31.xyz", "jonathan@ten31.xyz"],
+          "accounts facet sorted")
+    facet_inv = {i["id"] for i in res["investors"]}
+    check(facet_inv == {"fund:inv-harbor", "fund:inv-pacific", "org:org-bridge", "contact:c-solo"},
+          f"investor facet now mirrors the list (grid + org + contact), not just grid; got {facet_inv}")
+    check([i["name"] for i in res["investors"]] == sorted(i["name"] for i in res["investors"]),
+          "facet sorted by display name")
+
+    # --- graveyard: hidden from the picker, but its email stays visible + findable ---
+    check("fund:inv-dead" not in facet_inv, "graveyard investor excluded from the facet dropdown")
+    check("e6" in ids(res), "graveyard investor's email still shows in the unfiltered list (audit completeness)")
+    e6 = next(e for e in res["emails"] if e["id"] == "e6")
+    check(e6["investors"] == [{"id": "fund:inv-dead", "name": "Dead Deal LP"}], "graveyard email still shows its investor chip")
+    check(ids(_db.query_email_activity(conn, investor_id="fund:inv-dead")) == ["e6"],
+          "explicit investor_id filter still works for a graveyard investor")
+    check(ids(_db.query_email_activity(conn, search="deaddeal")) == ["e6"],
+          "graveyard email remains findable by free-text search")
+
+    # --- truncation ---
+    tr = _db.query_email_activity(conn, limit=2)
+    check(tr["count"] == 2 and tr["truncated"] is True, "limit=2 -> truncated")
+
+    # --- detail view (full body + recipients + attachments + identity) ---
+    d = _db.query_email_detail(conn, "e2")
+    check(d is not None and d["body_text"] == "Thanks for the deck — one question on the carry.",
+          "detail returns the full body")
+    check(d["direction"] == "inbound" and set(d["mailboxes"]) == {"grant@ten31.xyz", "jonathan@ten31.xyz"},
+          "detail direction + mailboxes")
+    check([(r["address"], r["kind"]) for r in d["recipients"]] ==
+          [("grant@ten31.xyz", "to"), ("jonathan@ten31.xyz", "cc")],
+          "detail recipients = to/cc only (from is excluded)")
+    check([a["filename"] for a in d["attachments"]] == ["term_sheet.pdf"], "detail lists attachments")
+    check(d["investors"] == [{"id": "fund:inv-harbor", "name": "Harbor & Vine"}], "detail resolves investor identity")
+    check(_db.query_email_detail(conn, "e5") is None,
+          "detail of a tombstoned-only email -> None (soft-delete on the sighting)")
+    check(_db.query_email_detail(conn, "nope") is None, "detail of a missing id -> None")
+
+    conn.close()
+
+    # --- route enforces admin server-side ---
+    test_route_admin_only()
+    # --- semantic content-search route (hydrate + soft-delete + 503) ---
+    test_search_route()
+
+    if FAILS:
+        print(f"\nFAILED ({len(FAILS)})")
+        for f in FAILS:
+            print("  - " + f)
+        sys.exit(1)
+    print("\nALL PASS (email-activity panel)")
+
+
+class FakeHandler:
+    def __init__(self, user, params=None):
+        self._user = user
+        self._params = params or {}
+        self.json = None
+        self.err = None
+        self.code = None
+
+    def get_user(self):
+        return self._user
+
+    def get_query_params(self):
+        return self._params
+
+    def send_json(self, obj):
+        self.json = obj
+
+    def send_error_json(self, msg, code):
+        self.err = msg
+        self.code = code
+
+
+def test_route_admin_only():
+    try:
+        from email_integration import routes
+    except Exception as e:  # pragma: no cover - optional deps missing in some dev envs
+        print(f"  SKIP route admin test (routes import failed: {e})")
+        return
+
+    h = FakeHandler(None)
+    routes._h_activity(h)
+    check(h.code == 401 and h.json is None, "route: no user -> 401")
+
+    h = FakeHandler({"role": "member", "user_id": "u1"})
+    routes._h_activity(h)
+    check(h.code == 403 and h.json is None, "route: member -> 403 (admin enforced server-side)")
+
+
+def test_search_route():
+    try:
+        from email_integration import routes
+    except Exception as e:  # pragma: no cover
+        print(f"  SKIP search route test (routes import failed: {e})")
+        return
+
+    # Hydration source = a fresh fully-populated in-memory DB each call (the handler
+    # opens + closes its own conn). Retrieval is stubbed — no Spark/Qdrant in tests.
+    routes._conn = make_db
+    routes._semantic_email_search = lambda query, top_k: [
+        {"score": 0.91, "text": "carry discussion\nand terms", "payload": {"source_id": "e2", "lp_name": "Harbor & Vine"}},
+        {"score": 0.80, "text": "gone", "payload": {"source_id": "e5", "lp_name": "Ghost"}},      # tombstoned -> drop
+        {"score": 0.70, "text": "n/a", "payload": {"source_id": "missing", "lp_name": "Nobody"}},  # missing -> drop
+    ]
+    h = FakeHandler({"role": "admin"}, {"q": "carry"})
+    routes._h_search(h)
+    check(h.json and [r["email_id"] for r in h.json["results"]] == ["e2"],
+          f"content search drops tombstoned + missing, keeps live e2; got {h.json and [r['email_id'] for r in h.json['results']]}")
+    top = h.json["results"][0]
+    check(top["lp_name"] == "Harbor & Vine" and top["score"] == 0.91 and top["subject"] == "Re: Fund III update",
+          "hit carries lp_name + score + hydrated subject")
+    check("\n" not in top["excerpt"], "excerpt is newline-flattened")
+
+    # empty query short-circuits (no retrieval call)
+    h = FakeHandler({"role": "admin"}, {"q": ""})
+    routes._h_search(h)
+    check(h.json == {"query": "", "results": []}, "empty query -> empty results")
+
+    # retrieval failure -> clean 503 (Spark/Qdrant down)
+    def _boom(query, top_k):
+        raise RuntimeError("spark down")
+    routes._semantic_email_search = _boom
+    h = FakeHandler({"role": "admin"}, {"q": "x"})
+    routes._h_search(h)
+    check(h.code == 503, f"retrieval failure -> 503, got {h.code}")
+
+    # admin enforced
+    h = FakeHandler({"role": "member"}, {"q": "x"})
+    routes._h_search(h)
+    check(h.code == 403, "content search admin-enforced server-side")
+
+
+if __name__ == "__main__":
+    main()

backend/email_integration/test_sync_ready.py

@@ -0,0 +1,72 @@
+#!/usr/bin/env python3
+"""Regression test for list_sync_ready_accounts (v0.1.0:104).
+
+Guards the Bug-B fix: a transient network timeout used to flip a mailbox to terminal
+sync_status='error', which the old `IN ('pending','active')` filter excluded forever —
+so a single blip dark-listed a mailbox until a manual kick. The new filter:
+  * always includes 'pending' / 'active' / 'retrying' (the transient-retry state), and
+  * re-includes 'error' accounts whose last attempt was over an hour ago (gentle backoff,
+    so a fixed credential self-heals and the pre-fix stuck mailboxes recover on deploy).
+Synthetic data only (guardrail #9).
+Run: cd backend && python3 email_integration/test_sync_ready.py
+"""
+import os
+import sqlite3
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+from email_integration import db as edb  # noqa: E402
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+def make_account(conn, email, status, *, enabled=1, last_synced_sql="NULL"):
+    aid = edb.upsert_account(conn, user_id="u-" + email,
+                             email_address=email, auth_method="dwd")
+    conn.execute(
+        f"UPDATE email_accounts SET sync_status=?, sync_enabled=?, "
+        f"last_synced_at={last_synced_sql} WHERE id=?",
+        (status, enabled, aid),
+    )
+    conn.commit()
+    return aid
+
+
+def main():
+    conn = sqlite3.connect(":memory:")
+    conn.row_factory = sqlite3.Row
+    edb.apply_migrations(conn.cursor())
+
+    make_account(conn, "active@t.xyz", "active", last_synced_sql="datetime('now','-5 minutes')")
+    make_account(conn, "retrying@t.xyz", "retrying", last_synced_sql="datetime('now','-5 minutes')")
+    make_account(conn, "pending@t.xyz", "pending", last_synced_sql="NULL")
+    make_account(conn, "error_stale@t.xyz", "error", last_synced_sql="datetime('now','-2 hours')")
+    make_account(conn, "error_recent@t.xyz", "error", last_synced_sql="datetime('now','-5 minutes')")
+    make_account(conn, "error_neversync@t.xyz", "error", last_synced_sql="NULL")
+    make_account(conn, "disabled@t.xyz", "active", enabled=0, last_synced_sql="datetime('now','-5 minutes')")
+
+    ready = {r["email_address"] for r in edb.list_sync_ready_accounts(conn)}
+
+    check("active@t.xyz" in ready, "healthy 'active' is ready")
+    check("retrying@t.xyz" in ready, "transient 'retrying' is ready (fast retry)")
+    check("pending@t.xyz" in ready, "'pending' is ready")
+    check("error_stale@t.xyz" in ready, "'error' last attempted >1h ago is ready (backoff elapsed → recovers stuck mailbox)")
+    check("error_neversync@t.xyz" in ready, "'error' never synced (NULL last attempt) is ready")
+    check("error_recent@t.xyz" not in ready, "'error' attempted <1h ago is NOT ready (gentle backoff, no hammering)")
+    check("disabled@t.xyz" not in ready, "sync_enabled=0 is never ready")
+
+    print()
+    if FAILS:
+        print(f"{len(FAILS)} FAILED")
+        sys.exit(1)
+    print("ALL PASS (sync ready filter)")
+
+
+if __name__ == "__main__":
+    main()

backend/ingest/chunking.py

@@ -4,7 +4,7 @@ Maps each CRM record type to one or more chunks per docs/EMBEDDINGS.md:
   * one chunk per communications row          (doc_type = the comm type)
   * one chunk per MATCHED email               (doc_type = email; body only when matched)
   * one chunk per fundraising_investors notes LINE (the outreach log; split per line)
-  * one chunk each for free-text fields: contacts.notes, lp_profiles.notes,
+  * one chunk each for free-text fields: contacts.notes,
     opportunities (description + next_step), organizations.description
 
 Each chunk carries a canonical `lp_id` (resolved via entity_links) and a `date_ts`
@@ -104,13 +104,6 @@ def build_chunks(conn):
         chunks.append(_mk(f"contacts.notes:{r['id']}", lp, lp_name, person,
                           "contact_note", to_epoch(r["updated_at"]), r["notes"], "contacts", r["id"]))
 
-    # lp_profiles.notes
-    for r in conn.execute("""SELECT lp.id, lp.contact_id, lp.notes, lp.updated_at
-                             FROM lp_profiles lp WHERE lp.notes IS NOT NULL AND lp.notes <> '' AND lp.deleted_at IS NULL"""):
-        lp, lp_name, person = _contact_lp(r["contact_id"], person_canon, org_canon, name, contact_org)
-        chunks.append(_mk(f"lp_profiles.notes:{r['id']}", lp, lp_name, person,
-                          "lp_note", to_epoch(r["updated_at"]), r["notes"], "lp_profiles", r["id"]))
-
     # opportunities (description + next_step)
     for r in conn.execute("""SELECT id, contact_id, name, description, next_step, updated_at
                              FROM opportunities WHERE deleted_at IS NULL"""):

backend/ingest/entity_resolution.py

@@ -8,7 +8,6 @@ layer created by migration 0001:
     fundraising_investors  ─┴─►  canonical_entities (entity_kind = lp | organization)
     contacts               ─┐
     fundraising_contacts   ─┴─►  canonical_entities (entity_kind = person)
-    lp_profiles            ───►  linked to its contact's person entity
 
 Every source row is recorded in `entity_links` so any name variant resolves to
 one canonical id. This is the DETERMINISTIC tier — it merges only what we can
@@ -184,7 +183,7 @@ def resolve_people(conn, org_canon_by_orgid, org_canon_by_fundinv, merge_map=Non
     people — each is matched to a contact-person and recorded only as a member_of
     edge to its investor entity (the grid's 'Contacts' column says who belongs to
     which investor). This is what stops the double-count.
-    Returns contact_id -> person canonical id (for lp_profiles)."""
+    Returns contact_id -> person canonical id."""
     merge_map = merge_map or {}
     contact_to_person = {}
     person_meta = {}
@@ -245,12 +244,6 @@ def resolve_people(conn, org_canon_by_orgid, org_canon_by_fundinv, merge_map=Non
             _link(conn, cid, "fundraising_contacts", r["id"], email or name_norm, mk, 0.95 if mk == "grid_link" else 0.9)
             _member_of(conn, cid, inv_canon)
 
-    # lp_profiles -> the person entity of its contact
-    for r in conn.execute("SELECT id, contact_id FROM lp_profiles WHERE deleted_at IS NULL"):
-        cid = contact_to_person.get(r["contact_id"])
-        if cid:
-            _link(conn, cid, "lp_profiles", r["id"], r["contact_id"], "contact_fk", 1.0)
-
     return person_meta
 
 

backend/ingest/llm.py

@@ -26,6 +26,31 @@ def chat(prompt, system=None, max_tokens=200, temperature=0.0):
     return (data["choices"][0]["message"].get("content") or "").strip()
 
 
+def chat_vision(prompt, image_b64, mime="image/jpeg", system=None, max_tokens=600, temperature=0.0):
+    """Multimodal chat: a text prompt + one base64 image to the local VL model via Spark Control.
+
+    Same endpoint and model as chat() — the daily-driver Qwen is vision-capable (capabilities
+    [vision, reasoning]); the only difference is the user message's `content` is the OpenAI
+    multimodal array (a text part + an image_url data-URI), which Spark Control forwards to vLLM
+    unchanged (it's a dumb passthrough). The server downscales to its max_pixels cap, so a
+    full-res phone photo is fine. Thinking stays off for fast, literal output."""
+    messages = []
+    if system:
+        messages.append({"role": "system", "content": system})
+    messages.append({"role": "user", "content": [
+        {"type": "text", "text": prompt},
+        {"type": "image_url", "image_url": {"url": f"data:{mime};base64,{image_b64}"}},
+    ]})
+    body = {"model": config.CHAT_MODEL, "messages": messages,
+            "temperature": temperature, "max_tokens": max_tokens,
+            "chat_template_kwargs": {"enable_thinking": False}}
+    status, data = http_util.request("POST", f"{config.SPARK_CONTROL_URL}/v1/chat/completions",
+                                     body, verify=config.SPARK_VERIFY_TLS)
+    if status != 200:
+        raise RuntimeError(f"/v1/chat/completions (vision) -> {status}: {data}")
+    return (data["choices"][0]["message"].get("content") or "").strip()
+
+
 def chat_json(prompt, system=None, max_tokens=200):
     """Chat and parse the first JSON object from the reply (tolerant of fences)."""
     raw = chat(prompt, system=system, max_tokens=max_tokens)

backend/ingest/sync.py

@@ -34,7 +34,7 @@ import entity_resolution as er
 import qdrant_io
 
 _CHANGE_TABLES = [("communications", "communications"), ("contacts", "contacts"),
-                  ("lp_profiles", "lp_profiles"), ("opportunities", "opportunities"),
+                  ("opportunities", "opportunities"),
                   ("organizations", "organizations"), ("fundraising_investors", "fundraising_investors")]
 
 
@@ -63,7 +63,7 @@ def _state_set(conn, key, value):
 def _deleted_source_ids(conn, since):
     """CRM records soft-deleted since the watermark — their chunks get pruned."""
     ids = set()
-    for tbl in ("contacts", "organizations", "opportunities", "communications", "lp_profiles"):
+    for tbl in ("contacts", "organizations", "opportunities", "communications"):
         try:
             for r in conn.execute(f"SELECT id FROM {tbl} WHERE deleted_at IS NOT NULL AND deleted_at > ?", (since,)):
                 ids.add(r["id"])

backend/ingest/test_entity_resolution.py

@@ -12,7 +12,7 @@ Asserts the SAFE fix:
   3. a grid contact that can't be PROVABLY matched mints NOTHING (no duplicate
      person, no cross-firm name guess) — the count stays correct,
   4. targeted cleanup soft-deletes a stale grid-only "twin" (person with no
-     contacts link) and a superseded 'lp'/'organization' row, with no enrichment,
+     contacts link), with no enrichment,
   5. cleanup PRESERVES a grid-only person that carries enrichment (guardrail #3),
   6. a re-emitted id is UN-tombstoned (no permanent burial),
   7. re-running is idempotent.
@@ -58,10 +58,9 @@ CREATE TABLE contacts (
 CREATE TABLE organizations (id TEXT PRIMARY KEY, name TEXT, email TEXT);
 CREATE TABLE fundraising_investors (id TEXT PRIMARY KEY, investor_name TEXT);
 CREATE TABLE fundraising_contacts (id TEXT PRIMARY KEY, full_name TEXT, email TEXT, investor_id TEXT, contact_id TEXT);
-CREATE TABLE lp_profiles (id TEXT PRIMARY KEY, contact_id TEXT, deleted_at TEXT);
 """
 
-SEEDED = ("per_TWIN", "per_ENR", "lp_OLD")
+SEEDED = ("per_TWIN", "per_ENR")
 
 
 def seed(db):
@@ -94,16 +93,14 @@ def seed(db):
               "('per_ENR','person','Enriched Orphan','entity_resolution','warm')")
     c.execute("INSERT INTO entity_links (id, canonical_id, source_model, source_id, match_value, match_kind, confidence, created_at) "
               "VALUES ('l_enr','per_ENR','fundraising_contacts','gy','enr','name_org',0.8,'t')")
-    # Superseded pre-:48 kind -> prune
-    c.execute("INSERT INTO canonical_entities (id, entity_kind, display_name, source) VALUES "
-              "('lp_OLD','lp','Old LP Row','entity_resolution')")
     c.commit()
     c.close()
 
 
 def resolved_persons(db):
     c = sqlite3.connect(db)
-    q = "SELECT COUNT(*) FROM canonical_entities WHERE entity_kind='person' AND deleted_at IS NULL AND id NOT IN (?,?,?)"
+    ph = ",".join("?" * len(SEEDED))
+    q = f"SELECT COUNT(*) FROM canonical_entities WHERE entity_kind='person' AND deleted_at IS NULL AND id NOT IN ({ph})"
     n = c.execute(q, SEEDED).fetchone()[0]
     c.close()
     return n
@@ -127,10 +124,11 @@ def grid_match_kinds(db):
 def minted_from_grid(db):
     """Persons minted directly from a grid row (the bug). Should be 0 after the fix."""
     c = sqlite3.connect(db)
-    n = c.execute("""SELECT COUNT(DISTINCT l.canonical_id) FROM entity_links l
+    ph = ",".join("?" * len(SEEDED))
+    n = c.execute(f"""SELECT COUNT(DISTINCT l.canonical_id) FROM entity_links l
                        JOIN canonical_entities ce ON ce.id=l.canonical_id AND ce.deleted_at IS NULL
                       WHERE l.source_model='fundraising_contacts' AND l.match_kind IN ('name_org','exact_email')
-                        AND l.canonical_id NOT IN (?,?,?)""", SEEDED).fetchone()[0]
+                        AND l.canonical_id NOT IN ({ph})""", SEEDED).fetchone()[0]
     c.close()
     return n
 
@@ -162,12 +160,11 @@ def main():
     check(mk.get("grid_assoc", 0) == 2, f"two grid contacts matched back via grid_assoc (got {mk.get('grid_assoc',0)})")
     check(mk.get("grid_link", 0) == 1, f"one grid contact linked via explicit contact_id (grid_link==1, got {mk.get('grid_link',0)})")
 
-    # Targeted cleanup: stale grid-only twin + superseded 'lp' row tombstoned...
+    # Targeted cleanup: stale grid-only twin tombstoned...
     check(deleted_at(db, "per_TWIN") is not None, "stale grid-only twin 'per_TWIN' tombstoned")
-    check(deleted_at(db, "lp_OLD") is not None, "superseded 'lp' row 'lp_OLD' tombstoned")
     # ...enriched grid-only person preserved.
     check(deleted_at(db, "per_ENR") is None, "enriched grid-only person 'per_ENR' PRESERVED (has segment)")
-    check(counts1.get("pruned_stale", 0) == 2, f"exactly 2 stale rows pruned (got {counts1.get('pruned_stale')})")
+    check(counts1.get("pruned_stale", 0) == 1, f"exactly 1 stale row pruned (got {counts1.get('pruned_stale')})")
 
     # Un-tombstone: soft-delete a real contact-person, then re-run -> it comes back.
     alice = er._eid("per", "e|alice@x.com")

backend/matrix_intake/Dockerfile

@@ -0,0 +1,22 @@
+# Container image for the Matrix intake bot — turns it from a bare nohup process into a managed
+# service (docker compose `restart: unless-stopped` survives a Spark reboot).
+#
+# Build context is the REPO ROOT (see ../../docker-compose.yml), not this directory: the bot is
+# NOT self-contained — spark.py reaches into backend/ingest/{llm,config,http_util}.py (stdlib
+# only) via sys.path, so the image must carry both trees with the repo layout preserved. That
+# keeps settings.load_env's REPO_ROOT (three dirs up from settings.py) = /app and spark.py's
+# ingest path = /app/backend/ingest both correct at runtime.
+FROM python:3.12-slim
+
+WORKDIR /app
+
+# The only third-party dep is matrix-nio; the reused ingest Spark client is pure stdlib.
+COPY backend/matrix_intake/requirements.txt ./requirements.txt
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY backend/matrix_intake/ ./backend/matrix_intake/
+COPY backend/ingest/ ./backend/ingest/
+
+# .env (Matrix + CRM + Spark creds) is mounted read-only at /app/.env at runtime — never baked.
+# `-u` keeps stdout/stderr unbuffered so `docker logs` shows the bot's lifecycle lines live.
+CMD ["python", "-u", "backend/matrix_intake/bot.py"]

backend/matrix_intake/README.md

@@ -0,0 +1,41 @@
+# Matrix intake bot
+
+Turns a typed message in a dedicated Matrix room into a proposed fundraising-grid add/edit,
+gated on in-thread human approval before any write. Runs as its own process (on the Spark),
+separate from the CRM. Full design + rules: `docs/guides/matrix-intake.md`.
+
+## Run
+
+```bash
+# 1. Install the one third-party dep (isolated to this component — NOT the CRM runtime)
+python3 -m pip install -r requirements.txt          # matrix-nio
+
+# 2. Fill the MATRIX_* and CRM_BOT_* vars in the repo .env (see ../../.env.example),
+#    and create a dedicated CRM user for CRM_BOT_USERNAME/PASSWORD (admin → invite user).
+
+# 3. Start the listener
+python3 bot.py
+```
+
+It primes the Matrix sync past history (no backlog replay), then listens. Post a message in
+the intake room; it replies in a thread with the parsed proposal. Reply **yes** to commit,
+**edit field=value** to change a field, or **no** to discard.
+
+## Layout
+
+- `bot.py` — entrypoint: connect, prime-then-listen, dispatch (lifts matrix-bridge's plumbing).
+- `parse.py` — message → structured proposal via local Qwen (`spark.py` → `backend/ingest/llm.py`).
+- `proposals.py` — in-memory pending-proposal store + the yes/edit/no state machine.
+- `crm_client.py` — login + `GET /api/intake/match` + write via `POST /api/fundraising/log-communication`.
+- `matrix_io.py` — message splitting, thread-root detection, threaded-reply sender.
+- `settings.py` — Matrix + CRM-API config (named `settings`, not `config`, to avoid shadowing `ingest/config`).
+
+## Test (offline)
+
+```bash
+python3 test_parse.py && python3 test_proposals.py && python3 test_crm_client.py
+# endpoint + create→match contract (boots the real server against a temp DB):
+cd ../ && python3 test_intake_endpoints.py
+```
+
+Live Matrix behavior needs creds + `matrix-nio` and can only be smoke-tested on the Spark.

backend/matrix_intake/__init__.py

@@ -0,0 +1,7 @@
+"""Matrix intake bot — a dedicated Matrix room that turns a typed message into a
+proposed fundraising-grid add/edit, gated on in-thread human approval before any write.
+
+Separate process from the CRM (its only third-party dep, matrix-nio, lives here, never
+in the stdlib CRM runtime). Parses with local Qwen via Spark Control; on approval, writes
+through the CRM's own API. See docs/guides/matrix-intake.md and ROADMAP.md.
+"""

backend/matrix_intake/bot.py

@@ -0,0 +1,441 @@
+#!/usr/bin/env python3
+"""Matrix intake bot — entrypoint.
+
+A top-level message in the dedicated intake room is parsed (local Qwen via Spark Control)
+into a proposed fundraising-grid add/edit and posted back IN A THREAD. The team member
+replies in that thread — **yes** / **edit field=value** / **no** — and only on **yes** does
+the bot write, through the CRM's own API. Nothing is ever written autonomously.
+
+Runs as its own process (its matrix-nio dep is isolated here, never in the CRM runtime).
+Lifts matrix-bridge's prime-then-listen + threaded-reply plumbing. Config: repo .env.
+"""
+import asyncio
+import base64
+
+from nio import AsyncClient, MatrixRoom, MessageDirection, RoomMessageImage, RoomMessageText
+
+import crm_client
+import email_proposals
+import matrix_io
+import parse
+import proposals
+import query
+import settings
+import spark
+
+UNCLEAR_HELP = (
+    "🤔 I couldn't tell what to record. Try e.g.\n"
+    "`New investor: Acme Capital — Jane Doe <jane@acme.com>, met at the Austin conf`\n"
+    "or a note like `Note for Acme Capital: wants the Q3 deck, follow up next week`."
+)
+
+EMAIL_POLL_SEC = 20  # how often the bot polls the CRM for new/decided email-activity proposals
+MAX_THREAD_SCAN_PAGES = 8  # how far back to scan for a resolved thread's replies before redacting
+
+
+async def main():
+    mx = settings.matrix_settings()
+    client = AsyncClient(mx["homeserver"], mx["user_id"])
+    client.restore_login(user_id=mx["user_id"], device_id=mx["device_id"], access_token=mx["token"])
+    say = matrix_io.make_say(client)
+    nudge = matrix_io.make_reply(client)
+    store = proposals.ProposalStore()
+    intake_room = mx["intake_room"]
+    roster = settings.team_roster()  # frames the parse: teammates do outreach, aren't prospects
+    if roster:
+        print(f"matrix-intake: team roster loaded ({len(roster)} names)", flush=True)
+    review_room = settings.email_review_room()  # CRM-drafted email proposals (empty → feature off)
+    query_room = settings.query_room()  # dedicated read-only Q&A room (empty → use the intake trigger)
+    email_threads = {}  # Matrix thread-root event_id -> {id, investor_name, note} for an email proposal
+
+    async def handle_intake(room_id, root, text, source="matrix_intake"):
+        # `source` tags provenance for the eventual commit: "matrix_intake" for a typed note,
+        # "matrix_card" when the text came from a scanned business card (on_image). Everything
+        # else about the flow is identical — that's the whole point of transcribe-then-reuse.
+        # A bare yes/no/approve typed in the MAIN timeline (not inside a proposal's thread) is
+        # an easy slip — point the user back to the thread rather than parse it as a new intake.
+        action, _ = proposals.interpret_reply(text)
+        if action in ("approve", "reject") and store.any_pending():
+            await nudge(room_id, "👉 To approve, reject, or edit a proposal, open its **thread** "
+                                 "and reply there — the note is in the thread.", root)
+            return
+        try:
+            proposal = await asyncio.to_thread(parse.parse_message, text, roster=roster)
+        except Exception as exc:  # Spark/Qwen unreachable or bad response
+            await say(room_id, f"⚠️ couldn't reach the local parser: {str(exc)[:200]}", root)
+            return
+        if proposal["intent"] == "unclear":
+            await say(room_id, UNCLEAR_HELP, root)
+            return
+        proposal["_source"] = source  # rides through to commit (control key, survives dict() copies)
+        # Resolve new-vs-existing against the CRM matcher (read-only). Degrade gracefully if the
+        # CRM is unreachable — still propose as new, just without match/candidate hints.
+        match, candidates = None, []
+        try:
+            res = await asyncio.to_thread(crm_client.match, proposal)
+            match = res.get("match")
+            candidates = res.get("candidates") or []
+        except Exception:
+            pass
+        if match:
+            # Confident exact match → auto-attach the note to that investor (no disambiguation).
+            proposal["intent"] = "meeting_note"
+            proposal["_match_id"] = match["id"]
+            proposal["_stage"] = "approval"
+            store.put(root, proposal)
+            hint = (f"\n\n🔎 Looks like an existing investor: **{match['name']}** — "
+                    "this will append a note to them.")
+            await say(room_id, proposals.render(proposal) + hint, root)
+            await nudge(room_id, proposals.summary_line(proposal), root)
+            return
+        if candidates:
+            # No exact match but near-misses exist → make the human pick one or confirm "new",
+            # so a typo'd/near-duplicate name can't silently create a second investor.
+            proposal["_stage"] = "disambiguate"
+            proposal["_candidates"] = candidates
+            store.put(root, proposal)
+            await say(room_id, proposals.render_disambiguation(proposal), root)
+            await nudge(room_id, proposals.disambiguation_nudge(proposal), root)
+            return
+        # Genuinely new — straight to the new-investor approval card.
+        proposal["_stage"] = "approval"
+        store.put(root, proposal)
+        await say(room_id, proposals.render(proposal), root)
+        # Also drop a brief, un-threaded reply in the main timeline so the proposal isn't
+        # easy to miss inside a thread (the full card + yes/edit/no stay in the thread).
+        await nudge(room_id, proposals.summary_line(proposal), root)
+
+    async def handle_card(room_id, event):
+        """A photo in the intake room → transcribe the business card on the local VL model, then
+        hand the transcription to the SAME intake flow as a typed note (parse → match → approve).
+        The only new step is image → text; everything downstream is reused. The transcription is
+        also the source text the email-integrity check runs against, so a mis-read address can't
+        slip in unapproved."""
+        mxc = getattr(event, "url", None)
+        if not mxc:
+            # Unencrypted images carry a plain mxc:// url; an encrypted room delivers a different
+            # event class entirely (we don't register for it), so this only guards the odd case.
+            await say(room_id, "📇 I can only read unencrypted images right now.", event.event_id)
+            return
+        await say(room_id, "📇 Reading the card…", event.event_id)  # vision is slower — ack first
+        try:
+            resp = await client.download(mxc=mxc)
+            data = getattr(resp, "body", None)
+            if not isinstance(data, (bytes, bytearray)):  # a DownloadError carries no bytes
+                raise RuntimeError(getattr(resp, "message", None) or "image download failed")
+            mime = getattr(resp, "content_type", None) or "image/jpeg"
+            b64 = base64.b64encode(data).decode("ascii")
+            text = await asyncio.to_thread(spark.transcribe_card, b64, mime)
+        except Exception as exc:
+            await say(room_id, f"⚠️ couldn't read the card: {str(exc)[:200]}", event.event_id)
+            return
+        if len(text.strip()) < 5:
+            await say(room_id, "📇 I couldn't read any text on that card — try a clearer, "
+                               "well-lit photo taken straight-on.", event.event_id)
+            return
+        # Frame the raw transcription so the existing extractor reads it as a new-investor intake;
+        # the transcription itself is what email-integrity is checked against.
+        framed = "New investor — from a business card:\n" + text.strip()
+        await handle_intake(room_id, event.event_id, framed, source="matrix_card")
+
+    async def handle_query(room_id, root, question):
+        """A read-only NL question ('@bot …' / '?…') — translate + run it on the BOX (local Qwen,
+        nothing leaves the box) and post the answer in a thread. No write path, no approval gate:
+        it only reads curated, parameterized queries. The endpoint returns its structured result
+        even on a soft no-match / model-down, so we render that; a transport/auth failure raises
+        and we show a brief error."""
+        try:
+            result = await asyncio.to_thread(crm_client.nl_query, question)
+        except Exception as exc:
+            await say(room_id, f"⚠️ couldn't run that query: {str(exc)[:200]}", root)
+            return
+        await say(room_id, query.render_answer(result), root)
+
+    async def handle_reply(room_id, root, text):
+        # Claim the proposal synchronously — BEFORE any await — so a second reply that
+        # arrives while a commit is in flight can't double-process it. asyncio is
+        # cooperative: nothing else runs between here and the first await below, so the
+        # pop is atomic w.r.t. other Matrix events.
+        proposal = store.pop(root)
+        if proposal is None:
+            return
+        if proposal.get("_stage") == "disambiguate":
+            await handle_disambiguation(room_id, root, text, proposal)
+            return
+
+        action, payload = proposals.interpret_reply(text)
+        if action == "approve":
+            try:
+                summary = await asyncio.to_thread(crm_client.commit, proposal)
+            except Exception as exc:
+                store.put(root, proposal)  # commit failed — restore so the user can retry
+                await say(room_id, f"⚠️ write failed, nothing committed: {exc}", root)
+                return
+            # Committed → clear the whole thread (card + ack + nudge + the user's note/photo),
+            # like the email-review room. The thread vanishing is the acknowledgment; a confirmation
+            # reply would just keep it alive (and need redacting too). Needs the bot's redact/mod
+            # power in the intake room to clear the user's own messages — else those linger.
+            await redact_thread(room_id, root)
+        elif action == "reject":
+            await redact_thread(room_id, root)
+        elif action == "edit":
+            field, value = payload
+            proposal = proposals.apply_edit(proposal, field, value)
+            store.put(root, proposal)  # keep it pending (edited) for the next reply
+            await say(room_id, "✏️ Updated:\n\n" + proposals.render(proposal), root)
+        else:
+            # Not yes/no/edit-grammar → treat it as a natural-language revision instruction and
+            # re-run it through local Qwen (no Claude, no scrub). The human still approves the
+            # revised card, so the draft→approve gate holds.
+            try:
+                revised = await asyncio.to_thread(parse.revise, proposal, text, roster=roster)
+            except Exception as exc:
+                store.put(root, proposal)
+                await say(room_id, f"⚠️ couldn't apply that change ({str(exc)[:200]}).\n\nReply **yes** "
+                                   "to commit, **no** to discard, **edit field=value**, or rephrase.", root)
+                return
+            if proposals.same_fields(proposal, revised):
+                store.put(root, proposal)
+                await say(room_id, "I didn't catch a change there. Reply **yes** to commit, **no** "
+                                   "to discard, **edit field=value**, or tell me what to change.", root)
+                return
+            store.put(root, revised)
+            await say(room_id, "✏️ Updated:\n\n" + proposals.render(revised), root)
+
+    async def handle_disambiguation(room_id, root, text, proposal):
+        cands = proposal.get("_candidates") or []
+        action, payload = proposals.interpret_disambiguation(text, len(cands))
+        if action == "pick":
+            updated = proposals.attach_to_candidate(proposal, cands[payload])
+            store.put(root, updated)
+            await say(room_id, "✏️ Will log against the existing investor:\n\n"
+                               + proposals.render(updated), root)
+        elif action == "new":
+            updated = proposals.promote_to_new(proposal)
+            store.put(root, updated)
+            await say(room_id, "➕ OK — adding as a new investor:\n\n"
+                               + proposals.render(updated), root)
+        elif action == "reject":
+            await redact_thread(room_id, root)  # discard → clear the thread, like an approve
+        else:  # unrecognized — re-show the shortlist
+            store.put(root, proposal)
+            await say(room_id, "I didn't catch that.\n\n" + proposals.render_disambiguation(proposal), root)
+
+    async def redact_card(room_id, event_id):
+        """Redact one event in `room_id` (best-effort). Redacting our OWN message needs no special
+        power; redacting someone else's message (a human reply, or the user's original card photo /
+        intake note) needs the bot to hold a redact/mod power level in that room."""
+        try:
+            await client.room_redact(room_id, event_id, reason="proposal resolved")
+        except Exception as exc:
+            print(f"matrix-intake: could not redact {event_id}: {exc}", flush=True)
+
+    async def redact_thread(room_id, root):
+        """Clear a resolved thread in `room_id`: redact the root AND every message that hangs off it
+        — the m.thread children (cards/acks/human replies) AND the main-timeline **nudge** (a plain
+        m.in_reply_to reply, not a thread child), so the thread drops out of both the threads view
+        and the timeline. For email-review the root is the bot's card; for intake it's the USER'S
+        own note/photo, so clearing it (and the human reply) needs the bot's redact/mod power in that
+        room — without it those just no-op and linger. Replies are found by scanning recent history
+        from the current sync token (the triggering reply is already synced, so a backward scan
+        includes it)."""
+        await redact_card(room_id, root)
+        token = getattr(client, "next_batch", None)
+        if not token:
+            return
+        try:
+            scanned = 0
+            for _ in range(MAX_THREAD_SCAN_PAGES):
+                resp = await client.room_messages(room_id, start=token,
+                                                  direction=MessageDirection.back, limit=100)
+                chunk = getattr(resp, "chunk", None)
+                if not chunk:
+                    break
+                for ev in chunk:
+                    rel = ((getattr(ev, "source", None) or {}).get("content", {}) or {}).get("m.relates_to") or {}
+                    in_reply = (rel.get("m.in_reply_to") or {}).get("event_id")
+                    # A thread child carries rel_type=m.thread + event_id==root (the cards/acks +
+                    # the human's yes/no replies — any sender). The un-threaded nudge is the BOT's
+                    # own plain reply to root (only m.in_reply_to==root, no rel_type); gate that
+                    # clause to our sender so we don't also redact an unrelated human plain-reply
+                    # to the same root (root itself is already redacted above).
+                    is_thread_child = rel.get("rel_type") == "m.thread" and rel.get("event_id") == root
+                    is_own_nudge = in_reply == root and getattr(ev, "sender", None) == mx["user_id"]
+                    if is_thread_child or is_own_nudge:
+                        await redact_card(room_id, ev.event_id)
+                token = getattr(resp, "end", None)
+                scanned += len(chunk)
+                if not token or scanned > 1000:
+                    break
+        except Exception as exc:
+            print(f"matrix-intake: thread reply cleanup failed for {root}: {exc}", flush=True)
+
+    async def handle_email_reply(room_id, root, text):
+        """An in-thread reply to a CRM-drafted email-proposal card: yes commits, no dismisses, and
+        anything else is a natural-language revision of the note (re-drafted by local Qwen; the
+        human still approves the revised note, so the draft→approve gate holds). On a conclusive
+        decision the card is redacted so the room clears down to only what still needs handling."""
+        item = email_threads.get(root)
+        if item is None:
+            return  # a threaded reply we don't own (or already resolved)
+        decision = email_proposals.interpret(text)
+        if decision == "approve":
+            # Claim before the await (double-approve guard, like the intake commit path).
+            email_threads.pop(root, None)
+            try:
+                await asyncio.to_thread(crm_client.decide_email_proposal, item["id"], "approve", item.get("note"))
+            except Exception as exc:
+                email_threads[root] = item  # restore for retry
+                await say(room_id, email_proposals.frame(f"⚠️ couldn't add it ({str(exc)[:200]}). Reply **yes** to retry, **no** to dismiss."), root)
+                return
+            # Success → clear the whole thread (card + replies). No confirmation: the thread
+            # vanishing is the acknowledgment, and a confirmation reply would keep it alive.
+            await redact_thread(review_room, root)
+        elif decision == "reject":
+            email_threads.pop(root, None)
+            try:
+                await asyncio.to_thread(crm_client.decide_email_proposal, item["id"], "dismiss")
+            except Exception as exc:
+                email_threads[root] = item
+                await say(room_id, email_proposals.frame(f"⚠️ couldn't dismiss it ({str(exc)[:200]}). Try again."), root)
+                return
+            await redact_thread(review_room, root)
+        else:
+            try:
+                new_note = await asyncio.to_thread(email_proposals.revise_note, item.get("note") or "", text)
+            except Exception as exc:
+                await say(room_id, email_proposals.frame(f"⚠️ couldn't revise that ({str(exc)[:200]}). Reply **yes** to add as-is, "
+                                   "**no** to dismiss, or rephrase."), root)
+                return
+            if not new_note:
+                await say(room_id, email_proposals.frame("I didn't catch a change. Reply **yes** to add the note as-is, **no** to "
+                                   "dismiss, or tell me how to change it."), root)
+                return
+            item["note"] = new_note
+            email_threads[root] = item
+            await say(room_id, email_proposals.frame(f"✏️ Updated draft note:\n\n{new_note}\n\nReply **yes** to add it, **no** to "
+                               "dismiss, or refine again."), root)
+
+    async def poll_email_proposals():
+        """Poll the CRM for email-activity proposals: post a review card for each new one, rebuild
+        the reply-routing map from already-posted threads (so replies still route after a restart),
+        and announce+close any decided on the web. One failing cycle logs and retries next tick."""
+        while True:
+            try:
+                lists = await asyncio.to_thread(crm_client.list_email_proposals)
+                for it in lists["open"]:  # rebuild routing for threads posted before (e.g. a restart)
+                    ev = it.get("event_id")
+                    if ev and ev not in email_threads:
+                        email_threads[ev] = {"id": it["id"], "investor_name": it.get("investor_name"),
+                                             "note": it.get("proposed_note") or ""}
+                for it in lists["to_post"]:
+                    try:
+                        resp = await client.room_send(
+                            review_room, "m.room.message",
+                            matrix_io.thread_content(email_proposals.render_card(it), None))
+                        ev = getattr(resp, "event_id", None)
+                        if not ev:
+                            print(f"matrix-intake: card send returned no event_id for {it['id']}", flush=True)
+                            continue
+                        await asyncio.to_thread(crm_client.mark_email_proposal_posted, it["id"], ev)
+                        email_threads[ev] = {"id": it["id"], "investor_name": it.get("investor_name"),
+                                             "note": it.get("proposed_note") or ""}
+                    except Exception as exc:
+                        print(f"matrix-intake: failed to post email proposal {it.get('id')}: {exc}", flush=True)
+                for it in lists["to_close"]:  # decided on the web → clear the thread, then close
+                    ev = it.get("event_id")
+                    if not ev:
+                        continue
+                    try:
+                        await redact_thread(review_room, ev)
+                        await asyncio.to_thread(crm_client.mark_email_proposal_closed, it["id"])
+                        email_threads.pop(ev, None)
+                    except Exception as exc:
+                        print(f"matrix-intake: failed to close email proposal {it.get('id')}: {exc}", flush=True)
+            except Exception as exc:
+                print(f"matrix-intake: email-proposal poll error: {exc}", flush=True)
+            await asyncio.sleep(EMAIL_POLL_SEC)
+
+    async def on_message(room: MatrixRoom, event: RoomMessageText):
+        if event.sender == mx["user_id"]:
+            return  # never react to our own messages (we post in-thread — this prevents loops)
+        text = (event.body or "").strip()
+        if not text:
+            return
+        root = matrix_io.thread_root_of(event)
+        # Email-proposal review room: only a threaded reply to a card we posted is actionable.
+        if review_room and room.room_id == review_room:
+            if root and root in email_threads:
+                await handle_email_reply(room.room_id, root, text)
+            return
+        # Dedicated Q&A room: every top-level message IS a question — no trigger needed. Threaded
+        # messages (the answers we post, or follow-ups) aren't acted on in v1.
+        if query_room and room.room_id == query_room:
+            if not root:
+                await handle_query(room.room_id, event.event_id, text)
+            return
+        if room.room_id != intake_room:
+            return
+        if root and store.has(root):
+            await handle_reply(room.room_id, root, text)
+        elif root:
+            return  # threaded message not tied to a live proposal — ignore
+        else:
+            # A top-level message is either an NL question (explicitly addressed with '?'/'@bot')
+            # or an intake note. The trigger is required, so plain notes still flow to intake.
+            q = query.parse_trigger(text)
+            if q is None:
+                await handle_intake(room.room_id, event.event_id, text)
+            elif not q:
+                await say(room.room_id, query.HELP, event.event_id)
+            else:
+                await handle_query(room.room_id, event.event_id, q)
+
+    async def on_image(room: MatrixRoom, event: RoomMessageImage):
+        # Business-card capture is intake-only: ignore our own uploads, images in the Q&A /
+        # email-review rooms, and an image dropped inside an existing thread (not a fresh card).
+        if event.sender == mx["user_id"]:
+            return
+        if room.room_id != intake_room:
+            return
+        if matrix_io.thread_root_of(event):
+            return
+        await handle_card(room.room_id, event)
+
+    # Prime the sync token past history, THEN register the callbacks — only react to messages
+    # arriving after startup (no backlog replay). (matrix-bridge pattern.)
+    print("matrix-intake: priming sync (skipping backlog)...", flush=True)
+    await client.sync(timeout=30000, full_state=False)
+    client.add_event_callback(on_message, RoomMessageText)
+    client.add_event_callback(on_image, RoomMessageImage)
+    who = await client.whoami()
+    print(f"matrix-intake: listening as {who.user_id} in room {intake_room}", flush=True)
+    tasks = [asyncio.create_task(client.sync_forever(timeout=30000))]
+    if review_room:
+        # "Invited" isn't "joined" — the bot must join before it can post cards (room_send to a
+        # room we're only invited to fails M_FORBIDDEN). Idempotent if already a member.
+        try:
+            await client.join(review_room)
+        except Exception as exc:
+            print(f"matrix-intake: could not join review room {review_room}: {exc}", flush=True)
+        tasks.append(asyncio.create_task(poll_email_proposals()))
+        print(f"matrix-intake: reviewing email proposals in room {review_room} (every {EMAIL_POLL_SEC}s)", flush=True)
+    if query_room:
+        # Read-only Q&A room — just join and listen (no poll task; questions are interactive).
+        # "Invited" isn't "joined": the bot must join before it can post answers (idempotent).
+        try:
+            await client.join(query_room)
+        except Exception as exc:
+            print(f"matrix-intake: could not join Q&A room {query_room}: {exc}", flush=True)
+        print(f"matrix-intake: answering questions in room {query_room}", flush=True)
+    try:
+        await asyncio.gather(*tasks)
+    finally:
+        await client.close()
+
+
+if __name__ == "__main__":
+    try:
+        asyncio.run(main())
+    except KeyboardInterrupt:
+        pass

backend/matrix_intake/crm_client.py

@@ -0,0 +1,209 @@
+"""CRM API client for the intake bot's write-back leg.
+
+The bot authenticates as a dedicated service user (Bearer JWT via /api/auth/login — the CRM
+has no service-key path) and reuses the CRM's OWN canonical write endpoint
+(/api/fundraising/log-communication) for both new-investor and existing-note cases, rather
+than mutating the grid itself. That endpoint creates the grid row (create_investor_if_missing),
+upserts the contact, logs the communication, appends a visible note, and re-syncs the
+relational tables + audit — exactly as a UI grid edit would. We only tag provenance
+(source="matrix_intake"). The payload builder is a pure function so it's unit-tested offline.
+"""
+import json
+import ssl
+import urllib.error
+import urllib.request
+from urllib.parse import urlencode
+
+import settings
+
+_token = None
+
+
+def _http(method, path, body=None, token=None):
+    s = settings.crm_settings()
+    url = s["base"] + path
+    data = json.dumps(body).encode("utf-8") if body is not None else None
+    headers = {"Content-Type": "application/json"}
+    if token:
+        headers["Authorization"] = f"Bearer {token}"
+    req = urllib.request.Request(url, data=data, method=method, headers=headers)
+    ctx = None
+    if url.lower().startswith("https") and not s["verify_tls"]:
+        ctx = ssl.create_default_context()
+        ctx.check_hostname = False
+        ctx.verify_mode = ssl.CERT_NONE
+    try:
+        with urllib.request.urlopen(req, timeout=30, context=ctx) as resp:
+            raw = resp.read()
+            return resp.status, (json.loads(raw) if raw else {})
+    except urllib.error.HTTPError as exc:
+        raw = exc.read()
+        try:
+            payload = json.loads(raw) if raw else {}
+        except Exception:
+            payload = {"raw": raw.decode("utf-8", "replace")}
+        return exc.code, payload
+
+
+def _login():
+    global _token
+    s = settings.crm_settings()
+    if not s["username"] or not s["password"]:
+        raise RuntimeError("CRM bot creds not set (CRM_BOT_USERNAME / CRM_BOT_PASSWORD)")
+    status, data = _http("POST", "/api/auth/login",
+                         {"username": s["username"], "password": s["password"]})
+    if status != 200 or not data.get("token"):
+        raise RuntimeError(f"CRM login failed ({status}): {data.get('error') or data}")
+    _token = data["token"]
+    return _token
+
+
+def _authed(method, path, body=None):
+    """Call the CRM with the cached token; re-login once on a 401 (token expiry)."""
+    global _token
+    token = _token or _login()
+    status, data = _http(method, path, body, token=token)
+    if status == 401:
+        token = _login()
+        status, data = _http(method, path, body, token=token)
+    return status, data
+
+
+def match(proposal):
+    """Resolve new-vs-existing for this proposal against the CRM matcher.
+
+    Returns {'match': {...}|None, 'candidates': [...]}:
+      - `match` is a confident EXACT existing investor — {'id', 'name'} — that the bot
+        auto-attaches a note to (no human disambiguation needed).
+      - `candidates` is a ranked list of fuzzy NEAR-matches — each {'id', 'name', 'score',
+        'matched_on'} — surfaced in-thread for the human to pick from (or confirm "new")
+        when there is no exact match, so a typo'd/near-duplicate name doesn't silently
+        create a second investor."""
+    q = proposal.get("investor_name") or proposal.get("contact_name") or ""
+    email = proposal.get("contact_email") or ""
+    if not q and not email:
+        return {"match": None, "candidates": []}
+    qs = urlencode({"q": q, "email": email})
+    status, data = _authed("GET", f"/api/intake/match?{qs}")
+    if status != 200:
+        raise RuntimeError(f"intake match failed ({status}): {data.get('error') or data}")
+    payload = data.get("data") or {}
+    m = payload.get("match")
+    match_out = {"id": m["id"], "name": m.get("investor_name") or q} if m else None
+    candidates = [
+        {"id": c["id"], "name": c.get("investor_name") or "?",
+         "score": c.get("score"), "matched_on": c.get("matched_on")}
+        for c in (payload.get("candidates") or []) if c.get("id")
+    ]
+    return {"match": match_out, "candidates": candidates}
+
+
+def list_email_proposals():
+    """Pull the email-activity review work-lists for the poll loop: {to_post, open, to_close}.
+    to_post = pending, un-posted (post a card); open = posted, awaiting a decision (rebuild the
+    reply-routing map after a restart); to_close = decided on the web (announce in-thread + close)."""
+    status, data = _authed("GET", "/api/intake/email-proposals")
+    if status != 200:
+        raise RuntimeError(f"email-proposals list failed ({status}): {data.get('error') or data}")
+    payload = data.get("data") or {}
+    return {k: (payload.get(k) or []) for k in ("to_post", "open", "to_close")}
+
+
+def mark_email_proposal_posted(proposal_id, event_id):
+    """Record the Matrix thread-root event id so the proposal's review state survives a restart."""
+    status, data = _authed("POST", f"/api/intake/email-proposals/{proposal_id}/matrix",
+                           {"event_id": event_id})
+    if status != 200:
+        raise RuntimeError(f"mark posted failed ({status}): {data.get('error') or data}")
+    return data.get("data") or {}
+
+
+def mark_email_proposal_closed(proposal_id):
+    """Mark the review thread resolved after announcing a web-side decision in it."""
+    status, data = _authed("POST", f"/api/intake/email-proposals/{proposal_id}/matrix",
+                           {"closed": True})
+    if status != 200:
+        raise RuntimeError(f"mark closed failed ({status}): {data.get('error') or data}")
+    return data.get("data") or {}
+
+
+def decide_email_proposal(proposal_id, decision, note=None):
+    """Relay an in-thread approve/dismiss (with the possibly-revised note) to the CRM. The server
+    appends the note to the grid on approve, tags source='matrix', and closes the thread."""
+    body = {"decision": decision}
+    if note is not None:
+        body["note"] = note
+    status, data = _authed("POST", f"/api/intake/email-proposals/{proposal_id}/decide", body)
+    if status not in (200, 201):
+        raise RuntimeError(f"email-proposal decide failed ({status}): {data.get('error') or data}")
+    return data.get("data") or {}
+
+
+def nl_query(question):
+    """Ask the read-only NL-query endpoint (POST /api/query/nl). Translation runs on the box's
+    LOCAL model — the question never leaves the box and no write is possible. Returns the
+    endpoint's structured result dict ({intent, slots, rows, summary, ...} or {error, detail});
+    the server returns that same body on a hit AND on the soft 503 (model down) / 500 (query
+    fault) status codes, so we hand it straight to the renderer. Any OTHER status — auth (403),
+    a malformed request (400), an unexpected shape — raises so the caller posts a brief error."""
+    status, data = _authed("POST", "/api/query/nl", {"question": question, "source": "matrix"})
+    if status not in (200, 500, 503):
+        raise RuntimeError(f"nl-query failed ({status}): {data.get('error') or data}")
+    return data.get("data") or {}
+
+
+def build_commit_payload(proposal):
+    """Pure: map a proposal to the /api/fundraising/log-communication request body.
+
+    Existing investor (carries _match_id) → target that exact grid row. Otherwise create the
+    investor if missing. The note becomes the communication body; the email is only sent when
+    it survived parse's source-text integrity check."""
+    contact = {
+        "name": proposal.get("contact_name") or proposal.get("investor_name") or "",
+        "email": proposal.get("contact_email") or "",
+        "title": proposal.get("contact_title") or "",
+        # city + linkedin_url + phone + mobile are honored by the server's contact upsert
+        # (_upsert_contact_from_fundraising); city also syncs to the grid contact pill, the
+        # rest land on the canonical contact record. phone = office/main line, mobile = cell.
+        "city": proposal.get("city") or "",
+        "linkedin_url": proposal.get("linkedin_url") or "",
+        "phone": proposal.get("phone") or "",
+        "mobile": proposal.get("mobile") or "",
+    }
+    note = proposal.get("note") or ""
+    # The CRM's grid note line uses subject-or-body for its one-line summary, so a non-empty
+    # subject hides the actual note text. Send a blank subject when there's a note (let the note
+    # itself show in the grid); fall back to a provenance label only when there's nothing to
+    # show. Provenance is recorded via source="matrix_intake" either way.
+    intent_label = "Note (Matrix)" if proposal.get("intent") == "meeting_note" else "Intake (Matrix)"
+    payload = {
+        "contact": contact,
+        "type": "note",
+        "body": note,
+        "subject": "" if note.strip() else intent_label,
+        "append_note": True,
+        # Provenance for the audit log: a typed note is "matrix_intake"; a scanned business card
+        # rides in on _source="matrix_card" (set by the bot's image handler). Default preserves
+        # the text path.
+        "source": proposal.get("_source") or "matrix_intake",
+    }
+    match_id = proposal.get("_match_id")
+    if match_id:
+        payload["row_id"] = match_id
+    else:
+        payload["investor_name"] = proposal.get("investor_name") or proposal.get("contact_name") or ""
+        payload["create_investor_if_missing"] = True
+    return payload
+
+
+def commit(proposal):
+    """Write the approved proposal to the CRM; return a short human summary for the thread."""
+    payload = build_commit_payload(proposal)
+    status, data = _authed("POST", "/api/fundraising/log-communication", payload)
+    if status not in (200, 201):
+        raise RuntimeError(f"log-communication failed ({status}): {data.get('error') or data}")
+    row = (data.get("data") or {}).get("row") or {}
+    name = row.get("investor_name") or payload.get("investor_name") or "investor"
+    if proposal.get("_match_id"):
+        return f"Logged a note on **{name}** (existing grid entry)."
+    return f"Created a new grid entry for **{name}**" + (" and logged a note." if payload.get("body") else ".")

backend/matrix_intake/email_proposals.py

@@ -0,0 +1,84 @@
+"""Email-activity proposal review over Matrix — the CRM→Matrix leg of the email-capture flow.
+
+The CRM (on the box) drafts a proposed grid note per newly-matched email (local model, no Claude)
+and queues it for human review. The CRM is stdlib-only and can't post to Matrix itself, so this
+bot PULLS the pending proposals (crm_client.list_email_proposals), posts a review card to the
+dedicated review room, and relays the human's in-thread reply back to the CRM. Same draft→approve
+discipline as the intake bot: nothing is appended to the grid until a human approves — here OR on
+the web Email Capture panel, the two surfaces kept in sync via the CRM's email_proposal_matrix row.
+
+This module is the PURE logic (card rendering, reply grammar, note revision) so it's unit-tested
+offline; the async poll/post/reply wiring lives in bot.py (network + Matrix, live-smoke only).
+"""
+import spark
+
+_YES = {"yes", "y", "approve", "approved", "ok", "confirm", "go", "add", "👍", "✅"}
+_NO = {"no", "n", "cancel", "discard", "reject", "skip", "stop", "👎", "❌"}
+
+_SNIPPET_MAX = 400  # email snippet shown on the card; the full body is in the web popup
+RULE = "-----------------------"  # top/bottom rule so threads don't bleed together on mobile
+
+
+def frame(text):
+    """Wrap a message in dash rules so each card/reply is visually bounded in the room."""
+    return f"{RULE}\n{text}\n{RULE}"
+
+
+def _truncate(s, n):
+    s = (s or "").strip()
+    return s if len(s) <= n else s[:n].rstrip() + "…"
+
+
+def render_card(item):
+    """The review card posted to the Matrix review room: who/when + a short email snippet + the
+    drafted note. Deliberately compact for mobile — the full scrollable body is in the web Email
+    Capture popup. Direction isn't a bare label anymore — the note itself names who emailed whom."""
+    name = item.get("investor_name") or "Unknown investor"
+    frm = item.get("from_name") or item.get("from_email") or "?"
+    lines = [f"📧 Proposed **grid note** for **{name}**"]
+    if item.get("email_subject"):
+        lines.append(f"· Subject: {item['email_subject']}")
+    if item.get("email_date"):
+        lines.append(f"· Date: {item['email_date']}")
+    lines.append(f"· From: {frm}")
+    snippet = _truncate(item.get("snippet"), _SNIPPET_MAX)
+    if snippet:
+        lines.append(f"· Email: {snippet}")
+    lines.append("")
+    lines.append(f"📝 Draft note: {item.get('proposed_note') or '(empty)'}")
+    lines.append("")
+    lines.append("Reply **yes** to add it to the grid, **no** to dismiss, or just tell me how to "
+                 "change the note (e.g. *say we discussed the Q3 raise*).")
+    return frame("\n".join(lines))
+
+
+def interpret(text):
+    """Classify an in-thread reply: 'approve' | 'reject' | 'revise' (anything else → revise the note)."""
+    t = (text or "").strip().lower()
+    if t in _YES:
+        return "approve"
+    if t in _NO:
+        return "reject"
+    return "revise"
+
+
+REVISE_SYSTEM = (
+    "You revise a single CRM note from a short instruction a venture-fund team member typed. "
+    "You are given the CURRENT note and an INSTRUCTION. Apply the instruction and reply with "
+    "ONLY a JSON object of the form {\"note\": \"<the full revised note>\"}. Keep it to one or two "
+    "factual sentences, no preamble. Output JSON only."
+)
+
+
+def revise_note(note, instruction, parse_fn=spark.parse_json):
+    """Re-draft the note via local Qwen from a free-form instruction (no Claude, no scrub — same
+    local-only basis as the intake parse). Returns the new note text, or None if the model gave
+    nothing usable / unchanged, in which case the caller re-prompts. `parse_fn` is injectable for
+    tests."""
+    prompt = "CURRENT:\n" + (note or "") + "\n\nINSTRUCTION:\n" + (instruction or "").strip()
+    raw = parse_fn(prompt, system=REVISE_SYSTEM, max_tokens=400) or {}
+    new = raw.get("note") if isinstance(raw, dict) else None
+    new = (new or "").strip()
+    if not new or new == (note or "").strip():
+        return None
+    return new

backend/matrix_intake/matrix_io.py

@@ -0,0 +1,73 @@
+"""Matrix plumbing lifted from matrix-bridge (src/bot.py): message splitting, thread-root
+detection, and a threaded-reply sender. Kept dependency-light so the rest of the bot is
+testable without a live homeserver."""
+
+MAX_MSG_CHARS = 30000  # well under Matrix's ~64KB event cap
+
+
+def split_message(text, limit=MAX_MSG_CHARS):
+    """Split text into <=limit-char chunks on newline boundaries (no truncation)."""
+    if len(text) <= limit:
+        return [text]
+    chunks, buf = [], ""
+    for line in text.splitlines(keepends=True):
+        while len(line) > limit:
+            if buf:
+                chunks.append(buf)
+                buf = ""
+            chunks.append(line[:limit])
+            line = line[limit:]
+        if len(buf) + len(line) > limit:
+            chunks.append(buf)
+            buf = ""
+        buf += line
+    if buf:
+        chunks.append(buf)
+    return chunks
+
+
+def thread_root_of(event):
+    """Return the thread root event_id if this message is a threaded reply, else None."""
+    relates = (getattr(event, "source", None) or {}).get("content", {}).get("m.relates_to") or {}
+    if relates.get("rel_type") == "m.thread":
+        return relates.get("event_id")
+    return None
+
+
+def thread_content(text, thread_root):
+    """Build an m.room.message content dict, threaded under thread_root when given."""
+    content = {"msgtype": "m.text", "body": text}
+    if thread_root:
+        content["m.relates_to"] = {
+            "rel_type": "m.thread",
+            "event_id": thread_root,
+            "is_falling_back": True,
+            "m.in_reply_to": {"event_id": thread_root},
+        }
+    return content
+
+
+def make_say(client):
+    """Return an async say(room_id, text, thread_root=None) bound to a matrix-nio client."""
+    async def say(room_id, text, thread_root=None):
+        for chunk in split_message(text):
+            await client.room_send(room_id, "m.room.message", thread_content(chunk, thread_root))
+    return say
+
+
+def reply_content(text, reply_to_event_id):
+    """Build a plain (non-threaded) reply: shows in the MAIN timeline as a reply to
+    reply_to_event_id, unlike thread_content() which lands the message inside a thread."""
+    content = {"msgtype": "m.text", "body": text}
+    if reply_to_event_id:
+        content["m.relates_to"] = {"m.in_reply_to": {"event_id": reply_to_event_id}}
+    return content
+
+
+def make_reply(client):
+    """Return an async reply(room_id, text, reply_to) that posts a plain main-timeline reply —
+    the brief 'proposed X — see thread' nudge alongside the in-thread proposal card."""
+    async def reply(room_id, text, reply_to):
+        for chunk in split_message(text):
+            await client.room_send(room_id, "m.room.message", reply_content(chunk, reply_to))
+    return reply

backend/matrix_intake/parse.py

@@ -0,0 +1,187 @@
+"""Turn a free-text intake message into a normalized proposal via local Qwen.
+
+The model only EXTRACTS structure; it never decides to write anything. New-vs-existing is
+finalized in M2 against the CRM matcher — here `intent` is the model's first read.
+
+`revise()` is the conversational-edit leg: a free-form correction the human types in the
+proposal thread (e.g. "add that we met June 14") is applied to the pending proposal via the
+same local Qwen — no Claude, no scrub. Email integrity is preserved: a changed address must
+literally appear in the instruction (or the original message); the model can never mint one.
+"""
+import json
+import re
+
+import spark
+
+SYSTEM = (
+    "You extract structured investor-intake data from a short message a venture-fund "
+    "team member typed about their fundraising outreach. The message is a note FROM a "
+    "team member ABOUT an investor or prospect they are contacting. Reply with ONLY a JSON "
+    "object, no prose, with these keys:\n"
+    '  "intent": "new_investor" if the message introduces a new investor or prospect, '
+    '"meeting_note" if it logs a note/update about an investor, else "unclear".\n'
+    '  "investor_name": the investing firm or entity name (e.g. "Acme Capital"), or null.\n'
+    '  "contact_name": the individual person mentioned, or null.\n'
+    '  "contact_email": the person\'s email if explicitly present, else null. Never invent one.\n'
+    '  "contact_title": the person\'s role/title if stated, else null.\n'
+    '  "city": the person\'s city or location if stated (e.g. "New York"), else null.\n'
+    '  "linkedin_url": the person\'s LinkedIn URL if explicitly present, else null. Never invent one.\n'
+    '  "phone": the office/main/direct phone number if present (a line labeled Phone/Tel/Office/'
+    'Direct, or a single unlabeled number); never a fax or a cell. Else null.\n'
+    '  "mobile": the cell/mobile number if present (a line labeled Cell/Mobile); never a fax. Else null.\n'
+    '  "note": any meeting note, context, or next step, else null.\n'
+    "Use null (not empty string) for anything not present."
+)
+
+# Appended when the team roster is known, so the model reads a teammate's name as the person
+# DOING the outreach, not the investor — fixes "Jonathan is chatting with Wyoming" extracting
+# the teammate instead of the prospect. Names come from settings.team_roster() (INTAKE_TEAM_ROSTER).
+ROSTER_FRAME = (
+    "These names and initials (case-insensitive) are our OWN team members — the people doing "
+    "the outreach, NOT investors or prospects. Never extract one as investor_name or "
+    "contact_name: {names}. When a team member is described talking with, meeting, or chasing "
+    'someone (e.g. "Jonathan is chatting with Wyoming"), the OTHER party (here "Wyoming") is '
+    "the investor or prospect to extract."
+)
+
+
+def build_system(roster=None, base=SYSTEM):
+    """Assemble the extraction system prompt. With a `roster` (team-member names) it appends
+    the outreach frame so a teammate's name is read as the person doing outreach, not the
+    investor. JSON-only stays the last line for recency. Pure + offline-testable."""
+    parts = [base]
+    if roster:
+        parts.append(ROSTER_FRAME.format(names=", ".join(roster)))
+    parts.append("Output JSON only.")
+    return "\n".join(parts)
+
+_EMAIL_RE = re.compile(r"[A-Za-z0-9._%+\-]+@[A-Za-z0-9.\-]+\.[A-Za-z]{2,}")
+_LINKEDIN_RE = re.compile(r"(?:https?://)?(?:[a-z]{2,3}\.)?linkedin\.com/[A-Za-z0-9_%/\-.]+", re.I)
+_VALID_INTENTS = {"new_investor", "meeting_note", "unclear"}
+_FIELDS = ("intent", "investor_name", "contact_name", "contact_email", "contact_title",
+           "city", "linkedin_url", "phone", "mobile", "note")
+
+
+def _digits(s):
+    """Bare digit run of a string (drops spaces/dashes/parens/dots), for phone-integrity checks."""
+    return re.sub(r"\D", "", str(s or ""))
+
+
+def _clean(v):
+    if v is None:
+        return None
+    s = str(v).strip()
+    if not s or s.lower() in ("null", "none", "n/a", "na", "unknown"):
+        return None
+    return s
+
+
+def normalize(raw, source_text=""):
+    """Coerce the model's dict into a stable proposal shape; salvage an email from the
+    source text if the model missed one. Returns a dict with all _FIELDS keys."""
+    raw = raw or {}
+    out = {k: _clean(raw.get(k)) for k in _FIELDS}
+
+    intent = (out["intent"] or "").lower().replace("-", "_").replace(" ", "_")
+    out["intent"] = intent if intent in _VALID_INTENTS else "unclear"
+
+    # Email integrity: only accept an address that literally appears in the source message.
+    # The model is unreliable for verbatim strings and must never mint an address — anything
+    # not present in what the human typed is dropped (a wrong email in the CRM is worse than
+    # none). This both salvages a missed address and rejects a hallucinated one.
+    m = _EMAIL_RE.search(source_text or "")
+    out["contact_email"] = m.group(0).rstrip(".,;:!?)]}>\"'") if m else None
+
+    # LinkedIn integrity: same rule as email — a profile URL identifies a specific person, so
+    # never let the model mint one; keep only a linkedin.com URL literally present in the source.
+    lm = _LINKEDIN_RE.search(source_text or "")
+    out["linkedin_url"] = lm.group(0).rstrip(".,;:!?)]}>\"'") if lm else None
+
+    # Phone integrity: keep a number (in its printed formatting) only if its digit run actually
+    # appears in the source — the model must never mint or "complete" a number. phone = the
+    # office/main line, mobile = the cell; which is which is the model's call (prompted), this
+    # only validates that the number is real. (≥7 digits avoids matching a stray short run.)
+    src_digits = _digits(source_text)
+    for f in ("phone", "mobile"):
+        d = _digits(out.get(f))
+        out[f] = out.get(f) if (len(d) >= 7 and d in src_digits) else None
+
+    # City is left as a plain extracted field (no source gate): a wrong city is low-harm and the
+    # human sees it on the card before approving, unlike a wrong email/LinkedIn.
+
+    # An intake with no firm AND no person is not actionable.
+    if not out["investor_name"] and not out["contact_name"]:
+        out["intent"] = "unclear"
+    return out
+
+
+def parse_message(text, parse_fn=spark.parse_json, roster=None):
+    """Parse one intake message. `parse_fn` is injectable for tests (defaults to Spark/Qwen);
+    `roster` is the team-member names that frame the extraction (see build_system).
+    Returns a normalized proposal dict. On a model/transport failure, raises (caller decides)."""
+    raw = parse_fn(text, system=build_system(roster), max_tokens=400)
+    proposal = normalize(raw, source_text=text)
+    # Stash the original message so a later revise() can re-check email integrity against it.
+    proposal["_source_text"] = text
+    return proposal
+
+
+REVISE_SYSTEM = (
+    "You revise a structured investor-intake proposal from a short correction a venture-fund "
+    "team member typed. You are given the CURRENT proposal as JSON and an INSTRUCTION. Apply "
+    "the instruction and reply with ONLY the full revised JSON object, these keys:\n"
+    '  "investor_name", "contact_name", "contact_email", "contact_title", "city", '
+    '"linkedin_url", "phone", "mobile", "note".\n'
+    "Change ONLY what the instruction asks; copy every other field through unchanged. Use null "
+    "for a field the instruction clears or that is genuinely absent. Never invent an email "
+    "address, a LinkedIn URL, or a phone number."
+)
+
+_REVISABLE = ("investor_name", "contact_name", "contact_title", "city", "note")
+
+
+def _apply_revision(proposal, model_out, instruction):
+    """Merge the model's revised fields onto the proposal. Pure + offline-testable.
+
+    Preserves control keys (_match_id / _stage / intent / _source_text). Enforces email
+    integrity: a revised address is taken only if it literally appears in the INSTRUCTION the
+    human typed; otherwise the existing (already integrity-checked) address is kept. The model's
+    own email field is never trusted — it must not mint an address."""
+    model_out = model_out or {}
+    out = dict(proposal)
+    for k in _REVISABLE:
+        if k in model_out:
+            out[k] = _clean(model_out.get(k))
+    m = _EMAIL_RE.search(instruction or "")
+    if m:
+        out["contact_email"] = m.group(0).rstrip(".,;:!?)]}>\"'")
+    # else: keep proposal's current contact_email (untouched above; control key copied by dict())
+    # LinkedIn follows the same rule: a revised URL is taken only if it appears in the instruction.
+    lm = _LINKEDIN_RE.search(instruction or "")
+    if lm:
+        out["linkedin_url"] = lm.group(0).rstrip(".,;:!?)]}>\"'")
+    # Phone/mobile too: a revised number is accepted only if its digits appear in the instruction
+    # (never let the model mint one); otherwise the existing value is kept.
+    instr_digits = _digits(instruction)
+    for f in ("phone", "mobile"):
+        if f in model_out:
+            cand = _clean(model_out.get(f))
+            d = _digits(cand)
+            out[f] = cand if (cand and len(d) >= 7 and d in instr_digits) else out.get(f)
+    # Don't let a revision strip the proposal down to nothing actionable.
+    if not out.get("investor_name") and not out.get("contact_name"):
+        out["investor_name"] = proposal.get("investor_name")
+        out["contact_name"] = proposal.get("contact_name")
+    return out
+
+
+def revise(proposal, instruction, parse_fn=spark.parse_json, roster=None):
+    """Apply a natural-language correction to a pending proposal via local Qwen; return the
+    revised proposal dict. `parse_fn` is injectable for tests (defaults to Spark/Qwen);
+    `roster` frames the revision the same way parse_message does (see build_system)."""
+    current = {k: proposal.get(k) for k in
+               ("investor_name", "contact_name", "contact_email", "contact_title", "note")}
+    prompt = ("CURRENT:\n" + json.dumps(current, ensure_ascii=False)
+              + "\n\nINSTRUCTION:\n" + (instruction or "").strip())
+    raw = parse_fn(prompt, system=build_system(roster, base=REVISE_SYSTEM), max_tokens=400)
+    return _apply_revision(proposal, raw, instruction)

backend/matrix_intake/proposals.py

@@ -0,0 +1,202 @@
+"""Pending-proposal store + the in-thread approval state machine.
+
+The one piece of state in the bot: a proposal awaiting a human's yes/edit/no, keyed by the
+Matrix thread root (the bot's proposal lives in a thread rooted at the user's message, and
+the user replies inside that thread). In-memory and ephemeral by design — a restart drops
+pending proposals (the user just re-sends), matching matrix-bridge's stateless-by-default
+ethos. Nothing here writes to the CRM; the bot calls the CRM client only after `approve`.
+
+A proposal carries a `_stage`: "approval" (the normal yes/edit/no card) or "disambiguate"
+(a fuzzy-match shortlist the human must resolve — pick a number / "new" / "no" — before it
+becomes an approval-stage proposal). The shortlist itself rides on `_candidates`.
+"""
+import re
+
+# field aliases accepted in `edit <field>=<value>`
+_EDIT_ALIASES = {
+    "name": "investor_name", "investor": "investor_name", "firm": "investor_name", "org": "investor_name",
+    "contact": "contact_name", "person": "contact_name",
+    "email": "contact_email",
+    "title": "contact_title", "role": "contact_title",
+    "city": "city", "location": "city",
+    "linkedin": "linkedin_url", "linkedin_url": "linkedin_url", "li": "linkedin_url",
+    "phone": "phone", "tel": "phone", "office": "phone",
+    "mobile": "mobile", "cell": "mobile",
+    "note": "note",
+}
+
+_YES = {"yes", "y", "approve", "approved", "ok", "confirm", "go", "👍", "✅"}
+_NO = {"no", "n", "cancel", "discard", "reject", "stop", "👎", "❌"}
+# "create a new investor anyway" replies to a disambiguation shortlist
+_NEW = {"new", "none", "new investor", "none of these", "create", "create new", "add new", "neither"}
+
+_CONTENT_FIELDS = ("intent", "investor_name", "contact_name", "contact_email", "contact_title",
+                   "city", "linkedin_url", "phone", "mobile", "note")
+
+
+class ProposalStore:
+    def __init__(self):
+        self._pending = {}  # thread_root -> proposal dict
+
+    def put(self, thread_root, proposal):
+        self._pending[thread_root] = proposal
+
+    def get(self, thread_root):
+        return self._pending.get(thread_root)
+
+    def pop(self, thread_root):
+        return self._pending.pop(thread_root, None)
+
+    def has(self, thread_root):
+        return thread_root in self._pending
+
+    def any_pending(self):
+        return bool(self._pending)
+
+
+def _parse_edit(text):
+    """Parse 'edit field=value' (also 'field: value'); return (canonical_field, value) or None."""
+    body = text.strip()
+    if body.lower().startswith("edit "):
+        body = body[5:].strip()
+    for sep in ("=", ":"):
+        if sep in body:
+            field, value = body.split(sep, 1)
+            field = field.strip().lower()
+            canon = _EDIT_ALIASES.get(field)
+            value = value.strip()
+            if canon and value:
+                return canon, value
+            # Not a known field on this separator — try the next one rather than bail,
+            # so e.g. "note: see deck=v2" still parses (split on ':' not the inner '=').
+            continue
+    return None
+
+
+def interpret_reply(text):
+    """Classify a threaded reply to a pending proposal.
+
+    Returns one of:
+      ("approve", None) | ("reject", None) | ("edit", (field, value)) | ("unknown", None)
+    """
+    t = (text or "").strip()
+    low = t.lower()
+    if low in _YES:
+        return ("approve", None)
+    if low in _NO:
+        return ("reject", None)
+    edit = _parse_edit(t)
+    if edit:
+        return ("edit", edit)
+    return ("unknown", None)
+
+
+def apply_edit(proposal, field, value):
+    """Return a copy of the proposal with one field changed."""
+    updated = dict(proposal)
+    updated[field] = value
+    return updated
+
+
+def same_fields(a, b):
+    """True if two proposals carry identical content (used to detect a no-op NL revision so we
+    don't tell the human 'Updated' when nothing changed)."""
+    return all((a or {}).get(k) == (b or {}).get(k) for k in _CONTENT_FIELDS)
+
+
+def interpret_disambiguation(text, n_candidates):
+    """Classify a reply to a fuzzy-match shortlist.
+
+    Returns ("pick", index) | ("new", None) | ("reject", None) | ("unknown", None). A bare
+    number selects that candidate; "new"/"none" creates a new investor; "no"/"cancel" discards."""
+    t = (text or "").strip().lower()
+    if not t:
+        return ("unknown", None)
+    if t in _NO:
+        return ("reject", None)
+    if t in _NEW:
+        return ("new", None)
+    m = re.fullmatch(r"#?\s*(\d{1,2})", t)
+    if m:
+        idx = int(m.group(1)) - 1
+        if 0 <= idx < n_candidates:
+            return ("pick", idx)
+    return ("unknown", None)
+
+
+def attach_to_candidate(proposal, candidate):
+    """Promote a disambiguation pick into an approval-stage meeting note on the chosen investor.
+    The note will target that existing grid row (via _match_id); the firm name is shown for
+    accuracy. Drops the shortlist."""
+    updated = dict(proposal)
+    updated.pop("_candidates", None)
+    updated["_stage"] = "approval"
+    updated["_match_id"] = candidate["id"]
+    updated["intent"] = "meeting_note"
+    if candidate.get("name"):
+        updated["investor_name"] = candidate["name"]
+    return updated
+
+
+def promote_to_new(proposal):
+    """Disambiguation 'new' — discard the shortlist and proceed as a new-investor proposal."""
+    updated = dict(proposal)
+    updated.pop("_candidates", None)
+    updated.pop("_match_id", None)
+    updated["_stage"] = "approval"
+    return updated
+
+
+def render_disambiguation(proposal):
+    """Render the fuzzy-match shortlist a human resolves before we create a new investor."""
+    name = proposal.get("investor_name") or proposal.get("contact_name") or "?"
+    cands = proposal.get("_candidates") or []
+    lines = [f"🔎 Before adding **{name}** as new — these existing investors look similar:"]
+    for i, c in enumerate(cands, 1):
+        lines.append(f"  **{i}.** {c.get('name') or '?'}")
+    lines.append("")
+    lines.append("Reply a **number** to log this against that investor, **new** to add it as a "
+                 "new investor, or **no** to discard.")
+    return "\n".join(lines)
+
+
+def disambiguation_nudge(proposal):
+    """Brief main-timeline pointer for a disambiguation proposal (the shortlist is in the thread)."""
+    name = proposal.get("investor_name") or proposal.get("contact_name") or "?"
+    return (f"🔎 **{name}** may match an existing investor — open the **thread** to pick one "
+            "or confirm it's new.")
+
+
+def render(proposal):
+    """Render a proposal as the in-thread message a human approves."""
+    if proposal.get("intent") == "meeting_note":
+        head = f"📝 Proposed **meeting note** for **{proposal.get('investor_name') or proposal.get('contact_name') or '?'}**"
+    else:
+        head = f"📇 Proposed **new investor**: **{proposal.get('investor_name') or proposal.get('contact_name') or '?'}**"
+    lines = [head]
+    fields = [
+        ("Investor", proposal.get("investor_name")),
+        ("Contact", proposal.get("contact_name")),
+        ("Email", proposal.get("contact_email")),
+        ("Title", proposal.get("contact_title")),
+        ("Phone", proposal.get("phone")),
+        ("Mobile", proposal.get("mobile")),
+        ("City", proposal.get("city")),
+        ("LinkedIn", proposal.get("linkedin_url")),
+        ("Note", proposal.get("note")),
+    ]
+    for label, val in fields:
+        if val:
+            lines.append(f"· {label}: {val}")
+    lines.append("")
+    lines.append("Reply **yes** to commit, **no** to discard, or just tell me what to change in "
+                 "plain words (e.g. \"change the email to a@b.com\", \"the firm is Acme Capital\").")
+    return "\n".join(lines)
+
+
+def summary_line(proposal):
+    """A brief one-liner for the main-timeline nudge; the full card lives in the thread."""
+    name = proposal.get("investor_name") or proposal.get("contact_name") or "?"
+    if proposal.get("intent") == "meeting_note":
+        return f"📝 Proposed a meeting note for **{name}** — see the thread to review & approve."
+    return f"📇 Proposed a new investor: **{name}** — see the thread to review & approve."

backend/matrix_intake/query.py

@@ -0,0 +1,189 @@
+"""NL-query Matrix surface (W2 step 5) — turn an '@bot <question>' message into a read-only
+answer from the CRM's curated NL-query endpoint, and render that answer for the chat room.
+
+This module is PURE (no network, no matrix-nio) so it's unit-testable offline; the async wiring
+(call the endpoint, post in a thread) lives in bot.py. The endpoint does the real work:
+translation runs on the box's LOCAL model (nothing leaves the box) and only the curated,
+parameterized queries can run — there is no write path here, so no approval gate applies.
+
+Trigger: a top-level message starting with '?' / '@bot' / '/ask' (see parse_trigger). We
+deliberately do NOT accept a bare leading 'ask', which would collide with intake notes like
+"Ask Jane to send the Q3 deck".
+"""
+
+# Markers a human wouldn't start an intake note with. '?' is handled separately (single char).
+QUERY_PREFIXES = ("@bot", "/ask", "/query", "/q")
+
+# Soft cap on rows rendered into a single chat answer. The endpoint already caps the SQL result
+# (server MAX_ROWS), but 500 rows is unreadable on mobile — show the first N and say how many
+# more there are (never a silent cut). Refine the question or use the web Ask box for the rest.
+MAX_DISPLAY_ROWS = 30
+
+# Column-name hints used only for nicer formatting (money / dates). Cosmetic — never affects
+# what's queried (that's fixed in intents.py).
+_MONEY_HINTS = ("amount", "invested", "total", "expected", "committed")
+# 0/1 flag columns: suppress when 0 (noise), show a label when 1.
+_FLAG_LABELS = {"graveyard": "retired", "overdue": "⚠️ overdue"}
+
+
+def parse_trigger(text):
+    """If `text` is addressed to the query bot, return the question (the remainder after the
+    trigger, possibly an empty string when the trigger is bare). Return None if it isn't a query,
+    so the caller routes it to intake instead."""
+    s = (text or "").strip()
+    if not s:
+        return None
+    if s[0] == "?":
+        return s[1:].strip()
+    low = s.lower()
+    for p in QUERY_PREFIXES:
+        if low.startswith(p):
+            rest = s[len(p):]
+            # Require a separator so '/asking …' isn't read as the '/ask' trigger.
+            if rest == "" or rest[0] in " \t\n:,":
+                return rest.lstrip(" \t\n:,").strip()
+    return None
+
+
+def _examples():
+    return ("Try things like:\n"
+            "• `?which investors haven't we contacted in 90 days?`\n"
+            "• `?top 10 investors by committed capital`\n"
+            "• `?when did we last reach out to Acme Capital?`\n"
+            "• `?how many emails has Grant sent this month?`")
+
+
+HELP = ("💬 Ask me about the fundraising database — start your message with `?` (or `@bot`).\n\n"
+        + _examples())
+
+
+def _is_money_col(name):
+    n = name.lower()
+    return any(h in n for h in _MONEY_HINTS)
+
+
+def _fmt_value(col, val):
+    """Format one scalar cell for chat: dates -> YYYY-MM-DD, money columns -> $1,234, else str."""
+    if val is None:
+        return ""
+    name = col.lower()
+    if name.endswith("_at") or name.endswith("date"):
+        return str(val)[:10]
+    if isinstance(val, (int, float)) and _is_money_col(col):
+        return f"${val:,.0f}"
+    return str(val)
+
+
+def _render_contacts(contacts):
+    """investor_lookup's nested contact dicts -> 'Name <email> (title · city, state)' lines."""
+    out = []
+    for c in contacts:
+        bits = c.get("full_name") or "?"
+        if c.get("email"):
+            bits += f" <{c['email']}>"
+        loc = ", ".join(x for x in (c.get("city"), c.get("state"), c.get("country")) if x)
+        extra = " · ".join(x for x in (c.get("title"), loc) if x)
+        if extra:
+            bits += f" ({extra})"
+        out.append(bits)
+    return out
+
+
+def _render_commitments(commitments):
+    """investor_lookup's nested per-fund commitments -> 'Fund: $amount' lines."""
+    out = []
+    for c in commitments:
+        fund = c.get("fund_name") or "?"
+        amt = c.get("amount")
+        out.append(f"{fund}: ${amt:,.0f}" if isinstance(amt, (int, float)) else f"{fund}: {amt}")
+    return out
+
+
+def _render_row(i, row, columns):
+    cols = columns or list(row.keys())
+    lead = None
+    scalars = []
+    sublines = []
+    for col in cols:
+        val = row.get(col)
+        if isinstance(val, list):
+            if not val:
+                continue
+            if col == "contacts":
+                sublines += [f"    – {x}" for x in _render_contacts(val)]
+            elif col == "commitments":
+                sublines += [f"    – {x}" for x in _render_commitments(val)]
+            else:  # generic list-of-dicts fallback (no intent uses this yet)
+                sublines += [f"    – {', '.join(f'{k}={v}' for k, v in d.items())}"
+                             for d in val if isinstance(d, dict)]
+            continue
+        if col in _FLAG_LABELS:
+            if val:
+                scalars.append(_FLAG_LABELS[col])
+            continue
+        s = _fmt_value(col, val)
+        if s == "":
+            continue
+        if lead is None:  # first non-empty column is the bold identifier for the row
+            lead = s
+        else:
+            scalars.append(f"{col}: {s}")
+    head = f"{i}. **{lead}**" if lead else f"{i}."
+    if scalars:
+        head += " — " + " · ".join(scalars)
+    return "\n".join([head] + sublines)
+
+
+def _render_interpretation(intent, slots):
+    if not intent:
+        return ""
+    if slots:
+        return f"read as: {intent} ({', '.join(f'{k}={v}' for k, v in slots.items())})"
+    return f"read as: {intent}"
+
+
+def _render_error(err, result):
+    detail = (result.get("detail") or "").strip()
+    if err == "no_match":
+        return "🤷 I couldn't map that to one of my saved queries.\n\n" + _examples()
+    if err == "model_unavailable":
+        return "⚠️ The local query model is unreachable right now — try again in a moment."
+    if err == "query_failed":
+        return f"⚠️ That query failed to run{(': ' + detail) if detail else ''}."
+    # unknown_intent / bad_slot / anything unexpected
+    return (f"⚠️ I couldn't run that ({err}){(': ' + detail) if detail else ''}.\n\n" + _examples())
+
+
+def render_answer(result):
+    """Render the NL-query endpoint's structured result into a Matrix markdown answer.
+
+    `result` is the endpoint body: a hit {intent, slots, columns, rows, summary, truncated} or
+    an error {error, detail}. Results never go back to any model — this is a deterministic format."""
+    result = result or {}
+    err = result.get("error")
+    if err:
+        return _render_error(err, result)
+
+    summary = (result.get("summary") or "").strip()
+    rows = result.get("rows") or []
+    columns = result.get("columns") or []
+    header = f"📊 {summary}" if summary else "📊 Done."
+    interp = _render_interpretation(result.get("intent"), result.get("slots") or {})
+    if interp:
+        header += f"\n_{interp}_"
+    if not rows:
+        return header + "\n\n(no matching records)"
+
+    shown = rows[:MAX_DISPLAY_ROWS]
+    blocks = [_render_row(i + 1, r, columns) for i, r in enumerate(shown)]
+    out = header + "\n\n" + "\n".join(blocks)
+
+    notes = []
+    extra = len(rows) - len(shown)
+    if extra > 0:
+        notes.append(f"+{extra} more not shown")
+    if result.get("truncated"):
+        notes.append("results hit the server cap")
+    if notes:
+        out += "\n\n_" + "; ".join(notes) + " — refine your question or use the web Ask box._"
+    return out

backend/matrix_intake/redact_intake.py

@@ -0,0 +1,86 @@
+#!/usr/bin/env python3
+"""One-time maintenance: clear the intake room's backlog of resolved/stale messages.
+
+Going forward the bot redacts each intake thread when it's approved/rejected (bot card + ack +
+nudge + the user's own note/photo). This clears the messages that piled up BEFORE that shipped.
+
+The intake room is single-purpose and the bot keeps **no durable pending state** (its proposal
+store is in-memory and is lost on every restart), so nothing in the room is "still live" after a
+restart — every message in it is safe to redact. This walks the room history and redacts every
+m.room.message event (text + business-card images), bot's and humans' alike.
+
+Redacting another user's message (the humans' notes/photos) needs the bot to hold a **redact /
+moderator power level** in the intake room — without it those just no-op and linger (the bot's own
+messages still clear). Make the bot a moderator of the intake room in Element first.
+
+Safe by default: prints what it WOULD redact and does nothing. Pass --apply to actually redact.
+Run on the Spark via the bot's own creds/image:
+    docker compose run --rm matrix-intake python -u backend/matrix_intake/redact_intake.py
+    docker compose run --rm matrix-intake python -u backend/matrix_intake/redact_intake.py --apply
+"""
+import asyncio
+import sys
+
+from nio import AsyncClient, MessageDirection
+
+import settings
+
+MAX_PAGES = 50  # 50 * 100 events is far more history than this room holds
+
+
+async def main(apply):
+    mx = settings.matrix_settings()
+    intake_room = mx.get("intake_room")
+    if not intake_room:
+        print("MATRIX_INTAKE_ROOM is not set — nothing to do.")
+        return
+    client = AsyncClient(mx["homeserver"], mx["user_id"])
+    client.restore_login(user_id=mx["user_id"], device_id=mx["device_id"], access_token=mx["token"])
+    try:
+        sync = await client.sync(timeout=10000, full_state=False)
+        token = sync.next_batch
+        targets = []  # (event_id, label)
+        seen = set()
+        for _ in range(MAX_PAGES):
+            resp = await client.room_messages(intake_room, start=token,
+                                              direction=MessageDirection.back, limit=100)
+            chunk = getattr(resp, "chunk", None)
+            if not chunk:
+                break
+            for ev in chunk:
+                src = getattr(ev, "source", None) or {}
+                if src.get("type") != "m.room.message":
+                    continue  # only chat messages + images; leave membership/state events alone
+                eid = getattr(ev, "event_id", None)
+                if not eid or eid in seen:
+                    continue
+                seen.add(eid)
+                content = src.get("content") or {}
+                if not content:
+                    continue  # already redacted (content stripped) — skip
+                msgtype = content.get("msgtype") or "?"
+                body = (content.get("body", "") or "").replace("\n", " ")
+                who = "bot " if getattr(ev, "sender", None) == mx["user_id"] else "user"
+                targets.append((eid, f"{who} [{msgtype}] {body[:60]}"))
+            token = getattr(resp, "end", None)
+            if not token:
+                break
+
+        print(f"messages to clear in the intake room: {len(targets)}")
+        fails = 0
+        for eid, label in targets:
+            print(("APPLY redact " if apply else "WOULD redact ") + eid + "  ::  " + label)
+            if apply:
+                r = await client.room_redact(intake_room, eid, reason="retroactive intake-room cleanup")
+                if not hasattr(r, "event_id"):
+                    fails += 1
+                    print(f"   ! redact failed (need mod power for others' messages?): {r}")
+        print(("done — redacted " if apply else "dry run — would redact ")
+              + f"{len(targets) - (fails if apply else 0)}/{len(targets)} event(s)"
+              + (f"; {fails} failed" if apply and fails else "") + ".")
+    finally:
+        await client.close()
+
+
+if __name__ == "__main__":
+    asyncio.run(main(apply="--apply" in sys.argv[1:]))

backend/matrix_intake/redact_resolved.py

@@ -0,0 +1,78 @@
+#!/usr/bin/env python3
+"""One-time maintenance: redact already-resolved email-proposal review cards.
+
+The bot redacts a card when it's decided going forward, but cards that were decided BEFORE that
+behavior shipped (e.g. smoke-test remnants) are already `closed` in the CRM, so the normal
+to_close sweep never touches them. This walks the review room's history, finds the bot's own
+"proposed grid note" cards, and redacts every one that is NOT still pending (i.e. not in the CRM
+`open` work-list) — leaving the room showing only what still needs handling.
+
+Safe by default: prints what it WOULD redact and does nothing. Pass --apply to actually redact.
+Run on the Spark via the bot's own creds/image:
+    docker compose run --rm matrix-intake python -u backend/matrix_intake/redact_resolved.py
+    docker compose run --rm matrix-intake python -u backend/matrix_intake/redact_resolved.py --apply
+"""
+import asyncio
+import sys
+
+from nio import AsyncClient, MessageDirection
+
+import crm_client
+import settings
+
+CARD_MARKER = "📧 Proposed"   # present in every review card (old and dash-framed)
+MAX_PAGES = 30                # 30 * 100 events is far more history than this room holds
+
+
+async def main(apply):
+    mx = settings.matrix_settings()
+    review_room = settings.email_review_room()
+    if not review_room:
+        print("MATRIX_EMAIL_REVIEW_ROOM is not set — nothing to do.")
+        return
+    client = AsyncClient(mx["homeserver"], mx["user_id"])
+    client.restore_login(user_id=mx["user_id"], device_id=mx["device_id"], access_token=mx["token"])
+    try:
+        # Cards still pending (must be KEPT) — their thread-root event id is the card event id.
+        open_ids = {it["event_id"] for it in crm_client.list_email_proposals().get("open", []) if it.get("event_id")}
+        print(f"pending cards to keep: {len(open_ids)}")
+
+        sync = await client.sync(timeout=10000, full_state=False)
+        token = sync.next_batch
+        cards = {}    # root event_id -> snippet (still-identifiable card bodies)
+        replies = {}  # reply event_id -> (thread_root, snippet)
+        for _ in range(MAX_PAGES):
+            resp = await client.room_messages(review_room, start=token,
+                                              direction=MessageDirection.back, limit=100)
+            chunk = getattr(resp, "chunk", None)
+            if not chunk:
+                break
+            for ev in chunk:
+                body = (getattr(ev, "body", "") or "").replace("\n", " ")
+                rel = ((getattr(ev, "source", None) or {}).get("content", {}) or {}).get("m.relates_to") or {}
+                if rel.get("rel_type") == "m.thread" and rel.get("event_id"):
+                    replies[ev.event_id] = (rel["event_id"], body[:50])  # a threaded reply (card already redacted)
+                elif getattr(ev, "sender", None) == mx["user_id"] and CARD_MARKER in body:
+                    cards[ev.event_id] = body[:70]                       # an un-redacted card root
+            token = getattr(resp, "end", None)
+            if not token:
+                break
+
+        # Redact card roots that aren't still pending, AND any reply whose thread isn't still pending.
+        targets = [(eid, "card  :: " + snip) for eid, snip in cards.items() if eid not in open_ids]
+        targets += [(eid, "reply :: " + snip) for eid, (root, snip) in replies.items() if root not in open_ids]
+        print(f"resolved cards: {sum(1 for e,_ in cards.items() if e not in open_ids)}; "
+              f"thread replies to clear: {sum(1 for _,(r,_) in replies.items() if r not in open_ids)}")
+        for eid, label in targets:
+            print(("APPLY redact " if apply else "WOULD redact ") + eid + "  ::  " + label)
+            if apply:
+                r = await client.room_redact(review_room, eid, reason="retroactive cleanup of resolved review threads")
+                if not hasattr(r, "event_id"):
+                    print(f"   ! redact failed: {r}")
+        print(("done — redacted " if apply else "dry run — would redact ") + f"{len(targets)} event(s).")
+    finally:
+        await client.close()
+
+
+if __name__ == "__main__":
+    asyncio.run(main(apply="--apply" in sys.argv[1:]))

backend/matrix_intake/requirements.txt

@@ -0,0 +1,4 @@
+# Matrix intake bot — isolated to this component's own process. matrix-nio is the ONLY
+# third-party runtime dep and MUST NOT be added to the stdlib CRM (backend/server.py).
+# The Spark/Qwen + CRM-API calls reuse the repo's stdlib HTTP client (backend/ingest/http_util).
+matrix-nio>=0.24

backend/matrix_intake/settings.py

@@ -0,0 +1,80 @@
+"""Config for the Matrix intake bot — Matrix creds + the dedicated intake room.
+
+Spark settings (SPARK_CONTROL_URL, CHAT_MODEL, …) are NOT read here; they come from the
+reused ingest client (see spark.py), which loads the same repo .env. This module only owns
+the Matrix connection and the CRM API target for the write-back leg (M2).
+"""
+import os
+
+REPO_ROOT = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+
+def load_env(path=None):
+    """Populate os.environ from the repo .env (setdefault — never clobber a real env var)."""
+    path = path or os.path.join(REPO_ROOT, ".env")
+    if not os.path.exists(path):
+        return
+    with open(path, "r", encoding="utf-8") as fh:
+        for line in fh:
+            line = line.strip()
+            if not line or line.startswith("#") or "=" not in line:
+                continue
+            k, v = line.split("=", 1)
+            os.environ.setdefault(k.strip(), v.strip())
+
+
+load_env()
+
+
+def _require(name):
+    val = os.environ.get(name, "").strip()
+    if not val:
+        raise RuntimeError(f"matrix_intake: required env var {name} is not set (see .env.example)")
+    return val
+
+
+# Matrix connection (resolved lazily so importing this module for tests never requires creds).
+def matrix_settings():
+    return {
+        "homeserver": _require("MATRIX_HOMESERVER"),
+        "user_id": _require("MATRIX_USER"),
+        "token": _require("MATRIX_ACCESS_TOKEN"),
+        "device_id": os.environ.get("MATRIX_DEVICE_ID", "ten31-intake-bot"),
+        "intake_room": _require("MATRIX_INTAKE_ROOM"),
+    }
+
+
+# CRM API target for the write-back leg (M2). The CRM has no service-key auth path — auth is
+# Bearer-JWT via /api/auth/login — so the bot logs in as a DEDICATED service user (a normal
+# CRM user, created by an admin) and reuses the existing auth. Creds live in .env, never code.
+def crm_settings():
+    return {
+        "base": os.environ.get("CRM_API_BASE", "http://127.0.0.1:8080").rstrip("/"),
+        "username": os.environ.get("CRM_BOT_USERNAME", "").strip(),
+        "password": os.environ.get("CRM_BOT_PASSWORD", ""),
+        "verify_tls": os.environ.get("CRM_API_VERIFY_TLS", "true").lower() in ("1", "true", "yes", "on"),
+    }
+
+
+# Team-member names (comma-separated in INTAKE_TEAM_ROSTER), fed to the parser so a teammate's
+# name reads as the person DOING outreach, not the investor (see parse.build_system). Optional —
+# unset/empty just means no roster framing, i.e. the prior behavior.
+def team_roster():
+    return [n.strip() for n in os.environ.get("INTAKE_TEAM_ROSTER", "").split(",") if n.strip()]
+
+
+# Dedicated room for reviewing CRM-drafted email-activity proposals (the CRM→Matrix push leg).
+# Separate from the intake room so high-volume email proposals don't drown the conversational
+# intake flow. Unset/empty disables the whole email-review poll loop (the bot just does intake).
+def email_review_room():
+    return os.environ.get("MATRIX_EMAIL_REVIEW_ROOM", "").strip()
+
+
+# Dedicated Q&A room for read-only natural-language queries (W2). In this room EVERY top-level
+# message is treated as a question — no '?'/'@bot' trigger needed (the trigger only exists to
+# disambiguate question-vs-note when Q&A shares the intake room; here that's unnecessary). The
+# '?'/'@bot' trigger still works in the intake room too, as a cross-room convenience. Unset/empty
+# just means no dedicated room (questions then go through the intake-room trigger). The bot must be
+# a member of this room. Read-only — no approval gate, no redaction, no special power level needed.
+def query_room():
+    return os.environ.get("MATRIX_QUERY_ROOM", "").strip()

backend/matrix_intake/spark.py

@@ -0,0 +1,56 @@
+"""Thin reuse of the in-repo local-Qwen client (backend/ingest/llm.py) via Spark Control.
+
+We import the ingest client rather than re-implementing the HTTP call so the intake bot
+speaks the exact same Spark contract (model, /v1/chat/completions, TLS verify, .env load).
+The intake message is real LP substance, but it goes ONLY to the local Qwen on Ten31 infra
+— never Claude — so no scrub boundary applies (same basis as the daily digest). Never call a
+Spark directly; everything goes through SPARK_CONTROL_URL.
+"""
+import os
+import sys
+
+_INGEST = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "ingest")
+if _INGEST not in sys.path:
+    sys.path.insert(0, _INGEST)
+
+import llm  # noqa: E402  (backend/ingest/llm.py — chat / chat_json over Spark Control)
+
+
+def parse_json(prompt, system=None, max_tokens=400):
+    """Send to local Qwen (temp 0, thinking off) and parse the first JSON object, or None."""
+    return llm.chat_json(prompt, system=system, max_tokens=max_tokens)
+
+
+# The vision model only TRANSCRIBES the card; the existing text-parse flow then extracts the
+# structured proposal from that transcription. Keeping the two steps separate (vs. asking the
+# vision model for JSON directly) is deliberate: the transcription becomes the source text the
+# email-integrity check runs against, so the "only keep an address that literally appears in the
+# source, never let the model mint one" rule (parse.normalize) protects card intake too.
+CARD_SYSTEM = (
+    "You are transcribing a photo of a business card. Copy the text EXACTLY as printed — never "
+    "paraphrase, translate, complete, normalize, or correct anything.\n"
+    "Read each of these character-by-character and reproduce every glyph precisely. Do NOT 'fix' "
+    "them toward a more common spelling or a well-known company's domain, and never add or drop a "
+    "character:\n"
+    "  - Email: check the local part, the @, and the domain separately (transcribe 'mara.com' as "
+    "'mara.com', never 'marac.com').\n"
+    "  - Phone, cell/mobile, and fax numbers — keep each on its own labeled line so they aren't "
+    "confused (put an office/main/direct number on Phone:, a cell/mobile number on Mobile:, and a "
+    "fax on Fax:).\n"
+    "  - Website / LinkedIn URL.\n"
+    "Then list, each on its own labeled line and ONLY if present on the card:\n"
+    "  Name:  Title:  Company:  Email:  Phone:  Mobile:  Fax:  LinkedIn:  City:\n"
+    "If a character is genuinely ambiguous, give your single best reading — never invent extra "
+    "characters to fill a gap. If the image is not a readable business card, reply with the single "
+    "word NONE. Output only the labeled lines, nothing else."
+)
+
+
+def transcribe_card(image_b64, mime="image/jpeg", chat_fn=None):
+    """Vision-transcribe a business card to faithful text via the local VL model (same model and
+    Spark Control endpoint as the text parse). Returns the transcription string, or '' if the model
+    saw no readable card. `chat_fn` is injectable for offline tests (defaults to Spark/VL)."""
+    chat_fn = chat_fn or llm.chat_vision
+    out = (chat_fn("Transcribe this business card.", image_b64, mime=mime,
+                   system=CARD_SYSTEM, max_tokens=600) or "").strip()
+    return "" if out.upper() == "NONE" else out

backend/matrix_intake/test_crm_client.py

@@ -0,0 +1,180 @@
+"""Tests for the CRM client's payload builder (pure logic, no network)."""
+import os
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+import crm_client  # noqa: E402
+
+
+def test_new_investor_payload():
+    p = {"intent": "new_investor", "investor_name": "Acme Capital",
+         "contact_name": "Jane Doe", "contact_email": "jane@acme.com",
+         "contact_title": "GP", "note": "met at conf"}
+    out = crm_client.build_commit_payload(p)
+    assert out["investor_name"] == "Acme Capital"
+    assert out["create_investor_if_missing"] is True
+    assert "row_id" not in out
+    assert out["contact"] == {"name": "Jane Doe", "email": "jane@acme.com", "title": "GP",
+                              "city": "", "linkedin_url": "", "phone": "", "mobile": ""}
+    assert out["body"] == "met at conf"
+    assert out["source"] == "matrix_intake"
+
+
+def test_contact_carries_card_fields_when_present():
+    p = {"intent": "new_investor", "investor_name": "Acme Capital", "contact_name": "Jane Doe",
+         "contact_email": "jane@acme.com", "city": "New York",
+         "linkedin_url": "linkedin.com/in/janedoe", "phone": "212-555-0100",
+         "mobile": "917-555-0199", "note": "met at conf"}
+    out = crm_client.build_commit_payload(p)
+    assert out["contact"]["city"] == "New York"
+    assert out["contact"]["linkedin_url"] == "linkedin.com/in/janedoe"
+    assert out["contact"]["phone"] == "212-555-0100"   # office/main line
+    assert out["contact"]["mobile"] == "917-555-0199"  # cell
+
+
+def test_existing_investor_uses_row_id_not_create():
+    p = {"intent": "meeting_note", "investor_name": "Acme Capital",
+         "contact_name": "Jane Doe", "contact_email": None, "note": "wants Q3 deck",
+         "_match_id": "rowAcme"}
+    out = crm_client.build_commit_payload(p)
+    assert out["row_id"] == "rowAcme"
+    assert "create_investor_if_missing" not in out
+    assert "investor_name" not in out  # targeted by row id, never re-matched by name
+    assert out["body"] == "wants Q3 deck"
+
+
+def test_contact_falls_back_to_investor_name_when_no_person():
+    p = {"intent": "new_investor", "investor_name": "Delta Fund",
+         "contact_name": None, "contact_email": None, "note": None}
+    out = crm_client.build_commit_payload(p)
+    assert out["contact"]["name"] == "Delta Fund"
+    assert out["body"] == ""
+
+
+def test_no_email_sends_empty_string_not_none():
+    p = {"intent": "new_investor", "investor_name": "Gamma", "contact_name": "Bob",
+         "contact_email": None, "note": "x"}
+    out = crm_client.build_commit_payload(p)
+    assert out["contact"]["email"] == ""
+
+
+def test_subject_blank_when_note_present_else_provenance_label():
+    # The CRM's grid note line uses subject-or-body, so a blank subject lets the note text show.
+    with_note = crm_client.build_commit_payload(
+        {"intent": "meeting_note", "investor_name": "Acme", "note": "sent the deck", "_match_id": "r1"})
+    assert with_note["subject"] == ""
+    assert with_note["body"] == "sent the deck"
+    # no note text → fall back to a provenance label so the grid line isn't empty
+    no_note = crm_client.build_commit_payload(
+        {"intent": "new_investor", "investor_name": "Beta", "contact_name": "X", "note": None})
+    assert no_note["subject"] == "Intake (Matrix)"
+
+
+def test_source_defaults_to_intake_and_card_overrides():
+    # Provenance: a typed note tags source="matrix_intake"; a scanned card rides in on
+    # _source="matrix_card" (set by the bot's image handler) so the audit log distinguishes them.
+    typed = crm_client.build_commit_payload(
+        {"intent": "new_investor", "investor_name": "Acme", "note": "x"})
+    assert typed["source"] == "matrix_intake"
+    card = crm_client.build_commit_payload(
+        {"intent": "new_investor", "investor_name": "Acme", "note": "x", "_source": "matrix_card"})
+    assert card["source"] == "matrix_card"
+
+
+def _with_stub_authed(reply, capture=None):
+    """Swap crm_client._authed for a canned (status, data); return a restorer."""
+    orig = crm_client._authed
+
+    def fake(method, path, body=None):
+        if capture is not None:
+            capture["path"] = path
+        return reply
+
+    crm_client._authed = fake
+    return orig
+
+
+def test_match_parses_exact_match():
+    cap = {}
+    orig = _with_stub_authed((200, {"data": {
+        "match": {"id": "rowAcme", "investor_name": "Acme Capital", "matched_on": "name"},
+        "candidates": [],
+    }}), cap)
+    try:
+        res = crm_client.match({"investor_name": "Acme Capital", "contact_email": ""})
+    finally:
+        crm_client._authed = orig
+    assert res["match"] == {"id": "rowAcme", "name": "Acme Capital"}
+    assert res["candidates"] == []
+    assert "q=Acme" in cap["path"]  # the query was forwarded
+
+
+def test_match_returns_ranked_candidates_when_no_exact():
+    orig = _with_stub_authed((200, {"data": {"match": None, "candidates": [
+        {"id": "rowCharlie", "investor_name": "Charlie Brown", "score": 0.92, "matched_on": "name"},
+        {"id": "rowBeta", "investor_name": "Beta Capital LLC", "score": 0.86, "matched_on": "name"},
+    ]}}))
+    try:
+        res = crm_client.match({"investor_name": "Charles Brown"})
+    finally:
+        crm_client._authed = orig
+    assert res["match"] is None
+    assert [c["id"] for c in res["candidates"]] == ["rowCharlie", "rowBeta"]
+    assert res["candidates"][0]["name"] == "Charlie Brown"
+    assert res["candidates"][0]["matched_on"] == "name"
+
+
+def test_match_no_query_skips_network():
+    def boom(*a, **k):
+        raise AssertionError("should not hit the network when there's nothing to match on")
+    orig = crm_client._authed
+    crm_client._authed = boom
+    try:
+        res = crm_client.match({"investor_name": None, "contact_name": None, "contact_email": None})
+    finally:
+        crm_client._authed = orig
+    assert res == {"match": None, "candidates": []}
+
+
+def test_nl_query_returns_endpoint_data():
+    cap = {}
+    orig = _with_stub_authed(
+        (200, {"data": {"intent": "top_investors_committed", "rows": [], "summary": "ok"}}), cap)
+    try:
+        res = crm_client.nl_query("top investors")
+    finally:
+        crm_client._authed = orig
+    assert res["intent"] == "top_investors_committed"
+    assert cap["path"] == "/api/query/nl"
+
+
+def test_nl_query_passes_through_soft_503():
+    # Model-down still carries a structured body (the endpoint 503s with the error in `data`) —
+    # return it for the renderer to surface, don't raise.
+    orig = _with_stub_authed((503, {"data": {"error": "model_unavailable"}}))
+    try:
+        res = crm_client.nl_query("anything")
+    finally:
+        crm_client._authed = orig
+    assert res["error"] == "model_unavailable"
+
+
+def test_nl_query_raises_on_auth_failure():
+    orig = _with_stub_authed((403, {"error": "Bot or admin required"}))
+    raised = False
+    try:
+        crm_client.nl_query("x")
+    except RuntimeError:
+        raised = True
+    finally:
+        crm_client._authed = orig
+    assert raised
+
+
+if __name__ == "__main__":
+    fns = [v for k, v in sorted(globals().items()) if k.startswith("test_") and callable(v)]
+    for fn in fns:
+        fn()
+        print(f"ok  {fn.__name__}")
+    print(f"\n{len(fns)} passed")

backend/matrix_intake/test_email_proposals.py

@@ -0,0 +1,78 @@
+"""Offline tests for the email-proposal review logic (card render, framing, reply grammar, note
+revision). The network/Matrix wiring lives in bot.py (live-smoke only); this covers pure functions."""
+import os
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+import email_proposals  # noqa: E402
+
+ITEM = {
+    "id": "p1", "investor_name": "Acme Capital", "direction": "received",
+    "from_name": "Jane Doe", "from_email": "jane@acme.com",
+    "email_subject": "Re: Fund III", "email_date": "2026-06-02",
+    "snippet": "thanks for the deck — one question on terms",
+    "proposed_note": "✉ Jane Doe emailed the team: asked about terms",
+}
+
+
+def test_interpret_yes_no_else():
+    assert email_proposals.interpret("yes") == "approve"
+    assert email_proposals.interpret(" Y ") == "approve"
+    assert email_proposals.interpret("✅") == "approve"
+    assert email_proposals.interpret("no") == "reject"
+    assert email_proposals.interpret("skip") == "reject"
+    # anything that isn't a clear yes/no is treated as a revision instruction
+    assert email_proposals.interpret("say we discussed the Q3 raise") == "revise"
+
+
+def test_frame_wraps_with_rules():
+    out = email_proposals.frame("hello")
+    lines = out.split("\n")
+    assert lines[0] == email_proposals.RULE and lines[-1] == email_proposals.RULE
+    assert "hello" in out
+
+
+def test_render_card_has_context_note_and_actions():
+    card = email_proposals.render_card(ITEM)
+    assert "Acme Capital" in card
+    assert "Jane Doe" in card
+    assert "Re: Fund III" in card and "2026-06-02" in card
+    assert "thanks for the deck" in card
+    assert "Jane Doe emailed the team: asked about terms" in card  # the clear, named note
+    assert "yes" in card.lower() and "no" in card.lower()
+
+
+def test_render_card_is_framed_and_dropless_direction():
+    card = email_proposals.render_card(ITEM)
+    assert card.startswith(email_proposals.RULE) and card.rstrip().endswith(email_proposals.RULE)
+    # the bare Sent/Received label is gone — the note itself names who emailed whom
+    assert "(Received)" not in card and "(Sent)" not in card
+
+
+def test_render_card_truncates_long_snippet():
+    card = email_proposals.render_card(dict(ITEM, snippet="x" * 1000))
+    assert "…" in card and len(card) < 1000
+
+
+def test_revise_note_applies_model_output():
+    out = email_proposals.revise_note(
+        "old note", "make it about the Q3 raise",
+        parse_fn=lambda prompt, system=None, max_tokens=400: {"note": "Discussed the Q3 raise."})
+    assert out == "Discussed the Q3 raise."
+
+
+def test_revise_note_noop_or_empty_returns_none():
+    # model echoes the same note unchanged -> None so the caller re-prompts (not "Updated")
+    assert email_proposals.revise_note("same", "x", parse_fn=lambda *a, **k: {"note": "same"}) is None
+    # model returns nothing usable -> None
+    assert email_proposals.revise_note("n", "y", parse_fn=lambda *a, **k: {}) is None
+    assert email_proposals.revise_note("n", "y", parse_fn=lambda *a, **k: None) is None
+
+
+if __name__ == "__main__":
+    fns = [v for k, v in sorted(globals().items()) if k.startswith("test_") and callable(v)]
+    for fn in fns:
+        fn()
+        print(f"ok  {fn.__name__}")
+    print(f"\n{len(fns)} passed")

backend/matrix_intake/test_matrix_io.py

@@ -0,0 +1,37 @@
+"""Tests for matrix_io content builders — pure dict shaping, no network."""
+import os
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+import matrix_io  # noqa: E402
+
+
+def test_reply_content_is_plain_main_timeline_reply():
+    c = matrix_io.reply_content("hi", "$evt1")
+    rel = c["m.relates_to"]
+    assert rel["m.in_reply_to"]["event_id"] == "$evt1"
+    # a plain reply must NOT carry a thread relation, or it'd land in the thread
+    # instead of the main timeline (the whole point of the nudge).
+    assert "rel_type" not in rel
+
+
+def test_reply_content_without_target_has_no_relation():
+    c = matrix_io.reply_content("hi", None)
+    assert "m.relates_to" not in c
+    assert c["body"] == "hi"
+
+
+def test_thread_content_stays_threaded():
+    c = matrix_io.thread_content("hi", "$root1")
+    rel = c["m.relates_to"]
+    assert rel["rel_type"] == "m.thread"
+    assert rel["event_id"] == "$root1"
+
+
+if __name__ == "__main__":
+    fns = [v for k, v in sorted(globals().items()) if k.startswith("test_") and callable(v)]
+    for fn in fns:
+        fn()
+        print(f"ok  {fn.__name__}")
+    print(f"\n{len(fns)} passed")

backend/matrix_intake/test_parse.py

@@ -0,0 +1,287 @@
+"""Tests for the intake parse/normalize layer — Spark/Qwen stubbed (no network)."""
+import os
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+import parse  # noqa: E402
+
+
+def _stub(reply):
+    """Return a parse_fn that ignores input and yields `reply` (simulating Qwen's JSON)."""
+    return lambda text, system=None, max_tokens=400: reply
+
+
+def test_new_investor_basic():
+    p = parse.parse_message(
+        "New investor Acme Capital, contact Jane Doe jane@acme.com, met at the Austin conf",
+        parse_fn=_stub({"intent": "new_investor", "investor_name": "Acme Capital",
+                        "contact_name": "Jane Doe", "contact_email": "jane@acme.com",
+                        "contact_title": None, "note": "met at the Austin conf"}),
+    )
+    assert p["intent"] == "new_investor"
+    assert p["investor_name"] == "Acme Capital"
+    assert p["contact_email"] == "jane@acme.com"
+
+
+def test_email_salvaged_from_source_when_model_misses():
+    p = parse.parse_message(
+        "add bob@example.org from Beta LP",
+        parse_fn=_stub({"intent": "new_investor", "investor_name": "Beta LP",
+                        "contact_name": "Bob", "contact_email": None}),
+    )
+    assert p["contact_email"] == "bob@example.org"
+
+
+def test_fabricated_email_dropped_when_not_in_source():
+    p = parse.parse_message(
+        "new prospect Gamma Partners, talked to their GP",
+        parse_fn=_stub({"intent": "new_investor", "investor_name": "Gamma Partners",
+                        "contact_name": "their GP", "contact_email": "made-up@nowhere.test"}),
+    )
+    # the model invented an address that isn't in the source → must be dropped
+    assert p["contact_email"] is None
+
+
+def test_email_extracted_without_surrounding_punctuation():
+    # "Name <addr>" is the most common contact format; parens / trailing period also occur.
+    # The salvage-from-source path must extract the bare address, never the brackets.
+    cases = [
+        ("New investor: Larch Capital — Dana Reed <dana@larchcap.com>, met at conf", "dana@larchcap.com"),
+        ("ping (sam@beta.io) re the deck", "sam@beta.io"),
+        ("reach kim@acme.co.", "kim@acme.co"),
+    ]
+    for src, expected in cases:
+        p = parse.parse_message(
+            src,
+            parse_fn=_stub({"intent": "new_investor", "investor_name": "X",
+                            "contact_name": "Y", "contact_email": None}),
+        )
+        assert p["contact_email"] == expected, (src, p["contact_email"])
+
+
+def test_meeting_note_intent_preserved():
+    p = parse.parse_message(
+        "Note for Acme Capital: wants the Q3 deck",
+        parse_fn=_stub({"intent": "meeting_note", "investor_name": "Acme Capital",
+                        "note": "wants the Q3 deck"}),
+    )
+    assert p["intent"] == "meeting_note"
+    assert p["note"] == "wants the Q3 deck"
+
+
+def test_unclear_when_no_entity():
+    p = parse.parse_message(
+        "hey what's up",
+        parse_fn=_stub({"intent": "new_investor", "investor_name": None, "contact_name": None}),
+    )
+    assert p["intent"] == "unclear"
+
+
+def test_null_strings_normalized():
+    p = parse.parse_message(
+        "Delta Fund",
+        parse_fn=_stub({"intent": "new_investor", "investor_name": "Delta Fund",
+                        "contact_name": "null", "contact_email": "N/A", "note": ""}),
+    )
+    assert p["contact_name"] is None
+    assert p["contact_email"] is None
+    assert p["note"] is None
+
+
+def test_bad_intent_falls_back_to_unclear():
+    p = parse.parse_message(
+        "Epsilon Capital",
+        parse_fn=_stub({"intent": "garbage", "investor_name": "Epsilon Capital"}),
+    )
+    assert p["intent"] == "unclear"
+
+
+def test_none_model_reply_is_unclear():
+    p = parse.parse_message("???", parse_fn=_stub(None))
+    assert p["intent"] == "unclear"
+
+
+def test_parse_message_stashes_source_text():
+    p = parse.parse_message("Acme Capital, Jane jane@acme.com",
+                            parse_fn=_stub({"intent": "new_investor", "investor_name": "Acme Capital",
+                                            "contact_name": "Jane", "contact_email": "jane@acme.com"}))
+    assert p["_source_text"] == "Acme Capital, Jane jane@acme.com"
+
+
+def test_revise_applies_note_change_and_preserves_control_keys():
+    proposal = parse.parse_message(
+        "New investor Acme Capital, Jane Doe jane@acme.com",
+        parse_fn=_stub({"intent": "new_investor", "investor_name": "Acme Capital",
+                        "contact_name": "Jane Doe", "contact_email": "jane@acme.com",
+                        "contact_title": None, "note": None}))
+    revised = parse.revise(
+        proposal, "add that we met on June 14",
+        parse_fn=_stub({"investor_name": "Acme Capital", "contact_name": "Jane Doe",
+                        "contact_email": "jane@acme.com", "contact_title": None,
+                        "note": "met on June 14"}))
+    assert revised["note"] == "met on June 14"
+    assert revised["investor_name"] == "Acme Capital"
+    assert revised["intent"] == "new_investor"                 # control key preserved
+    assert revised["_source_text"] == proposal["_source_text"]  # preserved for email integrity
+
+
+def test_revise_email_taken_only_from_instruction():
+    proposal = {"intent": "new_investor", "investor_name": "Acme", "contact_name": "Jane",
+                "contact_email": "jane@acme.com", "contact_title": None, "note": None,
+                "_source_text": "Acme, Jane jane@acme.com"}
+    # instruction literally carries the new address → accepted
+    r1 = parse.revise(proposal, "her email is jane@newfirm.com",
+                      parse_fn=_stub({"contact_email": "jane@newfirm.com"}))
+    assert r1["contact_email"] == "jane@newfirm.com"
+    # model tries to change the email but the instruction has no address → keep the existing one
+    r2 = parse.revise(proposal, "set her title to GP",
+                      parse_fn=_stub({"contact_email": "totally@madeup.test", "contact_title": "GP"}))
+    assert r2["contact_email"] == "jane@acme.com"  # model's email ignored (not in instruction)
+    assert r2["contact_title"] == "GP"
+
+
+def test_revise_preserves_match_id():
+    proposal = {"intent": "meeting_note", "investor_name": "Acme", "contact_name": None,
+                "contact_email": None, "contact_title": None, "note": "old",
+                "_match_id": "rowAcme", "_stage": "approval", "_source_text": "note for Acme: old"}
+    revised = parse.revise(proposal, "change the note to: sent the deck",
+                           parse_fn=_stub({"note": "sent the deck"}))
+    assert revised["note"] == "sent the deck"
+    assert revised["_match_id"] == "rowAcme"
+    assert revised["intent"] == "meeting_note"
+
+
+def test_build_system_appends_roster_frame_only_when_roster_given():
+    base = parse.build_system()
+    assert base.strip().endswith("Output JSON only.")
+    assert "doing the outreach" not in base  # no roster → no outreach frame
+
+    framed = parse.build_system(["Grant", "Jonathan", "Marty"])
+    assert "Grant" in framed and "Jonathan" in framed and "Marty" in framed
+    assert "doing the outreach" in framed  # the outreach frame is present
+    assert framed.strip().endswith("Output JSON only.")  # JSON-only stays last for recency
+
+
+def test_parse_message_injects_roster_into_system_prompt():
+    # Capture the system prompt the model is handed, and confirm the teammate ("jonathan")
+    # is framed as outreach while the prospect ("wyoming") is what gets extracted.
+    seen = {}
+
+    def cap(text, system=None, max_tokens=400):
+        seen["system"] = system
+        return {"intent": "meeting_note", "investor_name": "Wyoming", "contact_name": None,
+                "note": "jonathan chatting with them"}
+
+    p = parse.parse_message("jonathan is chatting with wyoming", parse_fn=cap,
+                            roster=["Grant", "Jonathan", "Marty"])
+    assert "Jonathan" in seen["system"]
+    assert "doing the outreach" in seen["system"]
+    assert p["investor_name"] == "Wyoming"
+
+
+def test_revise_injects_roster_into_system_prompt():
+    proposal = {"intent": "meeting_note", "investor_name": "Wyoming", "contact_name": None,
+                "contact_email": None, "contact_title": None, "note": "x",
+                "_source_text": "jonathan is chatting with wyoming"}
+    seen = {}
+
+    def cap(prompt, system=None, max_tokens=400):
+        seen["system"] = system
+        return {"note": "sent the deck"}
+
+    parse.revise(proposal, "note: sent the deck", parse_fn=cap, roster=["Grant", "Jonathan"])
+    assert "Jonathan" in seen["system"]
+    assert "doing the outreach" in seen["system"]
+
+
+def test_city_kept_as_plain_field_and_linkedin_salvaged_from_source():
+    # A card transcription carries labeled lines; city is kept as-is, LinkedIn is salvaged from
+    # the source text (verbatim) the same way email is.
+    src = ("New investor — from a business card:\nName: Jane Doe\nCompany: Acme Capital\n"
+           "Email: jane@acme.com\nLinkedIn: linkedin.com/in/janedoe\nCity: New York")
+    p = parse.parse_message(
+        src,
+        parse_fn=_stub({"intent": "new_investor", "investor_name": "Acme Capital",
+                        "contact_name": "Jane Doe", "contact_email": "jane@acme.com",
+                        "city": "New York", "linkedin_url": None}),  # model missed the URL
+    )
+    assert p["city"] == "New York"
+    assert p["linkedin_url"] == "linkedin.com/in/janedoe"  # salvaged from source
+
+
+def test_fabricated_linkedin_dropped_when_not_in_source():
+    p = parse.parse_message(
+        "new prospect Gamma Partners, talked to their GP",
+        parse_fn=_stub({"intent": "new_investor", "investor_name": "Gamma Partners",
+                        "contact_name": "their GP", "linkedin_url": "linkedin.com/in/madeup"}),
+    )
+    assert p["linkedin_url"] is None  # model invented a URL not in the source → dropped
+
+
+def test_revise_linkedin_taken_only_from_instruction():
+    proposal = {"intent": "new_investor", "investor_name": "Acme", "contact_name": "Jane",
+                "contact_email": "jane@acme.com", "contact_title": None, "city": None,
+                "linkedin_url": None, "note": None, "_source_text": "Acme Jane jane@acme.com"}
+    r1 = parse.revise(proposal, "her linkedin is linkedin.com/in/janedoe",
+                      parse_fn=_stub({"linkedin_url": "linkedin.com/in/janedoe"}))
+    assert r1["linkedin_url"] == "linkedin.com/in/janedoe"
+    # model tries to set a URL but the instruction carries none → keep existing (None)
+    r2 = parse.revise(proposal, "set her title to GP",
+                      parse_fn=_stub({"linkedin_url": "linkedin.com/in/fake", "contact_title": "GP"}))
+    assert r2["linkedin_url"] is None
+    assert r2["contact_title"] == "GP"
+
+
+def test_phone_and_mobile_kept_when_digits_in_source():
+    # A card transcription separates Phone/Mobile/Fax; the model maps office->phone, cell->mobile.
+    src = ("New investor — from a business card:\nName: Daniel Raupp\nCompany: Fortitude\n"
+           "Phone: 631-474-5610\nFax: 631-474-1806\nCell: 631-922-1195")
+    p = parse.parse_message(
+        src,
+        parse_fn=_stub({"intent": "new_investor", "investor_name": "Fortitude",
+                        "contact_name": "Daniel Raupp", "phone": "631-474-5610",
+                        "mobile": "631-922-1195"}),
+    )
+    assert p["phone"] == "631-474-5610"
+    assert p["mobile"] == "631-922-1195"  # the cell, kept in its printed formatting
+
+
+def test_fabricated_phone_dropped_when_digits_not_in_source():
+    p = parse.parse_message(
+        "new prospect Gamma Partners, talked to their GP",
+        parse_fn=_stub({"intent": "new_investor", "investor_name": "Gamma Partners",
+                        "contact_name": "their GP", "phone": "555-867-5309"}),
+    )
+    assert p["phone"] is None  # number not in the source → never minted
+
+
+def test_revise_phone_taken_only_from_instruction():
+    proposal = {"intent": "new_investor", "investor_name": "Acme", "contact_name": "Jane",
+                "contact_email": None, "contact_title": None, "city": None, "linkedin_url": None,
+                "phone": None, "mobile": None, "note": None, "_source_text": "Acme Jane"}
+    r1 = parse.revise(proposal, "her cell is 917-555-0199",
+                      parse_fn=_stub({"mobile": "917-555-0199"}))
+    assert r1["mobile"] == "917-555-0199"
+    # model tries to set a number but the instruction has none → keep existing (None)
+    r2 = parse.revise(proposal, "set her title to GP",
+                      parse_fn=_stub({"mobile": "000-000-0000", "contact_title": "GP"}))
+    assert r2["mobile"] is None
+    assert r2["contact_title"] == "GP"
+
+
+def test_revise_cannot_empty_the_proposal():
+    proposal = {"intent": "new_investor", "investor_name": "Acme", "contact_name": "Jane",
+                "contact_email": None, "contact_title": None, "note": "x", "_source_text": "Acme Jane"}
+    revised = parse.revise(proposal, "clear it",
+                           parse_fn=_stub({"investor_name": None, "contact_name": None,
+                                           "contact_title": None, "note": None}))
+    assert revised["investor_name"] == "Acme" and revised["contact_name"] == "Jane"
+
+
+if __name__ == "__main__":
+    fns = [v for k, v in sorted(globals().items()) if k.startswith("test_") and callable(v)]
+    for fn in fns:
+        fn()
+        print(f"ok  {fn.__name__}")
+    print(f"\n{len(fns)} passed")

backend/matrix_intake/test_proposals.py

@@ -0,0 +1,213 @@
+"""Tests for the proposal store + approval state machine (pure logic, no network)."""
+import copy
+import os
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+import proposals  # noqa: E402
+
+SAMPLE = {"intent": "new_investor", "investor_name": "Acme Capital",
+          "contact_name": "Jane Doe", "contact_email": "jane@acme.com",
+          "contact_title": None, "note": "met at conf"}
+
+
+def test_store_put_get_pop():
+    s = proposals.ProposalStore()
+    assert not s.has("$root")
+    s.put("$root", SAMPLE)
+    assert s.has("$root")
+    assert s.get("$root")["investor_name"] == "Acme Capital"
+    assert s.pop("$root")["investor_name"] == "Acme Capital"
+    assert not s.has("$root")
+    assert s.pop("$missing") is None
+
+
+def test_store_any_pending():
+    s = proposals.ProposalStore()
+    assert not s.any_pending()
+    s.put("$r", SAMPLE)
+    assert s.any_pending()
+    s.pop("$r")
+    assert not s.any_pending()
+
+
+def test_interpret_yes_variants():
+    for t in ("yes", "Y", "approve", "  ok ", "👍"):
+        assert proposals.interpret_reply(t)[0] == "approve", t
+
+
+def test_interpret_no_variants():
+    for t in ("no", "N", "cancel", "discard", "❌"):
+        assert proposals.interpret_reply(t)[0] == "reject", t
+
+
+def test_interpret_edit_equals():
+    action, payload = proposals.interpret_reply("edit email=new@acme.com")
+    assert action == "edit"
+    assert payload == ("contact_email", "new@acme.com")
+
+
+def test_interpret_edit_colon_and_alias():
+    action, payload = proposals.interpret_reply("firm: Acme Capital LLC")
+    assert action == "edit"
+    assert payload == ("investor_name", "Acme Capital LLC")
+
+
+def test_interpret_edit_city_and_linkedin_aliases():
+    a1, p1 = proposals.interpret_reply("city: New York")
+    assert (a1, p1) == ("edit", ("city", "New York"))
+    a2, p2 = proposals.interpret_reply("linkedin=linkedin.com/in/jane")
+    assert (a2, p2) == ("edit", ("linkedin_url", "linkedin.com/in/jane"))
+
+
+def test_render_shows_city_and_linkedin_when_present():
+    p = {**SAMPLE, "city": "New York", "linkedin_url": "linkedin.com/in/jane"}
+    out = proposals.render(p)
+    assert "City: New York" in out
+    assert "LinkedIn: linkedin.com/in/jane" in out
+
+
+def test_interpret_edit_phone_and_mobile_aliases():
+    assert proposals.interpret_reply("phone=212-555-0100") == ("edit", ("phone", "212-555-0100"))
+    assert proposals.interpret_reply("cell: 917-555-0199") == ("edit", ("mobile", "917-555-0199"))
+
+
+def test_render_shows_phone_and_mobile_when_present():
+    p = {**SAMPLE, "phone": "212-555-0100", "mobile": "917-555-0199"}
+    out = proposals.render(p)
+    assert "Phone: 212-555-0100" in out
+    assert "Mobile: 917-555-0199" in out
+
+
+def test_interpret_unknown():
+    assert proposals.interpret_reply("maybe later")[0] == "unknown"
+
+
+def test_interpret_edit_colon_value_contains_equals():
+    # the '=' inside the value must not break parsing — split on ':' first, keep the rest
+    action, payload = proposals.interpret_reply("note: see deck=v2")
+    assert action == "edit"
+    assert payload == ("note", "see deck=v2")
+
+
+def test_claim_once_pop_guards_double_approve():
+    # the double-approve guard relies on pop() yielding the proposal exactly once;
+    # a second claim returns None so a racing second 'yes' is a no-op
+    s = proposals.ProposalStore()
+    s.put("$r", SAMPLE)
+    assert s.pop("$r") is not None
+    assert s.pop("$r") is None
+
+
+def test_edit_with_unknown_field_is_not_an_edit():
+    # an unknown field name must not silently become an edit
+    assert proposals.interpret_reply("edit zipcode=90210")[0] == "unknown"
+
+
+def test_apply_edit_is_nondestructive():
+    updated = proposals.apply_edit(SAMPLE, "contact_email", "x@y.com")
+    assert updated["contact_email"] == "x@y.com"
+    assert SAMPLE["contact_email"] == "jane@acme.com"  # original untouched
+
+
+def test_render_includes_fields_and_instructions():
+    text = proposals.render(SAMPLE)
+    assert "Acme Capital" in text
+    assert "jane@acme.com" in text
+    assert "yes" in text.lower() and "no" in text.lower()
+
+
+def test_render_meeting_note_variant():
+    note = dict(SAMPLE, intent="meeting_note")
+    assert "meeting note" in proposals.render(note).lower()
+
+
+def test_summary_line_new_vs_note():
+    new_line = proposals.summary_line(SAMPLE)
+    assert "Acme Capital" in new_line and "new investor" in new_line.lower()
+    note_line = proposals.summary_line(dict(SAMPLE, intent="meeting_note"))
+    assert "Acme Capital" in note_line and "meeting note" in note_line.lower()
+    # the nudge must point the user to the thread, where the actual action lives
+    assert "thread" in new_line.lower()
+
+
+# --- fuzzy-match disambiguation + conversational-revision helpers ---
+
+DISAMBIG = {"intent": "new_investor", "investor_name": "Charles Brown",
+            "contact_name": "Charles Brown", "contact_email": None, "contact_title": None,
+            "note": "met at conf", "_stage": "disambiguate",
+            "_candidates": [{"id": "rowCharlie", "name": "Charlie Brown", "score": 0.92, "matched_on": "name"},
+                            {"id": "rowBeta", "name": "Beta Capital LLC", "score": 0.7, "matched_on": "name"}]}
+
+
+def test_interpret_disambiguation_pick_number():
+    assert proposals.interpret_disambiguation("1", 2) == ("pick", 0)
+    assert proposals.interpret_disambiguation(" 2 ", 2) == ("pick", 1)
+    assert proposals.interpret_disambiguation("#1", 2) == ("pick", 0)
+
+
+def test_interpret_disambiguation_out_of_range_is_unknown():
+    assert proposals.interpret_disambiguation("3", 2)[0] == "unknown"
+    assert proposals.interpret_disambiguation("0", 2)[0] == "unknown"
+
+
+def test_interpret_disambiguation_new_and_no():
+    assert proposals.interpret_disambiguation("new", 2)[0] == "new"
+    assert proposals.interpret_disambiguation("none of these", 2)[0] == "new"
+    assert proposals.interpret_disambiguation("no", 2)[0] == "reject"
+
+
+def test_interpret_disambiguation_freeform_is_unknown():
+    # a free-form reply in the shortlist stage isn't guessed at — re-prompt instead
+    assert proposals.interpret_disambiguation("the first one", 2)[0] == "unknown"
+
+
+def test_attach_to_candidate_promotes_to_meeting_note():
+    out = proposals.attach_to_candidate(DISAMBIG, DISAMBIG["_candidates"][0])
+    assert out["_match_id"] == "rowCharlie"
+    assert out["intent"] == "meeting_note"
+    assert out["_stage"] == "approval"
+    assert out["investor_name"] == "Charlie Brown"  # canonical existing name shown
+    assert "_candidates" not in out
+    assert "_candidates" in DISAMBIG  # original untouched
+
+
+def test_promote_to_new_clears_shortlist_and_match():
+    out = proposals.promote_to_new(dict(DISAMBIG, _match_id="rowX"))
+    assert out["_stage"] == "approval"
+    assert "_candidates" not in out
+    assert "_match_id" not in out
+
+
+def test_disambiguation_pick_then_yes_reaches_approval():
+    # Closes the seam between the two state machines: a shortlist pick promotes the proposal to
+    # approval stage carrying the chosen investor's row id, and a following 'yes' classifies as
+    # approve (the normal commit path) — so pick -> yes lands the note on the existing investor.
+    picked = proposals.attach_to_candidate(copy.deepcopy(DISAMBIG), DISAMBIG["_candidates"][0])
+    assert picked["_stage"] == "approval"
+    assert picked["_match_id"] == "rowCharlie"
+    assert picked["intent"] == "meeting_note"
+    assert proposals.interpret_reply("yes") == ("approve", None)
+
+
+def test_render_disambiguation_lists_numbered_candidates():
+    text = proposals.render_disambiguation(DISAMBIG)
+    assert "Charlie Brown" in text and "Beta Capital LLC" in text
+    assert "1." in text and "2." in text
+    assert "new" in text.lower() and "no" in text.lower()
+
+
+def test_same_fields_ignores_control_keys():
+    a = dict(SAMPLE)
+    assert proposals.same_fields(a, dict(a))
+    assert not proposals.same_fields(a, dict(a, note="different"))
+    assert proposals.same_fields(a, dict(a, _match_id="r1", _stage="approval"))
+
+
+if __name__ == "__main__":
+    fns = [v for k, v in sorted(globals().items()) if k.startswith("test_") and callable(v)]
+    for fn in fns:
+        fn()
+        print(f"ok  {fn.__name__}")
+    print(f"\n{len(fns)} passed")

backend/matrix_intake/test_query.py

@@ -0,0 +1,112 @@
+"""Tests for the NL-query Matrix surface: trigger detection + answer rendering (pure, no network)."""
+import os
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+import query  # noqa: E402
+
+
+# ── parse_trigger ───────────────────────────────────────────────────────────────────────
+def test_trigger_question_mark():
+    assert query.parse_trigger("?who are our top investors") == "who are our top investors"
+    assert query.parse_trigger("  ? spaced out ") == "spaced out"
+
+
+def test_trigger_at_bot():
+    assert query.parse_trigger("@bot top 10 investors") == "top 10 investors"
+    assert query.parse_trigger("@bot: top 10 investors") == "top 10 investors"   # pill-style colon
+    assert query.parse_trigger("@BOT  spaced") == "spaced"                       # case-insensitive
+
+
+def test_trigger_slash_forms():
+    assert query.parse_trigger("/ask when did we last email Acme?") == "when did we last email Acme?"
+    assert query.parse_trigger("/query top investors") == "top investors"
+    assert query.parse_trigger("/q top investors") == "top investors"
+
+
+def test_trigger_bare_returns_empty_string():
+    # A bare trigger is matched (so we show help) but carries no question.
+    assert query.parse_trigger("@bot") == ""
+    assert query.parse_trigger("?") == ""
+
+
+def test_non_trigger_routes_to_intake():
+    assert query.parse_trigger("New investor: Acme — Jane <jane@acme.com>") is None
+    # 'ask' as a note verb must NOT trigger (would collide with real intake notes).
+    assert query.parse_trigger("Ask Jane to send the Q3 deck") is None
+    assert query.parse_trigger("/asking for a friend") is None  # needs a separator after /ask
+    assert query.parse_trigger("") is None
+    assert query.parse_trigger("   ") is None
+
+
+# ── render_answer ───────────────────────────────────────────────────────────────────────
+def test_render_scalar_rows():
+    out = query.render_answer({
+        "intent": "top_investors_committed", "slots": {"limit": 2},
+        "summary": "Top 2 investor(s) by committed capital.",
+        "columns": ["investor_name", "total_invested", "lead"],
+        "rows": [{"investor_name": "Acme Capital", "total_invested": 5000000, "lead": "Grant"},
+                 {"investor_name": "Beta Fund", "total_invested": 2500000, "lead": "Jonathan"}],
+        "truncated": False})
+    assert "Top 2 investor(s)" in out
+    assert "**Acme Capital**" in out
+    assert "$5,000,000" in out                       # money formatting
+    assert "read as: top_investors_committed" in out  # interpretation footer
+
+
+def test_render_nested_contacts_and_commitments():
+    out = query.render_answer({
+        "intent": "investor_lookup", "slots": {"name": "Acme"},
+        "summary": '1 investor(s) matching "Acme".',
+        "columns": ["investor_name", "lead", "total_invested", "graveyard", "contacts", "commitments"],
+        "rows": [{"investor_name": "Acme Capital", "lead": "Grant", "total_invested": 5000000,
+                  "graveyard": 0,
+                  "contacts": [{"full_name": "Jane Doe", "email": "jane@acme.com", "title": "GP",
+                                "city": "Austin", "state": "TX", "country": ""}],
+                  "commitments": [{"fund_name": "Fund I", "amount": 5000000}]}],
+        "truncated": False})
+    assert "Jane Doe <jane@acme.com>" in out
+    assert "Fund I: $5,000,000" in out
+    assert "graveyard" not in out  # 0-valued flag column suppressed
+
+
+def test_render_flag_when_set():
+    out = query.render_answer({
+        "intent": "investors_follow_up", "slots": {},
+        "summary": "1 investor(s) with an open follow-up reminder.",
+        "columns": ["investor_name", "title", "due_date", "status", "overdue"],
+        "rows": [{"investor_name": "Acme", "title": "Send deck", "due_date": "2026-01-01",
+                  "status": "open", "overdue": 1}]})
+    assert "⚠️ overdue" in out
+    assert "2026-01-01" in out  # date truncated to YYYY-MM-DD
+
+
+def test_render_no_rows():
+    out = query.render_answer({"intent": "investors_by_city", "slots": {"city": "Nowhere"},
+                               "summary": '0 investor contact(s) in "Nowhere".',
+                               "columns": [], "rows": []})
+    assert "no matching" in out.lower()
+
+
+def test_render_overflow_note():
+    rows = [{"investor_name": f"Inv {i}", "total_invested": i}
+            for i in range(query.MAX_DISPLAY_ROWS + 5)]
+    out = query.render_answer({"intent": "top_investors_committed", "slots": {}, "summary": "many",
+                               "columns": ["investor_name", "total_invested"], "rows": rows})
+    assert "+5 more not shown" in out
+
+
+def test_render_errors():
+    assert "couldn't map" in query.render_answer({"error": "no_match", "question": "huh"}).lower()
+    assert "unreachable" in query.render_answer({"error": "model_unavailable"}).lower()
+    assert "failed" in query.render_answer({"error": "query_failed", "detail": "boom"}).lower()
+    assert "bad_slot" in query.render_answer({"error": "bad_slot", "detail": "x"})
+
+
+if __name__ == "__main__":
+    fns = [v for k, v in sorted(globals().items()) if k.startswith("test_") and callable(v)]
+    for fn in fns:
+        fn()
+        print(f"ok  {fn.__name__}")
+    print(f"\n{len(fns)} passed")

backend/matrix_intake/test_spark.py

@@ -0,0 +1,45 @@
+"""Tests for the business-card vision wrapper (pure logic, no network — chat_fn is stubbed)."""
+import os
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+import spark  # noqa: E402
+
+
+def test_transcribe_card_returns_faithful_text():
+    captured = {}
+
+    def fake_chat(prompt, image_b64, mime="image/jpeg", system=None, max_tokens=600):
+        captured["image_b64"] = image_b64
+        captured["mime"] = mime
+        captured["system"] = system
+        return "Jane Doe\nGeneral Partner\nAcme Capital LLC\njane@acme.com\n+1 555 123 4567"
+
+    out = spark.transcribe_card("Zm9vYmFy", mime="image/png", chat_fn=fake_chat)
+    # The transcription is passed through verbatim — email survives for the integrity check.
+    assert "jane@acme.com" in out
+    assert "Acme Capital LLC" in out
+    # The image + mime reached the vision call; the card system prompt was used.
+    assert captured["image_b64"] == "Zm9vYmFy"
+    assert captured["mime"] == "image/png"
+    assert "business card" in (captured["system"] or "").lower()
+
+
+def test_transcribe_card_none_sentinel_becomes_empty():
+    # The model replies NONE for an unreadable / non-card image → we return "" so the bot can
+    # ask for a clearer photo instead of feeding garbage into the intake parser.
+    assert spark.transcribe_card("x", chat_fn=lambda *a, **k: "NONE") == ""
+    assert spark.transcribe_card("x", chat_fn=lambda *a, **k: " none ") == ""
+
+
+def test_transcribe_card_strips_whitespace():
+    assert spark.transcribe_card("x", chat_fn=lambda *a, **k: "  Acme\n  ") == "Acme"
+
+
+if __name__ == "__main__":
+    fns = [v for k, v in sorted(globals().items()) if k.startswith("test_") and callable(v)]
+    for fn in fns:
+        fn()
+        print(f"ok  {fn.__name__}")
+    print(f"\n{len(fns)} passed")

backend/mcp/architect_agent.py

@@ -58,7 +58,22 @@ def _render_thesis(thesis):
 def _system(thesis):
     text = ("You are the Architect, the in-house copilot that sharpens Ten31's investment "
             "thesis with the partners. Ten31 invests in critical infrastructure across bitcoin, "
-            "AI, energy, and freedom technologies, with scarcity as the connecting idea. "
+            "AI, energy, and freedom technologies. The spine of the thesis: fiat is being debased "
+            "while AI drives the marginal cost of anything reproducible toward zero, so durable "
+            "value migrates to what stays provably scarce and verifiable. Bitcoin is the apex form "
+            "of that, a fixed-supply, non-debasable, verifiable reserve asset. AI is the abundance "
+            "engine and bitcoin is the scarcity anchor, two faces of one megatrend. The throughline "
+            "is an asset-value and capital-flow claim: as money debases and AI commoditizes the "
+            "reproducible, value accrues to the scarce side of one supply chain and the monetary "
+            "premium accrues to bitcoin as the non-debasable reserve asset. This is not a claim "
+            "that the world transacts, settles, or clears in bitcoin. The structure runs on three "
+            "seams: Energy and Compute (the same scarce firm power feeds both AI and bitcoin), "
+            "Debasement and Bitcoin (bitcoin as reserve and as pristine collateral for credit, "
+            "never payments), and AI and Data-Ownership (sovereign data and confidential inference, "
+            "the own-your-stack and own-your-inference layer). Strike is a financial-services and "
+            "reserve re-rate, never a payments story. Proof standard: every proof point must be "
+            "falsifiable as scaled substance with a number, never a first-instance milestone. Do "
+            "not invent proof and do not over-expose sensitive deal or return specifics. "
             f"VOICE RULES (follow exactly): {VOICE}\n\n"
             "Here is the current working thesis:\n" + _render_thesis(thesis))
     # Cache the thesis context so iterating across requests is cheap.

backend/mcp/outreach_agent.py

@@ -0,0 +1,258 @@
+"""Outreach drafting agent — tailored LP outreach in Ten31's voice, grounded in the
+thesis + the LP's DE-IDENTIFIED context, through the redaction boundary.
+
+Draft-only: a human reviews, edits, and sends (guardrails #4 and #6 — no auto-send,
+no cold/outbound automation until counsel defines the solicitation posture). Sovereignty:
+the thesis is Ten31's own non-sensitive messaging and goes to Claude as-is; the LP's
+context (CRM notes + email history) is scrubbed first, so the LP list never reaches the
+API in the clear, and the draft is re-hydrated locally for the human.
+"""
+import json
+import os
+import sys
+
+_HERE = os.path.dirname(os.path.abspath(__file__))
+sys.path.insert(0, _HERE)  # backend/mcp on path for sibling imports (architect_grounding, architect_agent)
+
+# outreach_type -> human description woven into the prompt
+OUTREACH_TYPES = {
+    "intro": "a first introduction to Ten31 and the fund",
+    "follow_up": "a warm follow-up that moves the conversation forward",
+    "fund_update": "a fund update / progress note",
+    "meeting_follow_up": "a follow-up after a recent meeting or call",
+    "nurture": "a light-touch note to stay in contact",
+}
+
+
+def _days_between(then_iso, now_iso):
+    from datetime import datetime
+    try:
+        a = datetime.strptime(str(then_iso)[:10], "%Y-%m-%d")
+        b = datetime.strptime(str(now_iso)[:10], "%Y-%m-%d")
+        return (b - a).days
+    except Exception:
+        return None
+
+
+def follow_up_radar(conn, our_addresses, now_iso, warm_days=45, limit=60):
+    """Deterministic scan: surface investors who need attention, each with a concrete,
+    checkable reason (no LLM guesswork in the *surfacing*). Tiers, most urgent first:
+      0  you owe a reply (their email is the most recent, unanswered)
+      1  flagged for follow-up and quiet
+      2  warm lead gone quiet (no contact in >= warm_days)
+    """
+    own = {(a or "").lower() for a in (our_addresses or [])}
+    try:
+        rows = conn.execute("SELECT * FROM fundraising_investors").fetchall()
+    except Exception:
+        return []
+    items = []
+    for r in rows:
+        d = dict(r)
+        inv_id, name = d.get("id"), d.get("investor_name")
+        if not inv_id:
+            continue
+        gv = d.get("graveyard")
+        if gv and str(gv).strip().lower() not in ("", "0", "false", "no"):
+            continue  # buried leads are out of scope
+        try:
+            erows = conn.execute(
+                "SELECT e.from_email, e.sent_at FROM emails e "
+                "JOIN email_investor_links l ON l.email_id = e.id "
+                "WHERE l.fundraising_investor_id = ? AND e.is_matched = 1 "
+                "ORDER BY e.sent_at DESC LIMIT 50", (inv_id,)).fetchall()
+        except Exception:
+            erows = []
+        if not erows:
+            continue  # no email history -> nothing to base a nudge on
+        last = erows[0]
+        days = _days_between(last["sent_at"], now_iso)
+        if days is None:
+            continue
+        inbound_last = (last["from_email"] or "").lower() not in own  # they emailed last
+        ff = d.get("follow_up")
+        flagged = bool(ff) and str(ff).strip().lower() not in ("", "0", "false", "no")
+
+        reason, tier, suggested = None, None, "follow_up"
+        if inbound_last and days >= 3:
+            reason, tier, suggested = f"You owe a reply — they emailed {days} days ago", 0, "follow_up"
+        elif flagged and days >= 14:
+            reason, tier, suggested = f"Flagged for follow-up, quiet {days} days", 1, "follow_up"
+        elif days >= warm_days and len(erows) >= 2:
+            reason, tier, suggested = f"No contact in {days} days", 2, "nurture"
+        if reason is None:
+            continue
+        if flagged and tier != 1:
+            reason += " · flagged"
+        items.append({"investor_id": inv_id, "name": name, "reason": reason,
+                      "days_since": days, "suggested_type": suggested, "tier": tier})
+    items.sort(key=lambda x: (x["tier"], -x["days_since"]))
+    return items[:limit]
+
+
+def _context(conn, investor_id):
+    """Assemble the recipient's context. Structured so the model replies to the ACTIVE
+    conversation (the most recent email thread) while still having earlier emails as
+    background. Returns (investor_name, context_text) or (None, None)."""
+    row = conn.execute("SELECT investor_name, notes FROM fundraising_investors WHERE id=?",
+                       (investor_id,)).fetchone()
+    if not row:
+        return None, None
+    name = row["investor_name"]
+    header = [f"Investor: {name}"]
+    notes = (row["notes"] or "").strip()
+    if notes:
+        header.append("CRM notes:\n" + notes)
+    try:
+        rows = conn.execute(
+            "SELECT e.subject, e.body_text, e.snippet, e.sent_at, e.thread_id FROM emails e "
+            "JOIN email_investor_links l ON l.email_id = e.id "
+            "WHERE l.fundraising_investor_id = ? AND e.is_matched = 1 "
+            "ORDER BY e.sent_at DESC LIMIT 20", (investor_id,)).fetchall()
+    except Exception:
+        rows = []  # email tables may be absent / not yet captured
+    active, background = [], []
+    if rows:
+        active_thread = rows[0]["thread_id"]
+        for em in rows:
+            body = (em["body_text"] or em["snippet"] or "")[:1500].strip()
+            block = f"({(em['sent_at'] or '')[:10]}) {em['subject'] or '(no subject)'}\n{body}"
+            in_active = active_thread is not None and em["thread_id"] == active_thread
+            (active if in_active else background).append(block)
+    sections = ["\n".join(header)]
+    if active:
+        sections.append("=== Active conversation (the most recent thread — this is what you are replying to) ===\n"
+                        + "\n\n".join(reversed(active[:6])))
+    if background:
+        sections.append("=== Earlier emails (background only, not the active thread) ===\n"
+                        + "\n\n".join(background[:4]))
+    return name, "\n\n".join(sections)
+
+
+# Keyword cues used to pick the sender's prior emails of the SAME PURPOSE as the draft
+# (so the voice few-shot matches what they're writing, not just whatever is most recent).
+PURPOSE_PATTERNS = {
+    "intro": ["introduc", "nice to meet", "reaching out", "wanted to connect", "by way of introduction", "e-meet"],
+    "follow_up": ["follow up", "following up", "circle back", "circling back", "checking in",
+                  "wanted to revisit", "any thoughts", "wanted to follow", "touching base"],
+    "fund_update": ["update", "progress", "quarter", "deployed", "portfolio", "milestone", "closing", "fund iii"],
+    "meeting_follow_up": ["great to meet", "great speaking", "thanks for the call", "thanks for your time",
+                          "after our", "following our", "enjoyed our", "great to connect", "great chatting"],
+    "nurture": ["checking in", "hope you", "thinking of you", "stay in touch", "wanted to share", "thought you"],
+}
+
+
+def _voice_examples(conn, sender_email, outreach_type=None, limit=8):
+    """The sender's OWN sent LP emails OF THE SAME PURPOSE — used as voice few-shot AND
+    surfaced for transparency (no black box). Larger sample, purpose-weighted (not just
+    recent). Returns (blocks_for_model, meta_for_ui); meta is the sender's own emails."""
+    if not sender_email:
+        return [], []
+    try:
+        rows = conn.execute(
+            "SELECT subject, body_text, snippet, sent_at, to_emails_json FROM emails "
+            "WHERE LOWER(from_email) = LOWER(?) AND is_matched = 1 "
+            "AND body_text IS NOT NULL AND TRIM(body_text) <> '' "
+            "ORDER BY sent_at DESC LIMIT 80", (sender_email,)).fetchall()
+    except Exception:
+        return [], []
+    pats = PURPOSE_PATTERNS.get(outreach_type or "", [])
+    scored = []
+    for idx, r in enumerate(rows):
+        text = ((r["subject"] or "") + " " + (r["body_text"] or r["snippet"] or "")).lower()
+        score = sum(1 for p in pats if p in text)
+        scored.append((score, -idx, r))  # purpose match first, then more recent
+    scored.sort(key=lambda x: (x[0], x[1]), reverse=True)
+    blocks, meta = [], []
+    for score, _neg_idx, r in scored[:limit]:
+        body = (r["body_text"] or r["snippet"] or "")[:900].strip()
+        if not body:
+            continue
+        blocks.append(f"Example — {r['subject'] or '(no subject)'}\n{body}")
+        to = ""
+        try:
+            arr = json.loads(r["to_emails_json"] or "[]")
+            if arr:
+                to = arr[0].get("email") if isinstance(arr[0], dict) else arr[0]
+        except Exception:
+            to = ""
+        meta.append({"subject": r["subject"] or "(no subject)", "date": (r["sent_at"] or "")[:10],
+                     "to": to, "on_topic": score > 0})
+    return blocks, meta
+
+
+def _draft_with_claude(aa, thesis, type_desc, deident_context, deident_voice, guidance):
+    voice_block = ""
+    if deident_voice:
+        voice_block = ("\n\nHere are examples of how THIS sender actually writes (de-identified). Match their "
+                       "voice, tone, sentence rhythm, openers, and sign-off — not just the rules above:\n\n"
+                       + "\n\n---\n\n".join(deident_voice))
+    system = (
+        "You are Ten31's outreach copilot. Draft ONE ready-to-send LP outreach email in the SENDER's voice. "
+        f"VOICE RULES (follow exactly): {aa.VOICE}" + voice_block + "\n\n"
+        "Ten31 invests in critical infrastructure across bitcoin, AI, energy, and freedom technologies. "
+        "The spine: fiat is being debased while AI drives the marginal cost of the reproducible toward "
+        "zero, so durable value accrues to what stays provably scarce, and the monetary premium accrues "
+        "to bitcoin as the apex non-debasable reserve asset. AI is the abundance engine and bitcoin is "
+        "the scarcity anchor. Ten31 owns the scarce links of that one supply chain. This is an "
+        "asset-value and capital-flow conviction, not a claim that the world transacts or settles in "
+        "bitcoin. Current working thesis:\n" + aa._render_thesis(thesis) + "\n\n"
+        "The recipient's context below is DE-IDENTIFIED: people, firms, and amounts appear as placeholders "
+        "like [PERSON_1], [ORG_1], [AMOUNT_1]. Keep every placeholder EXACTLY as written and NEVER invent new "
+        "ones — they are swapped back to real values after you reply. Reply to the ACTIVE conversation; use the "
+        "earlier emails only as background. Output a subject line, then the email body. Do NOT fabricate facts, "
+        "numbers, returns, or commitments that are not present in the context or the thesis.")
+    user = (f"Outreach type: {type_desc}\n\n"
+            f"Recipient context (de-identified):\n{deident_context}\n\n"
+            + (f"Additional guidance from the sender: {guidance}\n\n" if (guidance or "").strip() else "")
+            + "Draft the email now.")
+    resp = aa._client().messages.create(
+        model=aa.MODEL, max_tokens=1200,
+        system=[{"type": "text", "text": system, "cache_control": {"type": "ephemeral"}}],
+        messages=[{"role": "user", "content": user}])
+    return "".join(b.text for b in resp.content if getattr(b, "type", None) == "text")
+
+
+def draft_outreach(conn, investor_id, outreach_type, guidance, db_path, sender_email=None):
+    """Draft tailored outreach for one investor, in the SENDER's voice (few-shot from
+    their own prior emails). FAILS CLOSED: if the scrub can't be prepared or Claude
+    hallucinates a placeholder, no de-anonymized draft is returned."""
+    name, context = _context(conn, investor_id)
+    if not name:
+        return {"status": "not_found"}
+    type_desc = OUTREACH_TYPES.get(outreach_type, OUTREACH_TYPES["follow_up"])
+    voice_blocks, voice_meta = _voice_examples(conn, sender_email, outreach_type)
+
+    # 1) Scrub the sender's voice examples + the recipient context TOGETHER (shared token
+    #    space). The recipient context is free-prose email bodies, so the dictionary+regex
+    #    floor is NOT enough — pass the local-Qwen NER backstop (as architect_grounding does)
+    #    to tokenize unknown people/firms not in the CRM. FAILS CLOSED: if the local model is
+    #    unreachable, _ner_local raises here and no de-anonymized draft is returned.
+    try:
+        sys.path.insert(0, os.path.dirname(_HERE))  # backend/ for the redaction package
+        from redaction.client import Boundary
+        from architect_grounding import _ner_local  # local-Qwen NER backstop (sibling module)
+        boundary = Boundary(db_path=db_path, actor="closer", ner_fn=_ner_local)
+        scrubbed = boundary.scrub(list(voice_blocks) + [context], bucket=False, conn=conn)
+    except Exception as exc:
+        return {"status": "scrub_unavailable", "reason": str(exc)}
+    items = scrubbed["items"]
+    deident_voice, deident_target = items[:-1], items[-1]
+    handle = scrubbed["handle"]
+
+    # 2) Claude drafts over the de-identified context + voice + (non-sensitive) thesis.
+    try:
+        import architect_agent as aa
+        thesis = aa.at.get_thesis("core", db=db_path)
+        raw = _draft_with_claude(aa, thesis, type_desc, deident_target, deident_voice, guidance)
+    except Exception as exc:
+        boundary.forget(handle)
+        return {"status": "claude_not_configured", "reason": str(exc)}
+
+    # 3) Re-hydrate locally (strict: a hallucinated placeholder quarantines the draft).
+    rehy = boundary.rehydrate(raw, handle, strict=True, conn=conn)
+    boundary.forget(handle)
+    if rehy.get("error"):
+        return {"status": "rehydrate_failed"}
+    return {"status": "ok", "draft": rehy["text"], "investor_name": name,
+            "scrub_stats": scrubbed.get("stats", {}), "voice_examples": voice_meta}

backend/mcp/test_outreach.py

@@ -0,0 +1,118 @@
+#!/usr/bin/env python3
+"""Test the outreach agent's context assembly: it pulls the investor's CRM notes +
+recent matched email into the de-identifiable context block. Synthetic data only
+(guardrail #9). The scrub/Claude/rehydrate round-trip is exercised live in the preview.
+Run: cd backend && python3 mcp/test_outreach.py
+"""
+import os
+import sqlite3
+import sys
+import tempfile
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import outreach_agent as oa  # noqa: E402
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+def main():
+    db = os.path.join(tempfile.mkdtemp(), "t.db")
+    c = sqlite3.connect(db)
+    c.row_factory = sqlite3.Row
+    c.executescript("""
+        CREATE TABLE fundraising_investors (id TEXT PRIMARY KEY, investor_name TEXT, notes TEXT);
+        CREATE TABLE emails (id TEXT PRIMARY KEY, subject TEXT, body_text TEXT, snippet TEXT, sent_at TEXT,
+            from_email TEXT, to_emails_json TEXT, thread_id TEXT, is_matched INT);
+        CREATE TABLE email_investor_links (id TEXT, email_id TEXT, fundraising_investor_id TEXT);
+    """)
+    c.execute("INSERT INTO fundraising_investors VALUES ('inv1','Harbor & Vine','Met at the conference; interested in Fund III.')")
+    c.executemany("INSERT INTO emails (id,subject,body_text,sent_at,thread_id,is_matched) VALUES (?,?,?,?,?,1)", [
+        ("e1", "Re: Fund III", "Thanks for the call. We are still weighing the lock-up terms.", "2026-06-02T10:00:00", "t1"),
+        ("e2", "Intro", "Good to meet you at the dinner.", "2026-05-01T10:00:00", "t0"),
+        ("e3", "Spam", "ignore me", "2026-04-01T10:00:00", "t9"),  # not linked -> excluded
+    ])
+    c.executemany("INSERT INTO email_investor_links (id,email_id,fundraising_investor_id) VALUES (?,?, 'inv1')",
+                  [("l1", "e1"), ("l2", "e2")])
+    c.commit()
+
+    name, ctx = oa._context(c, "inv1")
+    check(name == "Harbor & Vine", f"resolves investor name (got {name!r})")
+    check("Met at the conference" in ctx, "includes CRM notes")
+    check("lock-up terms" in ctx, "active-thread email present")
+    check("Good to meet you" in ctx, "earlier email present as background")
+    check("ignore me" not in ctx, "excludes email not linked to this investor")
+    check("Active conversation" in ctx and "Earlier emails" in ctx
+          and ctx.index("lock-up terms") < ctx.index("Good to meet you"),
+          "active thread is separated from background, active first")
+
+    # voice examples: the sender's own sent emails (few-shot + transparency)
+    c.execute("INSERT INTO emails (id,subject,body_text,sent_at,from_email,to_emails_json,thread_id,is_matched) "
+              "VALUES ('v1','My note','Hi there, quick update on the fund. Best, Grant',"
+              "'2026-06-01T10:00:00','grant@ten31.xyz','[{\"email\":\"lp@x.example\"}]','tv',1)")
+    c.commit()
+    blocks, meta = oa._voice_examples(c, "grant@ten31.xyz")
+    check(len(blocks) == 1 and "quick update on the fund" in blocks[0], "voice example pulls the sender's own email")
+    check(len(meta) == 1 and meta[0]["subject"] == "My note" and meta[0]["to"] == "lp@x.example",
+          "voice meta carries subject + recipient for transparency")
+    check(oa._voice_examples(c, None) == ([], []), "no sender -> no voice examples")
+
+    n2, c2 = oa._context(c, "missing")
+    check(n2 is None and c2 is None, "unknown investor -> (None, None)")
+
+    # type catalogue is intact
+    check(set(["intro", "follow_up", "fund_update", "meeting_follow_up", "nurture"]) <= set(oa.OUTREACH_TYPES),
+          "outreach types catalogue present")
+
+    # ── follow-up radar ──
+    rc = sqlite3.connect(os.path.join(tempfile.mkdtemp(), "radar.db"))
+    rc.row_factory = sqlite3.Row
+    rc.executescript("""
+        CREATE TABLE fundraising_investors (id TEXT PRIMARY KEY, investor_name TEXT, follow_up TEXT, graveyard TEXT);
+        CREATE TABLE emails (id TEXT PRIMARY KEY, from_email TEXT, sent_at TEXT, is_matched INT);
+        CREATE TABLE email_investor_links (id TEXT, email_id TEXT, fundraising_investor_id TEXT);
+    """)
+    rc.executemany("INSERT INTO fundraising_investors (id,investor_name,follow_up,graveyard) VALUES (?,?,?,?)", [
+        ("owe", "Owe Reply LP", None, None),     # they emailed last, 5d ago -> tier 0
+        ("warm", "Warm Quiet LP", None, None),   # we emailed last, 60d ago -> tier 2
+        ("fresh", "Fresh LP", None, None),       # we emailed 4d ago -> not surfaced
+        ("buried", "Buried LP", None, "1"),      # graveyard -> excluded
+    ])
+    OURS = "grant@ten31.xyz"
+    em = [
+        ("o1", "lp@owe.example", "2026-06-04T10:00:00", "owe"),     # inbound, 5 days before 06-09
+        ("w1", OURS, "2026-04-10T10:00:00", "warm"),                # outbound, ~60 days
+        ("w0", "lp@warm.example", "2026-04-01T10:00:00", "warm"),   # 2nd email for history
+        ("f1", OURS, "2026-06-05T10:00:00", "fresh"),               # outbound, 4 days -> too recent
+        ("b1", "lp@buried.example", "2026-01-01T10:00:00", "buried"),
+    ]
+    for eid, frm, sent, inv in em:
+        rc.execute("INSERT INTO emails (id,from_email,sent_at,is_matched) VALUES (?,?,?,1)", (eid, frm, sent))
+        rc.execute("INSERT INTO email_investor_links (id,email_id,fundraising_investor_id) VALUES (?,?,?)", (eid + "l", eid, inv))
+    rc.commit()
+    radar = oa.follow_up_radar(rc, [OURS], "2026-06-09T00:00:00", warm_days=45)
+    names = [x["name"] for x in radar]
+    check("Owe Reply LP" in names and "Warm Quiet LP" in names, f"surfaces owe-reply + warm-quiet (got {names})")
+    check("Fresh LP" not in names, "recent contact not surfaced")
+    check("Buried LP" not in names, "graveyard excluded")
+    check(radar[0]["name"] == "Owe Reply LP" and radar[0]["tier"] == 0, "owe-a-reply ranked first (tier 0)")
+    owe = next(x for x in radar if x["name"] == "Owe Reply LP")
+    check("owe a reply" in owe["reason"] and owe["suggested_type"] == "follow_up", "owe-reply reason + suggested type")
+    warm = next(x for x in radar if x["name"] == "Warm Quiet LP")
+    check(warm["tier"] == 2 and warm["suggested_type"] == "nurture", "warm-quiet is tier 2, suggests nurture")
+
+    if FAILS:
+        print(f"\nFAILED ({len(FAILS)})")
+        for f in FAILS:
+            print("  - " + f)
+        sys.exit(1)
+    print("\nALL PASS (outreach context assembly)")
+
+
+if __name__ == "__main__":
+    main()

backend/mcp/test_outreach_redaction.py

@@ -0,0 +1,123 @@
+#!/usr/bin/env python3
+"""Regression test for the outreach NER-backstop wiring (v0.1.0:74).
+
+The outreach draft path scrubs free-prose LP context (CRM notes + email bodies) before
+it reaches Claude. The dictionary+regex floor only tokenizes KNOWN CRM entities, so an
+UNKNOWN person/firm mentioned in an email body would otherwise reach Claude in the clear.
+The v74 fix wired the local-Qwen NER backstop into draft_outreach (outreach_agent.py:
+`Boundary(..., ner_fn=_ner_local)`) and made it FAIL CLOSED when the local model is down.
+
+This drives the real draft_outreach with Claude and the NER model stubbed (offline,
+synthetic — guardrail #9) and proves:
+  (1) an unknown name in an email body is tokenized AWAY from the Claude payload;
+  (2) it is re-hydrated locally so the human still sees the real name;
+  (3) the interaction_log captures no sensitive value;
+  (4) when the local NER model raises (unreachable), the path returns scrub_unavailable
+      and Claude is never called.
+
+Run: cd backend && python3 mcp/test_outreach_redaction.py
+"""
+import os
+import sqlite3
+import sys
+import tempfile
+
+_HERE = os.path.dirname(os.path.abspath(__file__))
+sys.path.insert(0, _HERE)                       # backend/mcp
+sys.path.insert(0, os.path.dirname(_HERE))      # backend (for the redaction package)
+
+import outreach_agent as oa            # noqa: E402
+import architect_grounding as G        # noqa: E402
+import architect_agent as aa           # noqa: E402  (imports OK offline; client is lazy)
+
+FAILS = []
+
+UNKNOWN = "Penelope Ashworth-Vane"      # a person in NO CRM table -> only NER can catch her
+INVESTOR = "Harbor & Vine"              # a known org (fundraising_investors) -> dictionary floor
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+def make_db():
+    path = os.path.join(tempfile.mkdtemp(), "crm.db")
+    c = sqlite3.connect(path)
+    c.row_factory = sqlite3.Row
+    c.executescript("""
+        CREATE TABLE fundraising_investors (id TEXT PRIMARY KEY, investor_name TEXT, notes TEXT);
+        CREATE TABLE emails (id TEXT PRIMARY KEY, subject TEXT, body_text TEXT, snippet TEXT, sent_at TEXT,
+            from_email TEXT, to_emails_json TEXT, thread_id TEXT, is_matched INT);
+        CREATE TABLE email_investor_links (id TEXT, email_id TEXT, fundraising_investor_id TEXT);
+        CREATE TABLE interaction_log (id TEXT PRIMARY KEY, ts TEXT, actor_type TEXT, actor_id TEXT, action TEXT,
+            target_type TEXT, target_id TEXT, payload TEXT, source TEXT, created_at TEXT);
+    """)
+    c.execute("INSERT INTO fundraising_investors VALUES ('inv1',?,?)",
+              (INVESTOR, "Warm on Fund III; weighing lock-up terms."))
+    # The active-thread email body names an UNKNOWN person in free prose.
+    c.execute("INSERT INTO emails (id,subject,body_text,sent_at,thread_id,is_matched) VALUES "
+              "('e1','Re: Fund III',?,?,'t1',1)",
+              (f"Thanks for the call. My partner {UNKNOWN} still has a lock-up objection.", "2026-06-02T10:00:00"))
+    c.execute("INSERT INTO email_investor_links (id,email_id,fundraising_investor_id) VALUES ('l1','e1','inv1')")
+    c.commit()
+    return path, c
+
+
+def main():
+    db_path, conn = make_db()
+
+    # Stub the thesis fetch (avoid the thesis DB dependency) and Claude. The NER stub stands
+    # in for the local-Qwen model; _draft_with_claude echoes the de-identified text back so
+    # re-hydration is exercised and we can inspect exactly what would have reached Claude.
+    aa.at.get_thesis = lambda *a, **k: {}
+    captured = {}
+
+    def fake_claude(aa_mod, thesis, type_desc, deident_target, deident_voice, guidance):
+        captured["target"] = deident_target
+        return deident_target   # passthrough -> rehydrate must restore the real name
+
+    oa._draft_with_claude = fake_claude
+    G._ner_local = lambda text: [(UNKNOWN, "PERSON")]   # local model UP, finds the unknown name
+
+    # ── A) unknown name is tokenized away from Claude, restored locally ──
+    print("\n[A — NER backstop tokenizes an unknown name in outreach]")
+    res = oa.draft_outreach(conn, "inv1", "follow_up", "", db_path, sender_email=None)
+    check(res.get("status") == "ok", f"draft ok (status={res.get('status')})")
+    sent = captured.get("target", "")
+    check(UNKNOWN not in sent, "unknown name absent from the Claude payload (NER tokenized it)")
+    check(INVESTOR not in sent, "known investor org absent from the Claude payload (dictionary floor)")
+    check("lock-up" in sent, "objection substance survives to Claude")
+    check(UNKNOWN in res.get("draft", ""), "unknown name re-hydrated locally for the human")
+
+    blob = " ".join(r[0] for r in conn.execute("SELECT payload FROM interaction_log WHERE payload IS NOT NULL"))
+    check(UNKNOWN not in blob and INVESTOR not in blob, "interaction_log carries NO sensitive value")
+
+    # ── B) FAIL CLOSED: local NER model unreachable -> no Claude call ──
+    print("\n[B — fail closed: local NER model down]")
+    called = {"claude": False}
+
+    def boom(text):
+        raise RuntimeError("Spark Control unreachable")
+
+    G._ner_local = boom
+    oa._draft_with_claude = lambda *a, **k: called.__setitem__("claude", True) or a[3]
+    res2 = oa.draft_outreach(conn, "inv1", "follow_up", "", db_path, sender_email=None)
+    check(res2.get("status") == "scrub_unavailable", f"status scrub_unavailable (got {res2.get('status')})")
+    check(bool(res2.get("reason")), "scrub_unavailable carries the propagated NER failure reason (non-vacuous)")
+    check(called["claude"] is False, "Claude was NOT called when the NER model is down (fail closed)")
+    check("draft" not in res2, "no draft returned when scrub fails closed")
+
+    conn.close()
+    print()
+    if FAILS:
+        print(f"FAILED ({len(FAILS)}):")
+        for f in FAILS:
+            print(f"  - {f}")
+        sys.exit(1)
+    print("ALL PASS (outreach NER-backstop wiring + fail-closed)")
+
+
+if __name__ == "__main__":
+    main()

backend/migrations/0001_phase0_foundation.sql

@@ -113,4 +113,8 @@ ALTER TABLE contacts       ADD COLUMN deleted_at TEXT;
 ALTER TABLE organizations  ADD COLUMN deleted_at TEXT;
 ALTER TABLE opportunities  ADD COLUMN deleted_at TEXT;
 ALTER TABLE communications ADD COLUMN deleted_at TEXT;
-ALTER TABLE lp_profiles    ADD COLUMN deleted_at TEXT;
+-- lp_profiles ALTER removed (v0.1.0:104): the lp_profiles table is dropped in
+-- 0008_drop_retired_tables.sql and is no longer created by init_db(), so this
+-- ALTER would fail "no such table" on a fresh install. Live DBs already applied
+-- this migration (with the original ALTER) before lp_profiles was dropped, so
+-- removing the line here only affects fresh DBs — same end state either way.

backend/migrations/0005_grid_pipeline_link.down.sql

@@ -0,0 +1,7 @@
+-- Reversal of 0005_grid_pipeline_link.sql (manual; .down files are never auto-applied).
+--
+-- SQLite < 3.35 cannot DROP COLUMN. The added column is nullable and ignored by any code
+-- path predating it, so leaving it in place is harmless. The index drops freely. On
+-- SQLite >= 3.35 the column itself may also be dropped.
+DROP INDEX IF EXISTS idx_opportunities_fr_investor;
+-- ALTER TABLE opportunities DROP COLUMN fundraising_investor_id;  -- SQLite >= 3.35 only

backend/migrations/0005_grid_pipeline_link.sql

@@ -0,0 +1,22 @@
+-- Grid → Pipeline adoption — a durable link from a fundraising-grid investor to its
+-- Pipeline opportunity row.
+--
+-- ADDITIVE + REVERSIBLE (CLAUDE.md guardrail #3): adds one nullable column + index.
+-- Until now the grid's "Create Opportunity" button fired a one-shot POST with no
+-- back-reference, so a grid investor could spawn unlimited duplicate opportunities and
+-- an opp never knew which grid row it belonged to. opportunities.fundraising_investor_id
+-- records the link (set by the new POST /api/fundraising/pipeline/link endpoint), making
+-- the relationship dedup-able and reconcilable. "Is this investor in the pipeline?" and
+-- "what stage?" are then DERIVED from a live join on this column — deliberately not a
+-- denormalized mirror flag on fundraising_investors, which would only reintroduce the
+-- two-model drift this CRM exists to fight.
+--
+-- fundraising_investor_id is a LOGICAL foreign key to fundraising_investors(id). It is
+-- intentionally NOT a declared SQLite FOREIGN KEY: opportunities are soft-deleted (never
+-- hard-deleted) and fundraising_investors rows are rebuilt on every grid save, so there
+-- is nothing to cascade; SQLite's ALTER TABLE ADD COLUMN cannot add an enforced FK
+-- cleanly anyway. Nullable so every existing opportunity stays valid — a manually-created,
+-- non-grid opportunity simply has NULL here.
+ALTER TABLE opportunities ADD COLUMN fundraising_investor_id TEXT;
+
+CREATE INDEX IF NOT EXISTS idx_opportunities_fr_investor ON opportunities(fundraising_investor_id);

backend/migrations/0006_reminders.down.sql

@@ -0,0 +1,8 @@
+-- Manual rollback for 0006_reminders.sql (never auto-applied).
+-- Drops the whole reminders feature table. Per the never-hard-delete guardrail this
+-- discards reminder history, so only run it to reverse a bad migration on a dev/copy DB.
+DROP INDEX IF EXISTS idx_reminders_assignee;
+DROP INDEX IF EXISTS idx_reminders_due;
+DROP INDEX IF EXISTS idx_reminders_status;
+DROP INDEX IF EXISTS idx_reminders_investor;
+DROP TABLE IF EXISTS reminders;

backend/migrations/0006_reminders.sql

@@ -0,0 +1,45 @@
+-- Reminders & follow-ups — a real tickler/task model tied to the fundraising grid.
+--
+-- ADDITIVE + REVERSIBLE (CLAUDE.md guardrail #3): one new table + indexes; nothing
+-- existing is touched. Until now the only follow-up surfaces were the grid's binary
+-- `follow_up` checkbox (no date, owner, or status) and communications.next_action_date
+-- (tied to a single logged comm). This gives investors first-class reminders with a due
+-- date, status lifecycle, assignee, and provenance — the foundation for "who needs a
+-- follow-up?" queries, the daily digest's due/overdue section, and (later) bot-proposed
+-- reminders behind the Matrix approval gate.
+--
+-- investor_id is a LOGICAL foreign key to fundraising_investors(id) — deliberately NOT a
+-- declared SQLite FOREIGN KEY, matching opportunities.fundraising_investor_id (migration
+-- 0005). fundraising_investors rows are upserted by source_row_id on every grid save with
+-- a STABLE id (so the link survives saves), but a row dropped from the grid is DELETEd —
+-- there is nothing to cascade, and reconcile_grid_reminders() cancels the orphans on the
+-- next save (the pipeline reconciler's twin). investor_name is denormalized so a reminder
+-- stays readable in history even after its grid row is gone. investor_id is nullable: a
+-- reminder can be a standalone team task not tied to any investor.
+--
+-- contact_id is an optional logical FK to contacts(id) (the specific person). assignee_id
+-- is a logical ref to users(id) (NULL = team-wide). created_by holds a users.id OR a
+-- non-user sentinel ('bot'/'automation'), so it is plain TEXT with no FK.
+CREATE TABLE IF NOT EXISTS reminders (
+    id            TEXT PRIMARY KEY,
+    investor_id   TEXT,                                   -- logical FK -> fundraising_investors.id (NULL = standalone task)
+    investor_name TEXT,                                   -- denormalized; survives grid-row deletion
+    contact_id    TEXT,                                   -- optional logical FK -> contacts.id
+    title         TEXT NOT NULL,
+    details       TEXT,
+    due_date      TEXT,                                   -- ISO date 'YYYY-MM-DD' (or datetime)
+    status        TEXT NOT NULL DEFAULT 'open',           -- open | done | snoozed | cancelled
+    snoozed_until TEXT,
+    assignee_id   TEXT,                                   -- logical ref -> users.id; NULL = team-wide
+    created_by    TEXT,                                   -- users.id, or 'bot' / 'automation'
+    source        TEXT NOT NULL DEFAULT 'human',          -- human | bot | automation
+    completed_at  TEXT,
+    created_at    TEXT DEFAULT (datetime('now')),
+    updated_at    TEXT DEFAULT (datetime('now')),
+    deleted_at    TEXT
+);
+
+CREATE INDEX IF NOT EXISTS idx_reminders_investor ON reminders(investor_id) WHERE deleted_at IS NULL;
+CREATE INDEX IF NOT EXISTS idx_reminders_status   ON reminders(status)      WHERE deleted_at IS NULL;
+CREATE INDEX IF NOT EXISTS idx_reminders_due      ON reminders(due_date)    WHERE deleted_at IS NULL;
+CREATE INDEX IF NOT EXISTS idx_reminders_assignee ON reminders(assignee_id) WHERE deleted_at IS NULL;

backend/migrations/0007_pipeline_stages_v2.down.sql

@@ -0,0 +1,14 @@
+-- Reversal of 0007_pipeline_stages_v2.sql (manual; .down files are never auto-applied).
+--
+-- BEST-EFFORT: the 6->4 stage collapse is lossy and cannot be perfectly inverted (the
+-- pattern other .down files here share -- e.g. 0005 cannot DROP COLUMN on old SQLite). It
+-- restores VALID legacy 6-stage values, choosing a representative for each collapsed pair:
+--   engaged    was outreach OR meeting  -> 'meeting'        (representative)
+--   diligence  -> 'due_diligence'                            (exact)
+--   commitment was committed OR funded  -> 'committed'      (representative)
+-- Opportunities archived from the stray 'lost' value still carry stage = 'lost' but cannot be
+-- re-identified as "archived by this migration" vs archived for other reasons, so they are
+-- left archived; un-archive (clear deleted_at) manually if a rollback truly needs them back.
+UPDATE opportunities SET stage = 'meeting'        WHERE stage = 'engaged';
+UPDATE opportunities SET stage = 'due_diligence'  WHERE stage = 'diligence';
+UPDATE opportunities SET stage = 'committed'      WHERE stage = 'commitment';

backend/migrations/0007_pipeline_stages_v2.sql

@@ -0,0 +1,25 @@
+-- Pipeline funnel v2 — collapse the inherited 6-stage opportunity funnel into the locked
+-- 4-stage per-investor funnel: lead -> engaged -> diligence -> commitment, terminal at
+-- commitment. See ROADMAP "Pipeline stages + investor flags/labels -- LOCKED SPEC" (2026-06-19)
+-- and server.PIPELINE_STAGES.
+--
+-- DATA-ONLY + DEPLOYMENT-STATE-INVARIANT (migrations guide): targets stage values
+-- structurally, so it is a no-op on a fresh DB (no opportunities) and remaps deterministically
+-- on a populated one.
+--   outreach, meeting   -> engaged      (a two-way conversation has begun; "meeting" was an
+--                                         activity, not a position, so it folds in here)
+--   due_diligence       -> diligence
+--   committed, funded   -> commitment   (terminal; post-commit $ lives in the grid fund cell,
+--                                         and fund admin owns post-commitment -- no "funded" stage)
+UPDATE opportunities SET stage = 'engaged'    WHERE stage IN ('outreach', 'meeting');
+UPDATE opportunities SET stage = 'diligence'  WHERE stage = 'due_diligence';
+UPDATE opportunities SET stage = 'commitment' WHERE stage IN ('committed', 'funded');
+
+-- The stray legacy 'lost' value is not in the new settable enum, and a lost deal is a dead
+-- deal: ARCHIVE (soft-delete) the opportunity rather than leave an un-settable stage on a live
+-- row. The grid investor row is left fully intact (the grid is canonical); graveyarding the
+-- investor stays a human action, never an auto-mutation (human-in-the-loop guardrail). The
+-- stage text is left as 'lost' on the archived row for provenance -- it is filtered out
+-- everywhere by deleted_at IS NULL.
+UPDATE opportunities SET deleted_at = datetime('now'), updated_at = datetime('now')
+    WHERE stage = 'lost' AND deleted_at IS NULL;

backend/migrations/0008_drop_retired_tables.down.sql

@@ -0,0 +1,42 @@
+-- 0008_drop_retired_tables.down.sql  (manual rollback only — never auto-applied)
+--
+-- Recreates the two dropped tables as EMPTY shells, matching the schema that existed
+-- immediately before 0008 (lp_profiles includes the deleted_at column that migration
+-- 0001 had added). Data is not restored — both tables were empty when dropped.
+CREATE TABLE IF NOT EXISTS lp_profiles (
+    id TEXT PRIMARY KEY,
+    contact_id TEXT NOT NULL UNIQUE REFERENCES contacts(id) ON DELETE CASCADE,
+    commitment_amount REAL DEFAULT 0,
+    funded_amount REAL DEFAULT 0,
+    commitment_date TEXT,
+    fund_name TEXT,
+    investor_type TEXT,
+    accredited INTEGER DEFAULT 0,
+    legal_docs_signed INTEGER DEFAULT 0,
+    signed_date TEXT,
+    wire_received INTEGER DEFAULT 0,
+    wire_date TEXT,
+    k1_sent INTEGER DEFAULT 0,
+    preferred_communication TEXT DEFAULT 'email',
+    notes TEXT,
+    created_at TEXT DEFAULT (datetime('now')),
+    updated_at TEXT DEFAULT (datetime('now')),
+    deleted_at TEXT
+);
+CREATE INDEX IF NOT EXISTS idx_lp_profiles_contact ON lp_profiles(contact_id);
+
+CREATE TABLE IF NOT EXISTS feature_requests (
+    id TEXT PRIMARY KEY,
+    title TEXT NOT NULL,
+    description TEXT,
+    page TEXT,
+    category TEXT DEFAULT 'general',
+    priority TEXT DEFAULT 'medium',
+    status TEXT DEFAULT 'new',
+    requested_by TEXT,
+    requested_by_user_id TEXT REFERENCES users(id),
+    created_at TEXT DEFAULT (datetime('now')),
+    updated_at TEXT DEFAULT (datetime('now'))
+);
+CREATE INDEX IF NOT EXISTS idx_feature_requests_status ON feature_requests(status);
+CREATE INDEX IF NOT EXISTS idx_feature_requests_created_at ON feature_requests(created_at);

backend/migrations/0008_drop_retired_tables.sql

@@ -0,0 +1,15 @@
+-- 0008_drop_retired_tables.sql  (v0.1.0:104)
+--
+-- ONE-OFF DESTRUCTIVE EXCEPTION to the never-hard-delete rule, explicitly approved.
+-- Both tables are EMPTY and fully removed from the application code:
+--   * lp_profiles    — the legacy single-fund LP model, retired v0.1.0:78; the
+--                      fundraising_* grid is the canonical commitment record now.
+--   * feature_requests — backed the in-app Feedback page, which was removed.
+--
+-- The never-hard-delete policy STILL STANDS for all real CRM and thesis data — this
+-- is a deliberate, documented exception for two empty, retired tables so they don't
+-- linger as dead schema. init_db() no longer creates either table, and migration
+-- 0001's lp_profiles ALTER was removed, so a fresh DB never creates them and this
+-- DROP is a harmless no-op there; on the live box it removes the existing empties.
+DROP TABLE IF EXISTS lp_profiles;
+DROP TABLE IF EXISTS feature_requests;

backend/nl_query/__init__.py

@@ -0,0 +1,9 @@
+"""nl_query — the safe, read-only natural-language query surface (W2).
+
+The LLM's job (added later) is only to map a question to a {intent, slots} pair; everything
+that touches the database lives here behind a strict validator and a fixed, hand-written,
+parameterized query catalog. See runner.py (the trust boundary) and intents.py (the catalog).
+"""
+from .runner import run_query, validate, catalog  # noqa: F401
+from .intents import INTENTS  # noqa: F401
+from .translate import translate, answer, build_system  # noqa: F401

backend/nl_query/intents.py

@@ -0,0 +1,440 @@
+"""NL-query intents — the curated, hand-written query catalog (W2, the safe core).
+
+Each intent is a FIXED, reviewed, parameterized SQL query with a small set of typed
+"slots" (the blanks a question fills in: a number of days, a name, a limit). There is NO
+generic SQL/AST compiler and NO dynamically-built identifiers: every table and column name
+is hardcoded in the query text, and every value the caller (or an LLM) supplies reaches
+SQLite only as a bound `?` parameter. That is the whole trust model — a malformed or
+hostile request can change a bound value, never the query structure. Adding a capability
+means adding a reviewed entry here, not widening a language.
+
+Soft-delete discipline (CLAUDE.md standing rule), per table:
+  - reminders / opportunities / communications carry `deleted_at` -> filter `deleted_at IS NULL`.
+  - emails have NO `deleted_at`; "live" means a non-tombstoned per-mailbox sighting exists
+    (`email_account_messages.deleted_at IS NULL`) — mirror the digest / query_email_activity.
+  - fundraising_investors/_contacts/_funds/_commitments are a HARD-REBUILT projection of the
+    grid blob with NO `deleted_at` column; the live/retired axis there is the `graveyard` flag.
+    Do NOT add `deleted_at IS NULL` to those tables — the column does not exist and the clause
+    would raise. Exclude `graveyard = 1` where the question means "live" investors.
+
+Each run_* returns {columns, rows, summary, truncated}. `summary` is a DETERMINISTIC local
+one-liner (never an LLM narrative) — results never leave the box to be summarized.
+"""
+import sqlite3
+from datetime import datetime, timedelta
+
+# Generous ceiling — the Matrix review room is two admins and the web app is internal, so
+# dumping the full book is acceptable (per Grant); this only guards against an unbounded
+# scan flooding a response. A list intent past this is reported truncated, never silently cut.
+MAX_ROWS = 500
+
+# 4-stage per-investor funnel in order, terminal at 'commitment' (mirrors server.PIPELINE_STAGES).
+# Kept here so the pipeline intents have a stable rank without importing the server module
+# (helpers take a conn; they never import server — house convention).
+_STAGE_ORDER = ['lead', 'engaged', 'diligence', 'commitment']
+_STAGE_RANK_SQL = (
+    "CASE stage WHEN 'lead' THEN 1 WHEN 'engaged' THEN 2 "
+    "WHEN 'diligence' THEN 3 WHEN 'commitment' THEN 4 ELSE 0 END")
+
+
+# ── helpers ────────────────────────────────────────────────────────────────────────────
+def _rows(cur):
+    """Materialize a cursor as a list of plain dicts, independent of the connection's
+    row_factory (works whether rows come back as tuples or sqlite3.Row)."""
+    cols = [c[0] for c in cur.description]
+    return [dict(zip(cols, r)) for r in cur.fetchall()]
+
+
+def like_contains(value):
+    """Build a safe LIKE pattern for a free-text contains match. Escapes the LIKE
+    wildcards so a user/LLM value of '%' or '_' is treated literally — paired with
+    `LIKE ? ESCAPE '\\'` in the SQL, this stops '%' from matching the entire table."""
+    v = value.replace("\\", "\\\\").replace("%", "\\%").replace("_", "\\_")
+    return f"%{v}%"
+
+
+def _last_activity_by_investor(conn):
+    """{fundraising_investors.id: latest activity ISO ts} across logged communications and
+    captured grid-linked emails — the per-investor recency signal behind the "gone quiet"
+    and "last contact" intents.
+
+    NB: this MIRRORS server.last_activity_by_investor() and its soft-delete joins (comms via
+    cm.deleted_at IS NULL; email via a live email_account_messages sighting). It is duplicated
+    rather than imported only to keep this module free of a server import (the main module runs
+    as __main__, so `import server` would re-execute it). Keep the two in sync; the soft-delete
+    test guards this copy. If a third caller appears, extract both to a shared module."""
+    out = {}
+
+    def _bump(inv_id, ts):
+        if inv_id and ts and (out.get(inv_id) is None or str(ts) > str(out[inv_id])):
+            out[inv_id] = ts
+
+    # Each leg is guarded: the comms/email tables can be absent on a minimal DB. This is a
+    # narrow, intentional tolerance for optional tables — NOT the broad error-swallowing the
+    # runner forbids (a failure in an intent's main query surfaces as query_failed).
+    try:
+        for r in conn.execute(
+            "SELECT fc.investor_id AS inv, MAX(cm.communication_date) AS last_ts "
+            "FROM communications cm JOIN fundraising_contacts fc ON fc.contact_id = cm.contact_id "
+            "WHERE cm.deleted_at IS NULL AND fc.contact_id IS NOT NULL GROUP BY fc.investor_id"):
+            _bump(r["inv"], r["last_ts"])
+    except sqlite3.OperationalError:
+        pass
+    try:
+        for r in conn.execute(
+            "SELECT eil.fundraising_investor_id AS inv, MAX(e.sent_at) AS last_ts "
+            "FROM email_investor_links eil JOIN emails e ON e.id = eil.email_id "
+            "WHERE eil.fundraising_investor_id IS NOT NULL AND EXISTS "
+            "(SELECT 1 FROM email_account_messages eam WHERE eam.email_id = e.id "
+            "AND eam.deleted_at IS NULL) GROUP BY eil.fundraising_investor_id"):
+            _bump(r["inv"], r["last_ts"])
+    except sqlite3.OperationalError:
+        pass
+    return out
+
+
+def _today():
+    return datetime.utcnow().date()
+
+
+def _days_since(ts):
+    """Whole days between an ISO date/datetime string and today (UTC). None if unparseable."""
+    if not ts:
+        return None
+    try:
+        d = datetime.fromisoformat(str(ts)[:10].replace("Z", "")).date()
+    except ValueError:
+        return None
+    return (_today() - d).days
+
+
+def _own_addresses(conn):
+    try:
+        return {(r[0] or "").lower().strip()
+                for r in conn.execute("SELECT email_address FROM email_accounts")} - {""}
+    except sqlite3.OperationalError:
+        return set()
+
+
+def _truncate(rows):
+    """Apply the global ceiling, returning (rows, truncated)."""
+    if len(rows) > MAX_ROWS:
+        return rows[:MAX_ROWS], True
+    return rows, False
+
+
+# ── investor intents ─────────────────────────────────────────────────────────────────────
+def run_investors_cold(conn, slots):
+    """Live investors not contacted in `days` days — never-contacted first, then oldest."""
+    days = slots["days"]
+    cutoff = (_today() - timedelta(days=days)).isoformat()
+    last = _last_activity_by_investor(conn)
+    invs = _rows(conn.execute(
+        "SELECT id, investor_name, lead, total_invested FROM fundraising_investors "
+        "WHERE graveyard = 0 ORDER BY investor_name"))
+    cold = []
+    for inv in invs:
+        ts = last.get(inv["id"])
+        if ts is None or str(ts)[:10] < cutoff:
+            cold.append({"investor_name": inv["investor_name"], "lead": inv["lead"],
+                         "total_invested": inv["total_invested"],
+                         "last_activity_at": ts, "days_since": _days_since(ts)})
+    # never-contacted (days_since None) first, then most-stale first
+    cold.sort(key=lambda r: (r["days_since"] is not None, -(r["days_since"] or 0)))
+    rows, trunc = _truncate(cold)
+    return {"columns": ["investor_name", "lead", "total_invested", "last_activity_at", "days_since"],
+            "rows": rows, "truncated": trunc,
+            "summary": f"{len(cold)} live investor(s) not contacted in {days}+ days."}
+
+
+def run_investor_lookup(conn, slots):
+    """One investor's profile: contacts (name/email/title/city), committed total, per-fund
+    commitments, lead. Name matched as a contains (an LLM/user may pass a partial)."""
+    pat = like_contains(slots["name"])
+    invs = _rows(conn.execute(
+        "SELECT id, investor_name, lead, lead_source, total_invested, follow_up, graveyard "
+        "FROM fundraising_investors WHERE investor_name LIKE ? ESCAPE '\\' "
+        "ORDER BY graveyard, investor_name LIMIT 25", (pat,)))
+    for inv in invs:
+        inv["contacts"] = _rows(conn.execute(
+            "SELECT full_name, email, title, city, state, country FROM fundraising_contacts "
+            "WHERE investor_id = ? ORDER BY sort_order, full_name", (inv["id"],)))
+        inv["commitments"] = _rows(conn.execute(
+            "SELECT f.fund_name, c.amount FROM fundraising_commitments c "
+            "JOIN fundraising_funds f ON f.id = c.fund_id WHERE c.investor_id = ? AND c.amount <> 0 "
+            "ORDER BY f.display_order", (inv["id"],)))
+        inv.pop("id", None)
+    return {"columns": ["investor_name", "lead", "lead_source", "total_invested",
+                        "follow_up", "graveyard", "contacts", "commitments"],
+            "rows": invs, "truncated": False,
+            "summary": f"{len(invs)} investor(s) matching \"{slots['name']}\"."}
+
+
+def run_investors_by_city(conn, slots):
+    """Investors with a contact located in `city` (contains match on the contact's city)."""
+    pat = like_contains(slots["city"])
+    rows = _rows(conn.execute(
+        "SELECT i.investor_name, c.full_name AS contact, c.city, c.state, c.country, i.lead "
+        "FROM fundraising_contacts c JOIN fundraising_investors i ON i.id = c.investor_id "
+        "WHERE i.graveyard = 0 AND c.city LIKE ? ESCAPE '\\' "
+        "ORDER BY i.investor_name, c.full_name LIMIT ?", (pat, MAX_ROWS + 1)))
+    rows, trunc = _truncate(rows)
+    return {"columns": ["investor_name", "contact", "city", "state", "country", "lead"],
+            "rows": rows, "truncated": trunc,
+            "summary": f"{len(rows)} investor contact(s) in \"{slots['city']}\"."}
+
+
+def run_investors_by_lead(conn, slots):
+    """Live investors owned by a given lead/team member (contains match on `lead`)."""
+    pat = like_contains(slots["lead"])
+    rows = _rows(conn.execute(
+        "SELECT investor_name, lead, total_invested, follow_up FROM fundraising_investors "
+        "WHERE graveyard = 0 AND lead LIKE ? ESCAPE '\\' "
+        "ORDER BY total_invested DESC, investor_name LIMIT ?", (pat, MAX_ROWS + 1)))
+    rows, trunc = _truncate(rows)
+    return {"columns": ["investor_name", "lead", "total_invested", "follow_up"],
+            "rows": rows, "truncated": trunc,
+            "summary": f"{len(rows)} live investor(s) led by \"{slots['lead']}\"."}
+
+
+def run_top_investors_committed(conn, slots):
+    """Top `limit` live investors by total committed capital across all funds."""
+    n = slots["limit"]
+    rows = _rows(conn.execute(
+        "SELECT investor_name, total_invested, lead FROM fundraising_investors "
+        "WHERE graveyard = 0 AND total_invested > 0 "
+        "ORDER BY total_invested DESC, investor_name LIMIT ?", (n,)))
+    return {"columns": ["investor_name", "total_invested", "lead"], "rows": rows,
+            "truncated": False, "summary": f"Top {len(rows)} investor(s) by committed capital."}
+
+
+def run_investors_follow_up(conn, slots):
+    """Investors we owe a follow-up to: those with an OPEN reminder, overdue first. Uses the
+    W1 reminders table (the richer follow-up layer) joined to the current grid name."""
+    today = _today().isoformat()
+    rows = _rows(conn.execute(
+        "SELECT COALESCE(i.investor_name, r.investor_name) AS investor_name, r.title, "
+        "r.due_date, r.status, r.assignee_id, "
+        "CASE WHEN r.due_date IS NOT NULL AND substr(r.due_date,1,10) < ? THEN 1 ELSE 0 END AS overdue "
+        "FROM reminders r LEFT JOIN fundraising_investors i ON i.id = r.investor_id "
+        "WHERE r.deleted_at IS NULL AND r.status = 'open' AND r.investor_id IS NOT NULL "
+        "ORDER BY (r.due_date IS NULL), r.due_date ASC LIMIT ?", (today, MAX_ROWS + 1)))
+    rows, trunc = _truncate(rows)
+    return {"columns": ["investor_name", "title", "due_date", "status", "overdue"],
+            "rows": rows, "truncated": trunc,
+            "summary": f"{len(rows)} investor(s) with an open follow-up reminder."}
+
+
+# ── pipeline intents ──────────────────────────────────────────────────────────────────────
+def run_pipeline_top(conn, slots):
+    """Top `limit` live pipeline opportunities by stage (furthest along first), with the
+    investor, owner, and most-recent activity."""
+    n = slots["limit"]
+    last = _last_activity_by_investor(conn)
+    rows = _rows(conn.execute(
+        "SELECT o.fundraising_investor_id AS inv_id, "
+        "COALESCE(i.investor_name, o.name) AS investor_name, o.stage, o.expected_amount, "
+        "o.probability, u.full_name AS owner FROM opportunities o "
+        "LEFT JOIN fundraising_investors i ON i.id = o.fundraising_investor_id "
+        "LEFT JOIN users u ON u.id = o.owner_id "
+        "WHERE o.deleted_at IS NULL "
+        f"ORDER BY {_STAGE_RANK_SQL} DESC, o.expected_amount DESC LIMIT ?", (n,)))
+    for r in rows:
+        r["last_activity_at"] = last.get(r.pop("inv_id"))
+    return {"columns": ["investor_name", "stage", "expected_amount", "probability", "owner",
+                        "last_activity_at"],
+            "rows": rows, "truncated": False,
+            "summary": f"Top {len(rows)} live pipeline opportunit(ies) by stage."}
+
+
+def run_pipeline_totals(conn, slots):
+    """Total pipeline dollars and the split across each stage."""
+    rows = _rows(conn.execute(
+        "SELECT stage, COUNT(*) AS count, COALESCE(SUM(expected_amount),0) AS expected_total, "
+        "COALESCE(SUM(commitment_amount),0) AS committed_total FROM opportunities "
+        f"WHERE deleted_at IS NULL GROUP BY stage ORDER BY {_STAGE_RANK_SQL}"))
+    total = sum(r["expected_total"] for r in rows)
+    count = sum(r["count"] for r in rows)
+    return {"columns": ["stage", "count", "expected_total", "committed_total"],
+            "rows": rows, "truncated": False,
+            "summary": f"${total:,.0f} expected across {count} live opportunit(ies) in "
+                       f"{len(rows)} stage(s)."}
+
+
+# ── email / communication intents ─────────────────────────────────────────────────────────
+def run_recent_emails(conn, slots):
+    """The most recent `limit` matched investor emails, optionally one direction.
+    Matched-only + soft-delete-correct (a live email_account_messages sighting), mirroring
+    the Communications panel's query_email_activity."""
+    n, direction = slots["limit"], slots["direction"]
+    where = ["EXISTS (SELECT 1 FROM email_account_messages eam WHERE eam.email_id = e.id "
+             "AND eam.deleted_at IS NULL)",
+             "EXISTS (SELECT 1 FROM email_investor_links l WHERE l.email_id = e.id)"]
+    params = []
+    own = _own_addresses(conn)
+    if direction in ("inbound", "outbound") and own:
+        op = "IN" if direction == "outbound" else "NOT IN"
+        where.append(f"LOWER(e.from_email) {op} ({','.join('?' for _ in own)})")
+        params.extend(sorted(own))
+    sql = ("SELECT e.subject, e.from_name, e.from_email, e.sent_at, "
+           "(SELECT fi.investor_name FROM email_investor_links l "
+           " JOIN fundraising_investors fi ON fi.id = l.fundraising_investor_id "
+           " WHERE l.email_id = e.id AND l.fundraising_investor_id IS NOT NULL LIMIT 1) AS investor "
+           "FROM emails e WHERE " + " AND ".join(where) + " ORDER BY e.sent_at DESC LIMIT ?")
+    rows = _rows(conn.execute(sql, params + [n]))
+    label = {"inbound": "received", "outbound": "sent"}.get(direction, "")
+    return {"columns": ["sent_at", "subject", "from_name", "from_email", "investor"],
+            "rows": rows, "truncated": False,
+            "summary": f"{len(rows)} most-recent {label + ' ' if label else ''}investor email(s)."}
+
+
+def run_investor_last_contact(conn, slots):
+    """When we last had any activity with investor X (matched by name)."""
+    pat = like_contains(slots["name"])
+    last = _last_activity_by_investor(conn)
+    invs = _rows(conn.execute(
+        "SELECT id, investor_name FROM fundraising_investors "
+        "WHERE investor_name LIKE ? ESCAPE '\\' ORDER BY graveyard, investor_name LIMIT 25", (pat,)))
+    rows = []
+    for inv in invs:
+        ts = last.get(inv["id"])
+        rows.append({"investor_name": inv["investor_name"], "last_activity_at": ts,
+                     "days_since": _days_since(ts)})
+    return {"columns": ["investor_name", "last_activity_at", "days_since"], "rows": rows,
+            "truncated": False, "summary": f"Last contact for {len(rows)} investor(s) "
+                                           f"matching \"{slots['name']}\"."}
+
+
+def run_comms_by_user(conn, slots):
+    """The most recent `limit` outbound **investor** emails sent by a given user (matched by
+    username or full name). MATCHED-ONLY: restricted to investor-linked email (an
+    email_investor_links row exists), mirroring query_email_activity / recent_emails — NOT the
+    user's entire sent corpus (internal/vendor/personal mail is captured but never surfaced
+    here). Soft-delete-correct (live sighting, is_sent)."""
+    n, pat = slots["limit"], like_contains(slots["user"])
+    rows = _rows(conn.execute(
+        "SELECT e.subject, e.sent_at, u.full_name AS sender, "
+        "(SELECT fi.investor_name FROM email_investor_links l "
+        " JOIN fundraising_investors fi ON fi.id = l.fundraising_investor_id "
+        " WHERE l.email_id = e.id AND l.fundraising_investor_id IS NOT NULL LIMIT 1) AS investor "
+        "FROM emails e JOIN email_account_messages eam ON eam.email_id = e.id "
+        "AND eam.deleted_at IS NULL AND eam.is_sent = 1 "
+        "JOIN email_accounts ea ON ea.id = eam.account_id JOIN users u ON u.id = ea.user_id "
+        "WHERE (u.username LIKE ? ESCAPE '\\' OR u.full_name LIKE ? ESCAPE '\\') "
+        "AND EXISTS (SELECT 1 FROM email_investor_links l2 WHERE l2.email_id = e.id) "
+        "ORDER BY e.sent_at DESC LIMIT ?", (pat, pat, n)))
+    return {"columns": ["sent_at", "subject", "sender", "investor"], "rows": rows,
+            "truncated": False,
+            "summary": f"{len(rows)} recent email(s) sent by \"{slots['user']}\"."}
+
+
+def run_email_counts_by_user(conn, slots):
+    """Per-user counts of outbound **investor** emails over this week / month / year-to-date.
+    MATCHED-ONLY: counts only investor-linked email (an email_investor_links row exists),
+    mirroring query_email_activity / recent_emails — not the user's entire sent corpus.
+    Windows are calendar-based: week = since Monday, month = since the 1st, ytd = since Jan 1."""
+    today = _today()
+    wk = (today - timedelta(days=today.weekday())).isoformat()
+    mo = today.replace(day=1).isoformat()
+    yr = today.replace(month=1, day=1).isoformat()
+    where = ("WHERE eam.deleted_at IS NULL AND eam.is_sent = 1 "
+             "AND EXISTS (SELECT 1 FROM email_investor_links l WHERE l.email_id = e.id)")
+    params = [wk, mo, yr]
+    if slots.get("user"):
+        pat = like_contains(slots["user"])
+        where += " AND (u.username LIKE ? ESCAPE '\\' OR u.full_name LIKE ? ESCAPE '\\')"
+        params.extend([pat, pat])
+    rows = _rows(conn.execute(
+        "SELECT u.full_name AS user, u.username, "
+        "SUM(CASE WHEN substr(e.sent_at,1,10) >= ? THEN 1 ELSE 0 END) AS this_week, "
+        "SUM(CASE WHEN substr(e.sent_at,1,10) >= ? THEN 1 ELSE 0 END) AS this_month, "
+        "SUM(CASE WHEN substr(e.sent_at,1,10) >= ? THEN 1 ELSE 0 END) AS ytd "
+        "FROM users u JOIN email_accounts ea ON ea.user_id = u.id "
+        "JOIN email_account_messages eam ON eam.account_id = ea.id "
+        "JOIN emails e ON e.id = eam.email_id " + where +
+        " GROUP BY u.id HAVING ytd > 0 ORDER BY ytd DESC", params))
+    return {"columns": ["user", "this_week", "this_month", "ytd"], "rows": rows,
+            "truncated": False, "summary": f"Outbound email counts for {len(rows)} user(s)."}
+
+
+# ── registry ──────────────────────────────────────────────────────────────────────────────
+# key -> {summary, slots, run, example}. `slots` is consumed by the runner's validator and
+# (later) surfaced to the local-model translator + the UI as the single source of truth for
+# what is queryable. SlotSpec: {type: int|enum|text, ...constraints}.
+INTENTS = {
+    "investors_cold": {
+        "summary": "Investors we haven't contacted in a while (default 90 days).",
+        "slots": {"days": {"type": "int", "default": 90, "min": 1, "max": 3650}},
+        "example": "Which investors haven't we reached out to in the last 3 months?",
+        "run": run_investors_cold,
+    },
+    "investor_lookup": {
+        "summary": "One investor's contacts, email, committed total and per-fund breakdown.",
+        "slots": {"name": {"type": "text", "required": True, "maxlen": 120}},
+        "example": "What is Acme Capital's email and how much have they committed across funds?",
+        "run": run_investor_lookup,
+    },
+    "investors_by_city": {
+        "summary": "Investors with a contact located in a given city.",
+        "slots": {"city": {"type": "text", "required": True, "maxlen": 80}},
+        "example": "Who are all the investors located in Austin?",
+        "run": run_investors_by_city,
+    },
+    "investors_by_lead": {
+        "summary": "Investors owned by a given lead / team member.",
+        "slots": {"lead": {"type": "text", "required": True, "maxlen": 80}},
+        "example": "Show me the investors led by Jonathan.",
+        "run": run_investors_by_lead,
+    },
+    "top_investors_committed": {
+        "summary": "Top investors by total committed capital.",
+        "slots": {"limit": {"type": "int", "default": 10, "min": 1, "max": MAX_ROWS}},
+        "example": "List our top 10 investors by committed capital.",
+        "run": run_top_investors_committed,
+    },
+    "investors_follow_up": {
+        "summary": "Investors we owe a follow-up to (have an open reminder), overdue first.",
+        "slots": {},
+        "example": "Which investors do we owe follow-ups to?",
+        "run": run_investors_follow_up,
+    },
+    "pipeline_top": {
+        "summary": "Top pipeline opportunities by stage, with investor, owner and last activity.",
+        "slots": {"limit": {"type": "int", "default": 10, "min": 1, "max": MAX_ROWS}},
+        "example": "List our top 10 pipeline investors by stage and last conversation.",
+        "run": run_pipeline_top,
+    },
+    "pipeline_totals": {
+        "summary": "Total pipeline dollars and the split across each stage.",
+        "slots": {},
+        "example": "What is our total pipeline in dollars, split by stage?",
+        "run": run_pipeline_totals,
+    },
+    "recent_emails": {
+        "summary": "The most recent investor emails (optionally inbound or outbound only).",
+        "slots": {"limit": {"type": "int", "default": 10, "min": 1, "max": 100},
+                  "direction": {"type": "enum", "choices": ["any", "inbound", "outbound"],
+                                "default": "any"}},
+        "example": "What were the last 10 investor emails we sent, and who to?",
+        "run": run_recent_emails,
+    },
+    "investor_last_contact": {
+        "summary": "When we last had any activity with a given investor.",
+        "slots": {"name": {"type": "text", "required": True, "maxlen": 120}},
+        "example": "When did we last reach out to Acme Capital?",
+        "run": run_investor_last_contact,
+    },
+    "comms_by_user": {
+        "summary": "Recent investor emails sent by a given team member.",
+        "slots": {"user": {"type": "text", "required": True, "maxlen": 80},
+                  "limit": {"type": "int", "default": 10, "min": 1, "max": 100}},
+        "example": "What were the last investor emails sent by Grant?",
+        "run": run_comms_by_user,
+    },
+    "email_counts_by_user": {
+        "summary": "How many investor emails each user sent this week / month / year-to-date.",
+        "slots": {"user": {"type": "text", "required": False, "maxlen": 80}},
+        "example": "How many emails has Grant sent this week, this month, and year to date?",
+        "run": run_email_counts_by_user,
+    },
+}

backend/nl_query/runner.py

@@ -0,0 +1,127 @@
+"""NL-query runner — validate a {intent, slots} request, run the curated query, return rows.
+
+This is the trust boundary. Whatever produced the request (a local model in W2, the web UI,
+or a test) is untrusted: the runner accepts ONLY a known intent key and slot VALUES, coerces
+each value to its declared type, and rejects anything off-spec — it never lets a caller name
+a table/column, write SQL, or choose an operator. The intents do the rest with fixed,
+parameterized SQL (see intents.py). All failure modes return a structured error dict; the
+runner never raises to the caller (a bad `limit=abc` must not crash the request thread).
+"""
+import sqlite3
+
+from .intents import INTENTS
+
+
+def _coerce_slot(name, spec, raw):
+    """Coerce/validate one slot value against its spec. Returns (value, error). Exactly one
+    of the two is meaningful: error is None on success, else a human-readable string."""
+    t = spec["type"]
+    provided = raw is not None and not (isinstance(raw, str) and raw.strip() == "")
+
+    if not provided:
+        if "default" in spec:
+            return spec["default"], None
+        if spec.get("required"):
+            return None, f"slot '{name}' is required"
+        return None, None  # optional, absent
+
+    if t == "int":
+        try:
+            v = int(raw)
+        except (TypeError, ValueError):
+            return None, f"slot '{name}' must be an integer (got {raw!r})"
+        if "min" in spec:
+            v = max(spec["min"], v)
+        if "max" in spec:
+            v = min(spec["max"], v)
+        return v, None
+
+    if t == "enum":
+        v = str(raw).strip().lower()
+        if v not in spec["choices"]:
+            if "default" in spec:
+                return spec["default"], None
+            return None, f"slot '{name}' must be one of {spec['choices']} (got {raw!r})"
+        return v, None
+
+    if t == "text":
+        v = str(raw).strip()
+        maxlen = spec.get("maxlen", 200)
+        if len(v) > maxlen:
+            v = v[:maxlen]
+        return v, None
+
+    return None, f"slot '{name}' has unknown type {t!r}"  # registry bug, fail visibly
+
+
+def validate(intent_key, raw_slots):
+    """Validate an intent + raw slots WITHOUT running. Returns (clean_slots, error_dict).
+    Useful to the translator/UI for a dry-run check. error_dict is None on success."""
+    if intent_key not in INTENTS:
+        return None, {"error": "unknown_intent", "intent": intent_key,
+                      "detail": f"unknown intent; known: {sorted(INTENTS)}"}
+    spec = INTENTS[intent_key]["slots"]
+    raw_slots = raw_slots or {}
+    # Reject unexpected slot keys rather than ignore them — a request shaped wrong is a
+    # misunderstanding worth surfacing, not silently dropping.
+    unexpected = [k for k in raw_slots if k not in spec]
+    if unexpected:
+        return None, {"error": "bad_slot", "intent": intent_key,
+                      "detail": f"unexpected slot(s): {unexpected}; allowed: {sorted(spec)}"}
+    clean = {}
+    for name, sspec in spec.items():
+        v, err = _coerce_slot(name, sspec, raw_slots.get(name))
+        if err:
+            return None, {"error": "bad_slot", "intent": intent_key, "detail": err}
+        if v is not None or "default" in sspec:
+            clean[name] = v
+    return clean, None
+
+
+def run_query(conn, intent_key, raw_slots=None, *, audit_fn=None, actor=None, source="api"):
+    """Validate and execute a curated NL query. Always returns a dict — either a result
+    {intent, slots, columns, rows, row_count, truncated, summary} or an error
+    {error, intent, detail}. Records an audit row via audit_fn (if given) so a query made
+    through a leaked/automated credential is detectable.
+
+    audit_fn signature: audit_fn({actor, source, intent, slots, row_count, error}).
+    """
+    clean, err = validate(intent_key, raw_slots)
+    if err:
+        if audit_fn:
+            try:
+                audit_fn({"actor": actor, "source": source, "intent": intent_key,
+                          "slots": raw_slots, "row_count": 0, "error": err["error"]})
+            except Exception:
+                pass
+        return err
+
+    try:
+        result = INTENTS[intent_key]["run"](conn, clean)
+    except sqlite3.Error as exc:
+        # Surface a query failure (e.g. a missing optional table) as a visible error — never
+        # swallow it and hand back an empty result that reads as an authoritative "none".
+        if audit_fn:
+            try:
+                audit_fn({"actor": actor, "source": source, "intent": intent_key,
+                          "slots": clean, "row_count": 0, "error": "query_failed"})
+            except Exception:
+                pass
+        return {"error": "query_failed", "intent": intent_key, "detail": str(exc)}
+
+    out = {"intent": intent_key, "slots": clean, "row_count": len(result.get("rows", [])),
+           **result}
+    if audit_fn:
+        try:
+            audit_fn({"actor": actor, "source": source, "intent": intent_key,
+                      "slots": clean, "row_count": out["row_count"], "error": None})
+        except Exception:
+            pass
+    return out
+
+
+def catalog():
+    """The queryable surface as data: every intent's key, summary, slot specs and example.
+    Single source of truth for the W2 translator prompt and any UI hint list."""
+    return [{"intent": k, "summary": v["summary"], "slots": v["slots"],
+             "example": v.get("example", "")} for k, v in INTENTS.items()]

backend/nl_query/test_nl_query.py

@@ -0,0 +1,237 @@
+#!/usr/bin/env python3
+"""Tests for the W2 safe NL-query runner (the model-free core).
+
+Boots the REAL schema (server.init_db against a temp DB — exact columns + all migrations),
+inserts synthetic fundraising/email/reminder/pipeline data, and exercises every intent plus
+the trust-boundary behaviour:
+  - each intent returns the right rows over the real schema;
+  - SOFT-DELETE is respected on both recency legs (a tombstoned communication and a tombstoned
+    email sighting never count), on reminders, and on opportunities; graveyard investors are
+    excluded from "live" intents;
+  - the validator rejects bad/unknown/unexpected slots WITHOUT crashing (the `?limit=abc` class);
+  - LIKE wildcards in a free-text slot are escaped (a city of "%" does NOT return everything);
+  - limits clamp to their caps; the audit hook fires with the intent + row count.
+Synthetic data only — no real LP substance, no network, no model.
+
+Run: cd backend && python3 nl_query/test_nl_query.py
+"""
+import os
+import sys
+import tempfile
+from datetime import datetime, timedelta
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+os.environ["CRM_GMAIL_INTEGRATION_ENABLED"] = "1"
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))  # backend/
+import server  # noqa: E402
+import nl_query  # noqa: E402
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+def _ago(days):
+    return (datetime.utcnow() - timedelta(days=days)).isoformat() + "Z"
+
+
+TODAY = datetime.utcnow().date()
+
+
+def seed(conn):
+    c = conn.execute
+    # users + mailboxes
+    c("INSERT INTO users (id, username, email, password_hash, full_name, role) VALUES "
+      "('u_grant','grant','grant@ten31.xyz','x','Grant Smith','admin'),"
+      "('u_jon','jonathan','jon@ten31.xyz','x','Jonathan Lee','member')")
+    c("INSERT INTO email_accounts (id, user_id, email_address, auth_method) VALUES "
+      "('a_grant','u_grant','grant@ten31.xyz','dwd'),"
+      "('a_jon','u_jon','jon@ten31.xyz','dwd')")
+    # funds
+    c("INSERT INTO fundraising_funds (id, column_id, fund_name, display_order) VALUES "
+      "('f1','c_f1','Fund I',1),('f2','c_f2','Fund II',2)")
+
+    # investors (graveyard flag is the live/retired axis; no deleted_at on this table)
+    def inv(iid, name, lead, total, grave=0):
+        c("INSERT INTO fundraising_investors (id, investor_name, lead, graveyard, "
+          "source_row_id, total_invested) VALUES (?,?,?,?,?,?)",
+          (iid, name, lead, grave, iid, total))
+    inv("i_acme", "Acme Capital", "Jonathan Lee", 5_000_000)
+    inv("i_beta", "Beta Partners", "Grant Smith", 2_000_000)
+    inv("i_cold", "Cold Co", "Grant Smith", 0)                  # never contacted
+    inv("i_delta", "Delta LP", "Grant Smith", 1_000_000)        # only a (comms) signal
+    inv("i_ghost", "Graveyard Ghost", "Grant Smith", 9_999_999, grave=1)
+
+    # contacts (grid pills) + classic contact rows for the comms leg
+    c("INSERT INTO fundraising_contacts (id, investor_id, full_name, email, title, city, "
+      "contact_id, sort_order) VALUES "
+      "('fc_a','i_acme','Alice Acme','alice@acme.com','GP','Austin','cc_alice',0),"
+      "('fc_b','i_beta','Bob Beta','bob@beta.com','LP','Denver',NULL,0),"
+      "('fc_d','i_delta','Dana Delta','dana@delta.com','CFO','Miami','cc_dana',0)")
+    c("INSERT INTO contacts (id, first_name, last_name, email) VALUES "
+      "('cc_alice','Alice','Acme','alice@acme.com'),"
+      "('cc_dana','Dana','Delta','dana@delta.com')")
+
+    # commitments — Acme across two funds (3M + 2M = 5M); Beta one fund
+    c("INSERT INTO fundraising_commitments (id, investor_id, fund_id, amount) VALUES "
+      "('cm1','i_acme','f1',3_000_000),('cm2','i_acme','f2',2_000_000),"
+      "('cm3','i_beta','f1',2_000_000)")
+
+    # emails: matched + a per-mailbox sighting. is_sent + from_email decide direction.
+    def email(eid, frm, frm_name, days, inv_id, account, is_sent, deleted=False):
+        c("INSERT INTO emails (id, rfc_message_id, from_email, from_name, sent_at, subject, "
+          "is_matched, match_status) VALUES (?,?,?,?,?,?,1,'matched')",
+          (eid, "rfc_" + eid, frm, frm_name, _ago(days), "Re: " + eid))
+        c("INSERT INTO email_account_messages (id, email_id, account_id, gmail_message_id, "
+          "gmail_thread_id, is_sent, deleted_at) VALUES (?,?,?,?,?,?,?)",
+          ("eam_" + eid, eid, account, "g_" + eid, "t_" + eid, is_sent,
+           _ago(days) if deleted else None))
+        c("INSERT INTO email_investor_links (id, email_id, fundraising_investor_id, "
+          "matched_address, match_kind) VALUES (?,?,?,?, 'exact_email')",
+          ("eil_" + eid, eid, inv_id, frm))
+    email("ea_recent", "grant@ten31.xyz", "Grant Smith", 0, "i_acme", "a_grant", 1)   # Acme: today
+    email("eb_old", "grant@ten31.xyz", "Grant Smith", 40, "i_beta", "a_grant", 1)     # Beta: 40d
+    email("edel", "grant@ten31.xyz", "Grant Smith", 0, "i_beta", "a_grant", 1, deleted=True)  # tombstoned
+    email("ej", "jon@ten31.xyz", "Jonathan Lee", 0, "i_acme", "a_jon", 1)             # jonathan today
+    email("ein", "alice@acme.com", "Alice Acme", 3, "i_acme", "a_grant", 0)           # inbound 3d
+    # an UNMATCHED sent email by Grant (NO email_investor_links row) — captured, but not to a
+    # known investor. The investor-email intents are matched-only, so it must be EXCLUDED from
+    # comms_by_user / email_counts_by_user; without the matched-only filter it would inflate both.
+    c("INSERT INTO emails (id, rfc_message_id, from_email, from_name, sent_at, subject, "
+      "is_matched, match_status) VALUES ('eunm','rfc_eunm','grant@ten31.xyz','Grant Smith',?,"
+      "'Internal: team lunch',0,'unmatched')", (_ago(0),))
+    c("INSERT INTO email_account_messages (id, email_id, account_id, gmail_message_id, "
+      "gmail_thread_id, is_sent, deleted_at) VALUES "
+      "('eam_eunm','eunm','a_grant','g_eunm','t_eunm',1,NULL)")
+
+    # communications (the other recency leg) — Delta has ONLY comms: one live (5d), one tombstoned
+    # (today). If the soft-delete filter broke, Delta would read as contacted today.
+    c("INSERT INTO communications (id, contact_id, type, communication_date, created_by) VALUES "
+      "('cmm_live','cc_dana','email',?,'u_grant')", (_ago(5),))
+    c("INSERT INTO communications (id, contact_id, type, communication_date, created_by, deleted_at) "
+      "VALUES ('cmm_del','cc_dana','email',?,'u_grant',?)", (_ago(0), _ago(0)))
+
+    # reminders — open(overdue) / open(future) / done / deleted / standalone
+    def rem(rid, inv_id, title, due, status="open", deleted=False):
+        c("INSERT INTO reminders (id, investor_id, investor_name, title, due_date, status, "
+          "deleted_at) VALUES (?,?,?,?,?,?,?)",
+          (rid, inv_id, title, title, due, status, _ago(0) if deleted else None))
+    rem("r_over", "i_beta", "Send deck", (TODAY - timedelta(days=1)).isoformat())       # overdue
+    rem("r_future", "i_acme", "Quarterly check-in", (TODAY + timedelta(days=10)).isoformat())
+    rem("r_done", "i_acme", "Old task", (TODAY - timedelta(days=2)).isoformat(), status="done")
+    rem("r_del", "i_acme", "Tombstoned", (TODAY - timedelta(days=2)).isoformat(), deleted=True)
+    rem("r_standalone", None, "Team chore", (TODAY - timedelta(days=1)).isoformat())
+
+    # opportunities — commitment / engaged (live) / two archived (the new model has no 'lost'
+    # stage: a dead deal is soft-deleted, so both excluded cases ride deleted_at)
+    def opp(oid, inv_id, contact, stage, expected, owner, deleted=False):
+        c("INSERT INTO opportunities (id, name, contact_id, stage, expected_amount, owner_id, "
+          "fundraising_investor_id, deleted_at) VALUES (?,?,?,?,?,?,?,?)",
+          (oid, oid, contact, stage, expected, owner, inv_id, _ago(0) if deleted else None))
+    # opp contact_id must reference a real contacts row (FK on); reuse the two we made
+    opp("o_acme", "i_acme", "cc_alice", "commitment", 4_000_000, "u_jon")
+    opp("o_beta", "i_beta", "cc_dana", "engaged", 1_000_000, "u_grant")
+    opp("o_lost", "i_acme", "cc_alice", "diligence", 9_000_000, "u_jon", deleted=True)
+    opp("o_del", "i_beta", "cc_dana", "diligence", 7_000_000, "u_grant", deleted=True)
+    conn.commit()
+
+
+def names(res):
+    return [r["investor_name"] for r in res["rows"]]
+
+
+def main():
+    server.init_db()
+    conn = server.get_db()
+    seed(conn)
+    run = lambda *a, **k: nl_query.run_query(conn, *a, **k)
+
+    print("investors_cold")
+    r = run("investors_cold", {"days": 30})
+    check(names(r) == ["Cold Co", "Beta Partners"], f"cold(30) never-first then stale: {names(r)}")
+    check(run("investors_cold", {"days": 90})["row_count"] == 1, "cold(90): only never-contacted")
+    check("Graveyard Ghost" not in names(run("investors_cold", {"days": 3650})),
+          "cold excludes graveyard investors")
+    check("Delta LP" in names(run("investors_cold", {"days": 3})), "cold(3) sees Delta (comms 5d)")
+    check("Delta LP" not in names(run("investors_cold", {"days": 7})),
+          "cold(7): Delta's tombstoned comm (today) did NOT count")
+
+    print("investor_lookup")
+    r = run("investor_lookup", {"name": "acme"})
+    check(r["row_count"] == 1 and r["rows"][0]["total_invested"] == 5_000_000, "lookup total committed")
+    check({c["fund_name"] for c in r["rows"][0]["commitments"]} == {"Fund I", "Fund II"},
+          "lookup per-fund breakdown")
+    check(r["rows"][0]["contacts"][0]["email"] == "alice@acme.com", "lookup surfaces contact email")
+
+    print("investors_by_city / by_lead / top / follow_up")
+    check(names(run("investors_by_city", {"city": "Austin"})) == ["Acme Capital"], "by_city")
+    check(set(names(run("investors_by_lead", {"lead": "Grant"}))) == {"Beta Partners", "Cold Co", "Delta LP"},
+          "by_lead excludes graveyard + other leads")
+    check(names(run("top_investors_committed", {"limit": 2})) == ["Acme Capital", "Beta Partners"],
+          "top by committed (graveyard + zero excluded)")
+    r = run("investors_follow_up")
+    check(names(r) == ["Beta Partners", "Acme Capital"], f"follow_up overdue-first, open-only: {names(r)}")
+    check(r["rows"][0]["overdue"] == 1 and r["rows"][1]["overdue"] == 0, "follow_up overdue flag")
+
+    print("pipeline")
+    r = run("pipeline_totals")
+    stages = {row["stage"]: row for row in r["rows"]}
+    check(set(stages) == {"commitment", "engaged"}, f"pipeline_totals excludes archived/deleted: {set(stages)}")
+    check(stages["commitment"]["expected_total"] == 4_000_000, "pipeline_totals stage sum")
+    r = run("pipeline_top", {"limit": 10})
+    check(names(r) == ["Acme Capital", "Beta Partners"], "pipeline_top furthest-stage first")
+    check(r["rows"][0]["last_activity_at"] is not None, "pipeline_top enriches last activity")
+
+    print("emails")
+    check(run("recent_emails", {"direction": "outbound"})["row_count"] == 3,
+          "recent_emails(outbound): 3 live (tombstoned sighting excluded)")
+    check(run("recent_emails", {"direction": "inbound"})["row_count"] == 1, "recent_emails(inbound)")
+    check(run("recent_emails")["row_count"] == 4, "recent_emails(any): 4 live")
+    r = run("investor_last_contact", {"name": "beta"})
+    check(r["rows"][0]["days_since"] >= 39, "investor_last_contact days_since")
+    check(run("comms_by_user", {"user": "Grant"})["row_count"] == 2,
+          "comms_by_user: grant's 2 live MATCHED outbound (tombstoned + unmatched excluded)")
+    r = run("email_counts_by_user", {"user": "grant"})
+    check(r["rows"][0]["this_week"] == 1,
+          "email_counts this_week = 1 live matched (tombstoned + unmatched excluded)")
+    check(r["rows"][0]["ytd"] >= 1, "email_counts ytd")
+
+    print("trust boundary")
+    check(run("investors_cold", {"days": "abc"})["error"] == "bad_slot", "bad int slot -> bad_slot, no crash")
+    check(run("nope")["error"] == "unknown_intent", "unknown intent rejected")
+    check(run("pipeline_totals", {"foo": 1})["error"] == "bad_slot", "unexpected slot rejected")
+    check(run("investor_lookup", {})["error"] == "bad_slot", "missing required slot rejected")
+    check(run("investors_by_city", {"city": "%"})["row_count"] == 0,
+          "LIKE wildcard escaped — '%' does not match every row")
+    check(run("investors_cold", {"days": 0})["slots"]["days"] == 1, "int slot clamps to min")
+    check(run("top_investors_committed", {"limit": 99999})["slots"]["limit"] == nl_query.INTENTS
+          ["top_investors_committed"]["slots"]["limit"]["max"], "int slot clamps to max")
+
+    print("audit hook + catalog")
+    seen = []
+    run("pipeline_totals", audit_fn=seen.append, actor="tester", source="test")
+    check(len(seen) == 1 and seen[0]["intent"] == "pipeline_totals" and seen[0]["error"] is None
+          and seen[0]["actor"] == "tester", "audit hook fires with intent/actor/no-error")
+    run("nope", audit_fn=seen.append)
+    check(seen[-1]["error"] == "unknown_intent", "audit hook fires on rejection too")
+    check(len(nl_query.catalog()) == len(nl_query.INTENTS), "catalog covers every intent")
+
+    conn.close()
+    print()
+    if FAILS:
+        print(f"{len(FAILS)} FAILED")
+        for f in FAILS:
+            print("  - " + f)
+        sys.exit(1)
+    print("ALL PASS")
+
+
+if __name__ == "__main__":
+    main()

backend/nl_query/test_nl_query_endpoint.py

@@ -0,0 +1,139 @@
+#!/usr/bin/env python3
+"""Endpoint tests for the W2 NL-query HTTP surface (POST /api/query/nl, GET /api/query/catalog).
+
+Boots the REAL server against a temp DB and exercises the wiring end-to-end: auth gating
+(bot/admin only), the direct {intent, slots} mode, the soft-error shape, and the status
+mapping. The local model is forced UNAVAILABLE by pointing SPARK_CONTROL_URL at a dead local
+port, so the {question} path exercises the 503 path deterministically without any Spark.
+Synthetic data only.
+
+Run: cd backend && python3 nl_query/test_nl_query_endpoint.py
+"""
+import http.client
+import json
+import os
+import sqlite3
+import sys
+import tempfile
+import threading
+from http.server import ThreadingHTTPServer
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+os.environ["CRM_GMAIL_INTEGRATION_ENABLED"] = "1"
+# Dead port -> the local-model leg fails fast, so the {question} path returns 503 deterministically
+# (set before server/config import; load_env uses setdefault so this wins over any repo .env).
+os.environ["SPARK_CONTROL_URL"] = "http://127.0.0.1:1"
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))  # backend/
+import server  # noqa: E402
+import nl_query  # noqa: E402
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Quiet(server.CRMHandler):
+    def log_message(self, *a):
+        pass
+
+
+def _req(port, method, path, token=None, body=None):
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    headers = {}
+    if token:
+        headers["Authorization"] = "Bearer " + token
+    payload = json.dumps(body) if body is not None else None
+    if payload is not None:
+        headers["Content-Type"] = "application/json"
+    conn.request(method, path, body=payload, headers=headers)
+    resp = conn.getresponse()
+    raw = resp.read().decode("utf-8", "replace")
+    conn.close()
+    data = json.loads(raw) if raw else None
+    return resp.status, data
+
+
+def _data(d):
+    return (d or {}).get("data") or {}
+
+
+def main():
+    server.init_db()
+    db = sqlite3.connect(os.environ["CRM_DB_PATH"])
+    db.execute("INSERT INTO users (id,username,email,password_hash,full_name,role,is_active) VALUES "
+               "('u_admin','grant','g@t.x','x','Grant','admin',1),"
+               "('u_mem','mem','m@t.x','x','Mem','member',1)")
+    db.execute("INSERT INTO fundraising_investors (id,investor_name,lead,graveyard,source_row_id,"
+               "total_invested) VALUES ('a','Acme Capital','Jon',0,'a',5000000),"
+               "('b','Beta Partners','Grant',0,'b',2000000),('g','Ghost','Grant',1,'g',9000000)")
+    db.commit()
+    db.close()
+    admin = server.create_token("u_admin", "grant", "admin")
+    member = server.create_token("u_mem", "mem", "member")
+
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), _Quiet)
+    port = httpd.server_address[1]
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        print("direct {intent, slots} mode")
+        st, d = _req(port, "POST", "/api/query/nl", admin,
+                     {"intent": "top_investors_committed", "slots": {"limit": 2}})
+        rows = _data(d).get("rows", [])
+        check(st == 200 and [r["investor_name"] for r in rows] == ["Acme Capital", "Beta Partners"],
+              f"admin direct query -> 200 + rows (got {st})")
+        check(_data(d).get("intent") == "top_investors_committed", "response echoes interpreted intent")
+
+        print("soft errors + validation")
+        st, d = _req(port, "POST", "/api/query/nl", admin, {"intent": "made_up"})
+        check(st == 200 and _data(d).get("error") == "unknown_intent",
+              f"bad intent -> 200 with data.error=unknown_intent (got {st}, {_data(d).get('error')})")
+        st, d = _req(port, "POST", "/api/query/nl", admin, {})
+        check(st == 400, f"neither question nor intent -> 400 (got {st})")
+
+        print("auth gating")
+        st, _ = _req(port, "POST", "/api/query/nl", member,
+                     {"intent": "top_investors_committed"})
+        check(st == 403, f"member -> 403 (got {st})")
+        st, _ = _req(port, "POST", "/api/query/nl", None, {"intent": "top_investors_committed"})
+        check(st == 401, f"unauthenticated -> 401 (got {st})")
+
+        print("catalog")
+        st, d = _req(port, "GET", "/api/query/catalog", admin)
+        check(st == 200 and isinstance(d.get("data"), list) and len(d["data"]) == len(nl_query.INTENTS),
+              f"catalog -> 200 with every intent (got {st})")
+        st, _ = _req(port, "GET", "/api/query/catalog", member)
+        check(st == 403, f"catalog member -> 403 (got {st})")
+
+        print("question path with the local model down")
+        st, d = _req(port, "POST", "/api/query/nl", admin,
+                     {"question": "who are our top investors by committed capital?"})
+        check(st == 503 and _data(d).get("error") == "model_unavailable",
+              f"question + dead model -> 503 model_unavailable (got {st}, {_data(d).get('error')})")
+        check(_data(d).get("question"), "question echoed back even on outage")
+
+        print("audit trail")
+        db = sqlite3.connect(os.environ["CRM_DB_PATH"])
+        n = db.execute("SELECT COUNT(*) FROM audit_log WHERE entity_type='nl_query'").fetchone()[0]
+        db.close()
+        check(n >= 2, f"executed queries are audited (entity_type=nl_query rows: {n})")
+    finally:
+        httpd.shutdown()
+
+    print()
+    if FAILS:
+        print(f"{len(FAILS)} FAILED")
+        for f in FAILS:
+            print("  - " + f)
+        sys.exit(1)
+    print("ALL PASS")
+
+
+if __name__ == "__main__":
+    main()

backend/nl_query/test_translate.py

@@ -0,0 +1,107 @@
+#!/usr/bin/env python3
+"""Tests for the W2 NL translator (question -> {intent, slots}) — the local-model leg.
+
+The model is stubbed via an injected chat_fn, so this runs fully offline (no Spark, no
+network). Covers:
+  - build_system() exposes the whole intent catalog as the model's closed vocabulary;
+  - translate() returns the parsed {intent, slots} and DROPS slot keys the intent doesn't
+    declare (model noise), while every surviving value is still validated downstream;
+  - the translation failure modes: no intent fit -> no_match; unparseable -> no_match;
+    local model unreachable -> model_unavailable (so the endpoint can 503);
+  - answer() chains translate + the validated runner end-to-end, and a HALLUCINATED intent
+    from the model is still rejected by the validator (the model output is never trusted).
+
+Run: cd backend && python3 nl_query/test_translate.py
+"""
+import os
+import sys
+import tempfile
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))  # backend/
+import server  # noqa: E402
+import nl_query  # noqa: E402
+
+T = nl_query  # exercise the public API (translate/answer/build_system are re-exported)
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+def main():
+    print("build_system")
+    sysprompt = nl_query.build_system()
+    check(all(k in sysprompt for k in nl_query.INTENTS), "system prompt lists every intent key")
+    check("days (integer, default 90)" in sysprompt, "system prompt renders int slot + default")
+    check("one of any|inbound|outbound" in sysprompt, "system prompt renders enum choices")
+
+    print("translate")
+    captured = {}
+
+    def fake(prompt, system):
+        captured["system"] = system
+        captured["prompt"] = prompt
+        return {"intent": "investors_cold", "slots": {"days": 90, "bogus": "x"}}
+
+    r = T.translate("who's gone quiet for 3 months?", chat_fn=fake)
+    check(r == {"intent": "investors_cold", "slots": {"days": 90}},
+          f"routes to intent + drops unknown slot 'bogus': {r}")
+    check(nl_query.INTENTS and "investors_cold" in captured["system"], "chat_fn received the catalog")
+    check(captured["prompt"] == "who's gone quiet for 3 months?", "chat_fn received the question")
+
+    check(T.translate("x", chat_fn=lambda q, s: {"intent": None})["error"] == "no_match",
+          "intent null -> no_match")
+    check(T.translate("x", chat_fn=lambda q, s: None)["error"] == "no_match",
+          "unparseable model reply -> no_match")
+    check(T.translate("", chat_fn=lambda q, s: {"intent": "x"})["error"] == "no_match",
+          "empty question -> no_match (no model call needed)")
+
+    def boom(q, s):
+        raise RuntimeError("spark down")
+
+    check(T.translate("x", chat_fn=boom)["error"] == "model_unavailable",
+          "local model unreachable -> model_unavailable")
+
+    print("answer (end-to-end through the validated runner)")
+    server.init_db()
+    conn = server.get_db()
+    conn.execute("INSERT INTO fundraising_investors (id, investor_name, lead, graveyard, "
+                 "source_row_id, total_invested) VALUES "
+                 "('a','Acme Capital','Jon',0,'a',5000000),"
+                 "('b','Beta Partners','Grant',0,'b',2000000),"
+                 "('g','Ghost','Grant',1,'g',9000000)")
+    conn.commit()
+
+    r = T.answer(conn, "top investors",
+                 chat_fn=lambda q, s: {"intent": "top_investors_committed", "slots": {"limit": 2}})
+    check([x["investor_name"] for x in r["rows"]] == ["Acme Capital", "Beta Partners"],
+          "answer() runs the translated query")
+    check(r["question"] == "top investors", "answer() echoes the original question")
+
+    r = T.answer(conn, "nonsense", chat_fn=lambda q, s: {"intent": "made_up_intent", "slots": {}})
+    check(r.get("error") == "unknown_intent", "hallucinated intent is rejected by the validator")
+    check(r["question"] == "nonsense", "answer() echoes question on error too")
+
+    r = T.answer(conn, "anything", chat_fn=boom)
+    check(r.get("error") == "model_unavailable", "answer() surfaces a model outage")
+
+    conn.close()
+    print()
+    if FAILS:
+        print(f"{len(FAILS)} FAILED")
+        for f in FAILS:
+            print("  - " + f)
+        sys.exit(1)
+    print("ALL PASS")
+
+
+if __name__ == "__main__":
+    main()

backend/nl_query/translate.py

@@ -0,0 +1,108 @@
+"""NL-query translator — plain-English question -> {intent, slots} on the LOCAL model.
+
+The model's ONLY job is to pick one curated intent and fill its typed slots; it never
+touches the database, never sees a row, and never writes SQL. Its output is untrusted and
+is handed straight to the runner's validator (runner.validate), which is the trust boundary.
+
+LOCAL-ONLY BY CONSTRUCTION. Translation runs on the local Qwen via Spark Control
+(SPARK_CONTROL_URL), the same sanctioned local leg as intake/digest — so the question never
+leaves the box and there is NO Claude path and NO redaction boundary to manage here (that
+was the whole point of the W2 simplification: the answer is sensitive and never leaves; the
+question is generic English and is translated locally). If the local model ever proves too
+weak, a Claude-behind-redaction translator could be slotted in as an alternative `chat_fn`
+WITHOUT changing the validator/executor — but it is deliberately not built.
+
+`chat_fn(prompt, system) -> dict|None` is injectable so the whole translation leg is testable
+offline without Spark. The default calls the ingest Spark client (lazy import — it ships in
+the Docker image, not the bare CRM).
+"""
+from .intents import INTENTS
+from .runner import run_query
+
+
+def _default_chat_json(prompt, system):
+    import os
+    import sys
+    sys.path.insert(0, os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "ingest"))
+    import llm  # noqa: E402  (ingest Spark client; raises if Spark is unreachable)
+    return llm.chat_json(prompt, system=system, max_tokens=400)
+
+
+def _render_slot(name, spec):
+    t = spec["type"]
+    if t == "int":
+        extra = f", default {spec['default']}" if "default" in spec else ""
+        return f"{name} (integer{extra})"
+    if t == "enum":
+        extra = f", default {spec['default']}" if "default" in spec else ""
+        return f"{name} (one of {'|'.join(spec['choices'])}{extra})"
+    req = ", required" if spec.get("required") else ", optional"
+    return f"{name} (text{req})"
+
+
+def build_system():
+    """The system prompt: the full intent catalog as the model's closed vocabulary."""
+    lines = [
+        "You translate a question about a venture fund's investor database into ONE "
+        "structured query. Respond with ONLY a JSON object and nothing else:",
+        '  {"intent": "<one key below, or null>", "slots": {<slot>: <value>}}',
+        "",
+        "Rules:",
+        "- Choose the single best-fitting intent. If none fits, return {\"intent\": null}.",
+        "- Use ONLY the slot names listed for the chosen intent; omit a slot to accept its default.",
+        "- Convert natural durations to the integer a slot wants: '3 months'->90, 'a quarter'->90, "
+        "'6 weeks'->42, 'a year'/'year to date'->365.",
+        "- Copy names, cities and people verbatim from the question into text slots.",
+        "- No commentary, no markdown, JSON only.",
+        "",
+        "Intents:",
+    ]
+    for key, spec in INTENTS.items():
+        slots = spec["slots"]
+        slot_str = "; ".join(_render_slot(n, s) for n, s in slots.items()) or "(none)"
+        lines.append(f"- {key}: {spec['summary']}")
+        lines.append(f"    slots: {slot_str}")
+        if spec.get("example"):
+            lines.append(f"    e.g. \"{spec['example']}\"")
+    return "\n".join(lines)
+
+
+def translate(question, *, chat_fn=None):
+    """Map a question to {intent, slots} on the local model. Returns that dict, or an error
+    dict {error, detail}: 'model_unavailable' (local model unreachable -> the endpoint 503s)
+    or 'no_match' (the model could not map the question to any intent)."""
+    chat_fn = chat_fn or _default_chat_json
+    q = (question or "").strip()
+    if not q:
+        return {"error": "no_match", "detail": "empty question"}
+    try:
+        data = chat_fn(q, build_system())
+    except Exception as exc:               # connection/runtime failure on the LOCAL model
+        return {"error": "model_unavailable", "detail": str(exc)}
+    if not isinstance(data, dict):
+        return {"error": "no_match", "detail": "model returned no parseable JSON"}
+    intent = data.get("intent")
+    if intent in (None, "", "null", "none"):
+        return {"error": "no_match", "detail": "no intent fit the question"}
+    slots = data.get("slots")
+    slots = slots if isinstance(slots, dict) else {}
+    # Drop slot KEYS the chosen intent doesn't declare — model noise, not a safety concern
+    # (every surviving VALUE still goes through full type validation in the runner). Unknown
+    # intents are left as-is so the runner rejects them as unknown_intent.
+    if intent in INTENTS:
+        allowed = INTENTS[intent]["slots"]
+        slots = {k: v for k, v in slots.items() if k in allowed}
+    return {"intent": intent, "slots": slots}
+
+
+def answer(conn, question, *, chat_fn=None, audit_fn=None, actor=None, source="api"):
+    """End-to-end: translate a question locally, then run it through the validated runner.
+    Returns the runner's result (with the interpreted intent/slots, so a human can see how
+    the question was read) plus the original question, or a translation error dict."""
+    t = translate(question, chat_fn=chat_fn)
+    if t.get("error"):
+        return {**t, "question": question}
+    result = run_query(conn, t["intent"], t["slots"],
+                       audit_fn=audit_fn, actor=actor, source=source)
+    result["question"] = question
+    return result

backend/nl_query/try_questions.py

@@ -0,0 +1,51 @@
+#!/usr/bin/env python3
+"""Dev harness — fire questions at the LOCAL model and print how each is translated.
+
+Lets you eyeball whether the local Qwen maps real questions to the right curated query
+(intent + slots), against your real Spark, with NO UI, auth, HTTP, or deploy. This is the
+cheap way to validate translation quality before building the web/Matrix surfaces. It only
+translates (it does not touch the DB), so no data is needed and nothing leaves the box.
+
+NOT shipped and NOT a test (no `test_` prefix) — a developer convenience.
+
+Needs SPARK_CONTROL_URL set (read from the repo .env) and the Spark reachable.
+Run:
+  python3 backend/nl_query/try_questions.py                 # the built-in sample set
+  python3 backend/nl_query/try_questions.py "when did we last email Acme?"
+"""
+import os
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))  # backend/
+import nl_query  # noqa: E402
+
+SAMPLES = [
+    "Which investors haven't we reached out to in the last 3 months?",
+    "Which investors do we owe follow-ups to?",
+    "What is Acme Capital's email and how much have they committed across funds?",
+    "When did we last reach out to Acme Capital?",
+    "What were the last 10 investor emails we sent, and who to?",
+    "What were the last 10 investor emails we received?",
+    "Who are all the investors located in Austin?",
+    "List our top 10 investors by committed capital.",
+    "List our top 10 pipeline investors by stage and the most recent conversation.",
+    "What is our total pipeline in dollars, split by stage?",
+    "What were the last investor emails sent by Grant?",
+    "How many emails has Jonathan sent this week, this month, and year to date?",
+]
+
+
+def main():
+    questions = sys.argv[1:] or SAMPLES
+    print(f"Translating {len(questions)} question(s) on the local model "
+          f"(SPARK_CONTROL_URL={os.environ.get('SPARK_CONTROL_URL', '(unset)')})\n")
+    for q in questions:
+        r = nl_query.translate(q)
+        if r.get("error"):
+            print(f"  ?  {q}\n     -> [{r['error']}] {r.get('detail', '')}\n")
+        else:
+            print(f"  ?  {q}\n     -> {r['intent']}  slots={r['slots']}\n")
+
+
+if __name__ == "__main__":
+    main()

backend/run_tests.py

@@ -0,0 +1,67 @@
+#!/usr/bin/env python3
+"""Aggregate test runner for the backend suite.
+
+The backend tests are standalone scripts (each with `if __name__ == "__main__"`, no
+pytest). This discovers every backend/**/test_*.py and runs each in its OWN subprocess
+(tests set os.environ and import `server` with different configs, so isolation matters),
+prints a one-line PASS/FAIL per test, dumps output only for failures, and exits non-zero
+if any test fails.
+
+Run:  python3 backend/run_tests.py        (from the repo root)
+  or: cd backend && python3 run_tests.py
+Filter: python3 backend/run_tests.py soft_delete redaction   # substring match on path
+"""
+import os
+import subprocess
+import sys
+import time
+
+BACKEND = os.path.dirname(os.path.abspath(__file__))
+
+
+def discover(filters):
+    found = []
+    for root, dirs, files in os.walk(BACKEND):
+        dirs[:] = [d for d in dirs if d != "__pycache__"]
+        for f in files:
+            if f.startswith("test_") and f.endswith(".py"):
+                path = os.path.join(root, f)
+                rel = os.path.relpath(path, BACKEND)
+                if not filters or any(flt in rel for flt in filters):
+                    found.append(path)
+    return sorted(found)
+
+
+def main():
+    filters = sys.argv[1:]
+    tests = discover(filters)
+    if not tests:
+        print("No tests matched.")
+        sys.exit(1)
+    print(f"Running {len(tests)} backend test(s)\n")
+
+    passed, failed = [], []
+    t0 = time.time()
+    for path in tests:
+        rel = os.path.relpath(path, BACKEND)
+        proc = subprocess.run([sys.executable, path], cwd=BACKEND,
+                              stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
+        if proc.returncode == 0:
+            passed.append(rel)
+            print(f"  PASS  {rel}")
+        else:
+            failed.append(rel)
+            print(f"  FAIL  {rel}")
+            sys.stdout.write(proc.stdout.decode("utf-8", "replace").rstrip() + "\n")
+
+    print(f"\n{len(passed)}/{len(tests)} passed in {time.time() - t0:.1f}s")
+    if failed:
+        print("FAILED:")
+        for f in failed:
+            print(f"  - {f}")
+        sys.exit(1)
+    print("ALL PASS")
+
+
+if __name__ == "__main__":
+    main()

backend/scripts/contacts_census.sql

@@ -0,0 +1,43 @@
+-- contacts_census.sql — A/B/C census for the contacts <-> fundraising_contacts consolidation.
+-- COUNTS ONLY: no names, emails, or amounts — nothing identifying leaves the box.
+--
+-- Run wherever you have the DB:
+--   on the box (container shell):  sqlite3 /data/crm.db < contacts_census.sql
+--   against a local copy:          sqlite3 data/crm.db  < contacts_census.sql
+--
+-- Canonical link: fundraising_contacts.contact_id -> contacts.id (migration 0004).
+--   A = contact linked to >=1 grid pill   (healthy overlap — already done)
+--   B = live contact with NO grid pill     (needs a grid row/pill created)
+--   C = grid pill with contact_id IS NULL  (needs a contacts row created)
+-- Sanity: A + B should equal "total live contacts".
+
+.headers on
+.mode column
+
+WITH
+linked AS (   -- contacts referenced by at least one grid pill
+  SELECT DISTINCT c.id
+  FROM contacts c
+  WHERE c.deleted_at IS NULL
+    AND EXISTS (SELECT 1 FROM fundraising_contacts fc WHERE fc.contact_id = c.id)
+),
+unlinked AS ( -- B: live contacts with no pill pointing at them
+  SELECT c.id, c.contact_type
+  FROM contacts c
+  WHERE c.deleted_at IS NULL
+    AND NOT EXISTS (SELECT 1 FROM fundraising_contacts fc WHERE fc.contact_id = c.id)
+)
+SELECT 'total live contacts'                              AS metric, (SELECT COUNT(*) FROM contacts WHERE deleted_at IS NULL) AS n
+UNION ALL SELECT 'A: linked (contact <-> grid pill)',     (SELECT COUNT(*) FROM linked)
+UNION ALL SELECT 'B: contacts-only (no grid pill)',       (SELECT COUNT(*) FROM unlinked)
+UNION ALL SELECT '   ...of B, contact_type=investor',     (SELECT COUNT(*) FROM unlinked WHERE contact_type='investor')
+UNION ALL SELECT '   ...of B, contact_type=prospect',     (SELECT COUNT(*) FROM unlinked WHERE contact_type='prospect')
+UNION ALL SELECT '   ...of B, with >=1 live communication', (SELECT COUNT(*) FROM unlinked u WHERE EXISTS (SELECT 1 FROM communications cm WHERE cm.contact_id=u.id AND cm.deleted_at IS NULL))
+UNION ALL SELECT '   ...of B, with >=1 live opportunity',   (SELECT COUNT(*) FROM unlinked u WHERE EXISTS (SELECT 1 FROM opportunities o WHERE o.contact_id=u.id AND o.deleted_at IS NULL))
+UNION ALL SELECT 'C: pill-only (contact_id IS NULL)',     (SELECT COUNT(*) FROM fundraising_contacts WHERE contact_id IS NULL)
+UNION ALL SELECT '   dangling pills (contact_id set, no live contact)',
+                 (SELECT COUNT(*) FROM fundraising_contacts fc
+                  WHERE fc.contact_id IS NOT NULL
+                    AND NOT EXISTS (SELECT 1 FROM contacts c WHERE c.id=fc.contact_id AND c.deleted_at IS NULL))
+UNION ALL SELECT 'context: total grid pills (fundraising_contacts)',   (SELECT COUNT(*) FROM fundraising_contacts)
+UNION ALL SELECT 'context: total grid rows (fundraising_investors)',   (SELECT COUNT(*) FROM fundraising_investors);

backend/scripts/seed_synthetic.py

@@ -11,8 +11,8 @@ What it builds (into a SEPARATE dev DB, never crm.db):
     core migration (backend/migrations/), so the canonical/interaction/graph
     tables exist.
   * A classic-model dataset: organizations, contacts (investors + prospects),
-    opportunities across pipeline stages, communications with entity-rich prose
+    opportunities across pipeline stages, and communications with entity-rich
-    notes, and lp_profiles.
+    prose notes.
   * A fundraising grid (fundraising_state.grid_json) populated via the real
     sync_fundraising_relational() code path, so the normalized mirror + the
     grid->classic bridge behave exactly as in production.
@@ -179,7 +179,7 @@ def main():
              f"Prospect sourced via {random.choice(['X DM', 'warm intro', 'podcast'])}.", uid, now()))
         contacts.append((cid, first, last, org_name, "prospect"))
 
-    # ── opportunities + lp_profiles + communications ──
+    # ── opportunities + communications ──
     stages = server.PIPELINE_STAGES
     for idx, (cid, first, last, org_name, ctype) in enumerate(contacts):
         person = f"{first} {last}"
@@ -193,25 +193,12 @@ def main():
                 "expected_amount, probability, fund_name, description, next_step, owner_id, priority, updated_at) "
                 "VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?)",
                 (gen(), f"{org_name or person} — {fund_label}", cid, org_ids.get(org_name) if org_name else None,
-                 stage, amt if stage in ("committed", "funded") else 0, amt,
+                 stage, amt if stage == "commitment" else 0, amt,
-                 {"lead": 10, "outreach": 25, "meeting": 40, "due_diligence": 60, "committed": 90, "funded": 100}[stage],
+                 {"lead": 10, "engaged": 35, "diligence": 60, "commitment": 90}[stage],
                  fund_label, f"Potential {fund_label} allocation for {person}.",
                  random.choice(["Send deck", "Schedule call", "Await IC", "Send subdocs"]),
                  uid, random.choice(["low", "medium", "high"]), now()))
 
-        # lp_profile for ~closed investors
-        if ctype == "investor" and idx % 2 == 0:
-            amt = random.choice(AMOUNTS)
-            conn.execute(
-                "INSERT INTO lp_profiles (id, contact_id, commitment_amount, funded_amount, commitment_date, "
-                "fund_name, investor_type, accredited, legal_docs_signed, wire_received, k1_sent, notes, updated_at) "
-                "VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?)",
-                (gen(), cid, amt, amt if idx % 4 == 0 else 0, past(120),
-                 random.choice(list(FUND_LABELS.values())),
-                 random.choice(["family_office", "institutional", "endowment", "individual"]),
-                 1, 1 if idx % 3 else 0, 1 if idx % 4 == 0 else 0, 0,
-                 f"Closed LP. Accreditation on file. Primary contact {person}.", now()))
-
         # 2-4 communications each, entity-rich prose
         for k in range(random.randint(2, 4)):
             ctype_comm, subj, body = random.choice(COMM_TEMPLATES)
@@ -275,7 +262,7 @@ def main():
 
     print(f"\nSynthetic dev DB written to: {db}")
     print("  Classic model:")
-    for t in ("organizations", "contacts", "opportunities", "communications", "lp_profiles"):
+    for t in ("organizations", "contacts", "opportunities", "communications"):
         print(f"    {t:<24} {count(t)}")
     print("  Fundraising grid (after real sync):")
     for t in ("fundraising_investors", "fundraising_contacts", "fundraising_funds",

backend/smtp_send.py

@@ -0,0 +1,99 @@
+"""Minimal outbound SMTP sender for the CRM (daily digest, test sends).
+
+Config comes ONLY from SMTP_* environment variables. Two ways those get set:
+  * dev / bare run: a local .env (see .env.example).
+  * Start9 box: docker_entrypoint.sh reads the files the "Configure Digest SMTP"
+    StartOS action writes under /data/secrets/smtp/ and exports them as env.
+The backend never reads those files directly, so dev and prod share one path.
+
+This is the package's OWN dedicated mailbox (per-package custom SMTP) — it is
+independent of any StartOS system-wide SMTP account; nothing here calls into the
+platform. Stdlib only (smtplib/ssl/email), consistent with the rest of runtime.
+"""
+import os
+import smtplib
+import ssl
+from email.message import EmailMessage
+
+
+class SMTPNotConfigured(Exception):
+    """Raised when SMTP_* env is absent — callers turn this into a clear 'not
+    configured' response rather than a 500."""
+
+
+def smtp_configured():
+    return bool(os.environ.get("SMTP_HOST", "").strip())
+
+
+def load_smtp_config():
+    host = os.environ.get("SMTP_HOST", "").strip()
+    if not host:
+        raise SMTPNotConfigured("SMTP_HOST is not set")
+    # Port/security come from a free-text action field; normalize defensively.
+    try:
+        port = int(str(os.environ.get("SMTP_PORT", "") or "587").strip())
+    except ValueError:
+        port = 587
+    security = (os.environ.get("SMTP_SECURITY", "") or "starttls").strip().lower()
+    if security not in ("starttls", "tls", "none"):
+        security = "starttls"
+    return {
+        "host": host,
+        "port": port,
+        "from_addr": os.environ.get("SMTP_FROM", "").strip(),
+        "username": os.environ.get("SMTP_USERNAME", "").strip(),
+        "password": os.environ.get("SMTP_PASSWORD", ""),
+        "security": security,
+    }
+
+
+def _connect(cfg, timeout):
+    """Open an authenticated SMTP connection per the configured security mode.
+    'tls' = implicit TLS (SMTPS, usually 465); 'starttls' = upgrade on 587;
+    'none' = plaintext (for a LAN relay that does its own transport security)."""
+    if cfg["security"] == "tls":
+        ctx = ssl.create_default_context()
+        server = smtplib.SMTP_SSL(cfg["host"], cfg["port"], timeout=timeout, context=ctx)
+    else:
+        server = smtplib.SMTP(cfg["host"], cfg["port"], timeout=timeout)
+        server.ehlo()
+        if cfg["security"] == "starttls":
+            server.starttls(context=ssl.create_default_context())
+            server.ehlo()
+    if cfg["username"]:
+        server.login(cfg["username"], cfg["password"])
+    return server
+
+
+def send_email(to_addrs, subject, body, *, html=None, cfg=None, timeout=30):
+    """Send one message. `to_addrs` is a str or list; `body` is plain text and
+    `html` an optional HTML alternative. Returns {'sent_to', 'from'} on success;
+    raises SMTPNotConfigured / ValueError / smtplib.SMTPException otherwise."""
+    cfg = cfg or load_smtp_config()
+    if isinstance(to_addrs, str):
+        to_addrs = [to_addrs]
+    to_addrs = [a for a in (str(x).strip() for x in to_addrs) if a]
+    if not to_addrs:
+        raise ValueError("no recipients")
+
+    from_addr = cfg["from_addr"] or cfg["username"]
+    if not from_addr:
+        raise SMTPNotConfigured("no SMTP_FROM or SMTP_USERNAME to use as sender")
+
+    msg = EmailMessage()
+    msg["From"] = from_addr
+    msg["To"] = ", ".join(to_addrs)
+    msg["Subject"] = subject
+    msg.set_content(body)
+    if html:
+        msg.add_alternative(html, subtype="html")
+
+    server = _connect(cfg, timeout)
+    try:
+        server.send_message(msg, from_addr=from_addr, to_addrs=to_addrs)
+    finally:
+        try:
+            server.quit()
+        except Exception:
+            pass
+    return {"sent_to": to_addrs, "from": from_addr}

backend/test_assets_traversal.py

@@ -0,0 +1,126 @@
+#!/usr/bin/env python3
+"""Regression test for the /assets/ path-traversal containment fix (v0.1.0:74).
+
+Before the fix, get_path()/urlparse did NOT normalize '..', so an unauthenticated
+GET /assets/../../data/crm.db (raw client, no client-side normalization) escaped the
+frontend root and read any file the process could — the LP DB, the JWT secret, the
+Gmail key. The fix resolves the target with os.path.realpath and 404s anything that
+does not stay under FRONTEND_ROOT (server.py, the `/assets/` branch of do_GET).
+
+This boots the REAL server in-process against a throwaway frontend root, plants a
+decoy "secret" OUTSIDE that root, and proves: (1) traversal vectors that resolve to a
+real readable file outside the root still 404 and leak no bytes; (2) the live crm.db
+path is 404'd; (3) URL-encoded separators don't help; (4) a legit in-bounds asset
+still serves 200 (the fix isn't over-broad). Synthetic only (guardrail #9).
+
+Run: cd backend && python3 test_assets_traversal.py
+"""
+import http.client
+import os
+import sys
+import tempfile
+import threading
+from http.server import ThreadingHTTPServer
+
+# Lay out a throwaway tree BEFORE importing server (FRONTEND_DIR/ROOT resolve at import):
+#   base/frontend/{index.html,assets/app.css}   <- the served root
+#   base/secret.txt                             <- a real file a traversal would target
+#   base/data/crm.db                            <- the live DB, created by init_db()
+_BASE = tempfile.mkdtemp()
+_FRONTEND = os.path.join(_BASE, "frontend")
+os.makedirs(os.path.join(_FRONTEND, "assets"))
+_DATA = os.path.join(_BASE, "data")
+os.makedirs(_DATA)
+with open(os.path.join(_FRONTEND, "index.html"), "w") as f:
+    f.write("<!doctype html><title>crm</title>")
+_CSS_MARKER = "/* legit-asset-marker-7f3a */"
+with open(os.path.join(_FRONTEND, "assets", "app.css"), "w") as f:
+    f.write(_CSS_MARKER)
+_SECRET_MARKER = "TOPSECRET-JWT-zq19"
+with open(os.path.join(_BASE, "secret.txt"), "w") as f:
+    f.write(_SECRET_MARKER)
+
+os.environ["CRM_FRONTEND_DIR"] = _FRONTEND
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import server  # noqa: E402
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Quiet(server.CRMHandler):
+    def log_message(self, *a):  # keep the test output clean
+        pass
+
+
+def _get(port, path):
+    """Raw GET with the path sent verbatim — http.client does NOT normalize '..',
+    which is exactly the unauthenticated raw-client threat the fix defends against."""
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    conn.request("GET", path)
+    resp = conn.getresponse()
+    body = resp.read().decode("utf-8", "replace")
+    conn.close()
+    return resp.status, body
+
+
+def main():
+    server.init_db()  # creates base/data/crm.db and the full schema
+    check(os.path.exists(os.environ["CRM_DB_PATH"]), "init_db created the live crm.db (a real traversal target)")
+
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), _Quiet)
+    port = httpd.server_address[1]
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        # ── legit in-bounds asset still serves (containment is not over-broad) ──
+        print("\n[legit asset]")
+        st, body = _get(port, "/assets/app.css")
+        check(st == 200, f"in-bounds /assets/app.css serves 200 (got {st})")
+        check(_CSS_MARKER in body, "in-bounds asset body is served intact")
+
+        # ── traversal to a REAL file outside the root: 404, zero bytes leaked ──
+        print("\n[traversal -> decoy secret outside the root]")
+        for vec in ["/assets/../../secret.txt",
+                    "/assets/../../../secret.txt",
+                    "/assets/..%2f..%2fsecret.txt",         # urlparse won't decode %2f
+                    "/assets/..%2F..%2Fsecret.txt"]:        # …nor uppercase %2F (some clients send it)
+            st, body = _get(port, vec)
+            check(st == 404, f"{vec} -> 404 (got {st})")
+            check(_SECRET_MARKER not in body, f"{vec} leaks no secret bytes")
+
+        # ── traversal to the live crm.db (the headline vector from the eval) ──
+        print("\n[traversal -> live crm.db]")
+        for vec in ["/assets/../../data/crm.db",
+                    "/assets/../data/crm.db",
+                    "/assets/..%2f..%2fdata%2fcrm.db"]:
+            st, body = _get(port, vec)
+            check(st == 404, f"{vec} -> 404 (got {st})")
+            check("SQLite format 3" not in body, f"{vec} leaks no DB header")
+
+        # ── deep absolute-style escape ──
+        print("\n[deep escape]")
+        st, body = _get(port, "/assets/../../../../../../../../etc/passwd")
+        check(st == 404, f"/assets/../../etc/passwd -> 404 (got {st})")
+        check("root:" not in body, "/etc/passwd not leaked")
+    finally:
+        httpd.shutdown()
+
+    print()
+    if FAILS:
+        print(f"FAILED ({len(FAILS)}):")
+        for f in FAILS:
+            print(f"  - {f}")
+        sys.exit(1)
+    print("ALL PASS (assets path-traversal containment)")
+
+
+if __name__ == "__main__":
+    main()

backend/test_contacts_grid_signals.py

@@ -0,0 +1,224 @@
+#!/usr/bin/env python3
+"""Tests for the mobile Contacts card's grid-derived signals (Phase 8a).
+
+GET /api/contacts enriches each classic contact with two read-only, live-derived fields
+sourced from the fundraising grid (the canonical investor model), for the mobile card:
+  - `committed`  -> the linked investor's total_invested (>0 drives the existing-LP avatar ring),
+                    mirroring existing_investor_by_source_row (committed capital, not graveyard);
+  - `pipeline_stage` -> that investor's live derived stage (drives the card's stage pill),
+                    or null when the investor isn't in the pipeline.
+  - `priority` -> that investor's priority flag (drives the mobile Contacts Priority sort, 8d).
+  - `source_row_id` -> that investor's grid row id (the "Open investor in Grid" deep-link target, 8h),
+                    present for ANY grid-linked contact (even a zero-commit prospect), null otherwise.
+A contact with no grid link (pure classic/legacy contact) gets committed 0 / stage null / priority false
+/ source_row_id null.
+Signals are derived fresh on read and never stored on the contact. Synthetic data only.
+
+Run: cd backend && python3 test_contacts_grid_signals.py
+"""
+import http.client
+import json
+import os
+import sqlite3
+import sys
+import tempfile
+import threading
+from http.server import ThreadingHTTPServer
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import server  # noqa: E402
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Quiet(server.CRMHandler):
+    def log_message(self, *a):
+        pass
+
+
+def _req(port, method, path, token=None, body=None):
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    headers = {}
+    if token:
+        headers["Authorization"] = "Bearer " + token
+    payload = None
+    if body is not None:
+        payload = json.dumps(body)
+        headers["Content-Type"] = "application/json"
+    conn.request(method, path, body=payload, headers=headers)
+    resp = conn.getresponse()
+    raw = resp.read().decode("utf-8", "replace")
+    conn.close()
+    data = None
+    if raw:
+        try:
+            data = json.loads(raw)
+        except ValueError:
+            pass
+    return resp.status, data
+
+
+# One fund column so a non-zero cell rolls up into total_invested (the "existing LP" signal).
+COLUMNS = [{"id": "fund1", "label": "Fund III", "isFund": True}]
+ROW_ACME = {"id": "rowAcme", "investor_name": "Acme Capital", "priority": True, "fund1": 250000,
+            "contacts": [{"name": "Jane Doe", "email": "jane@acme.com", "title": "GP"}]}
+ROW_BETA = {"id": "rowBeta", "investor_name": "Beta Capital", "fund1": 0,
+            "contacts": [{"name": "Pat Roe", "email": "pat@beta.com", "title": ""}]}
+
+
+def _db():
+    return sqlite3.connect(os.environ["CRM_DB_PATH"])
+
+
+def seed():
+    c = _db()
+    c.execute("INSERT INTO users (id,username,email,password_hash,full_name,role,is_active) "
+              "VALUES ('u1','grant','grant@ten31.example','x','Grant','admin',1)")
+    # A pure classic contact with NO fundraising-grid link (not an investor).
+    c.execute("INSERT INTO contacts (id,first_name,last_name,email,contact_type,status) "
+              "VALUES ('cLegacy','Vendor','Vince','vince@vendor.com','other','active')")
+    c.commit()
+    c.close()
+
+
+def _by_email(contacts, email):
+    return next((c for c in contacts if (c.get("email") or "").lower() == email), None)
+
+
+def main():
+    server.init_db()
+    seed()
+    token = server.create_token("u1", "grant", "admin")
+
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), _Quiet)
+    port = httpd.server_address[1]
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        st, _ = _req(port, "PUT", "/api/fundraising/state", token,
+                     {"grid": {"columns": COLUMNS, "rows": [ROW_ACME, ROW_BETA]}, "views": []})
+        check(st == 200, f"seed grid via PUT /state (got {st})")
+
+        # Put Acme into the pipeline at 'engaged' so its contact's card shows a stage pill.
+        st, d = _req(port, "POST", "/api/fundraising/pipeline/link", token,
+                     {"source_row_id": "rowAcme", "stage": "engaged"})
+        check(st in (200, 201), f"link Acme to pipeline @engaged (got {st}, {d})")
+
+        st, d = _req(port, "GET", "/api/contacts?limit=500", token)
+        contacts = (d or {}).get("data") or []
+        check(st == 200 and contacts, f"GET /api/contacts (got {st}, {len(contacts)} contacts)")
+
+        jane = _by_email(contacts, "jane@acme.com")
+        pat = _by_email(contacts, "pat@beta.com")
+        vince = _by_email(contacts, "vince@vendor.com")
+        check(jane is not None, "Acme's synced contact Jane Doe is in the directory")
+        check(pat is not None, "Beta's synced contact Pat Roe is in the directory")
+        check(vince is not None, "the pure classic contact Vince is in the directory")
+
+        # ── existing-LP ring signal: committed reflects the linked investor's rollup ──
+        print("\n[committed: existing-LP ring driven by the linked investor's total_invested]")
+        check((jane or {}).get("committed") == 250000,
+              f"Jane.committed == 250000 (existing LP) (got {(jane or {}).get('committed')})")
+        check((pat or {}).get("committed") == 0,
+              f"Pat.committed == 0 (zero-commit prospect, no ring) (got {(pat or {}).get('committed')})")
+        check((vince or {}).get("committed") == 0,
+              f"Vince.committed == 0 (no grid link) (got {(vince or {}).get('committed')})")
+
+        # ── stage-pill signal: pipeline_stage is the investor's live derived stage ──
+        print("\n[pipeline_stage: stage pill driven by the investor's live opp stage]")
+        check((jane or {}).get("pipeline_stage") == "engaged",
+              f"Jane.pipeline_stage == 'engaged' (got {(jane or {}).get('pipeline_stage')!r})")
+        check((pat or {}).get("pipeline_stage") is None,
+              f"Pat.pipeline_stage is None (not in pipeline) (got {(pat or {}).get('pipeline_stage')!r})")
+        check((vince or {}).get("pipeline_stage") is None,
+              f"Vince.pipeline_stage is None (no grid link) (got {(vince or {}).get('pipeline_stage')!r})")
+
+        # ── priority signal: flagged investor → contact's Priority-sort key (8d) ──
+        print("\n[priority: Contacts Priority sort driven by the investor's priority flag]")
+        check((jane or {}).get("priority") is True,
+              f"Jane.priority is True (Acme flagged) (got {(jane or {}).get('priority')!r})")
+        check((pat or {}).get("priority") is False,
+              f"Pat.priority is False (Beta not flagged) (got {(pat or {}).get('priority')!r})")
+        check((vince or {}).get("priority") is False,
+              f"Vince.priority is False (no grid link) (got {(vince or {}).get('priority')!r})")
+
+        # ── source_row_id signal: the "Open investor in Grid" deep-link target (8h) ──
+        print("\n[source_row_id: Open-in-Grid deep-link target = the linked investor's grid row id]")
+        check((jane or {}).get("source_row_id") == "rowAcme",
+              f"Jane.source_row_id == 'rowAcme' (got {(jane or {}).get('source_row_id')!r})")
+        check((pat or {}).get("source_row_id") == "rowBeta",
+              f"Pat.source_row_id == 'rowBeta' (present for a zero-commit linked contact) (got {(pat or {}).get('source_row_id')!r})")
+        check((vince or {}).get("source_row_id") is None,
+              f"Vince.source_row_id is None (no grid link) (got {(vince or {}).get('source_row_id')!r})")
+
+        # ── the get-by-id endpoint carries the same signals (mobile detail sheet, 8b) ──
+        print("\n[get-by-id: /api/contacts/{id} also injects committed + pipeline_stage]")
+        st, d = _req(port, "GET", f"/api/contacts/{jane['id']}", token)
+        detail = (d or {}).get("data") or {}
+        check(st == 200 and detail.get("committed") == 250000 and detail.get("pipeline_stage") == "engaged"
+              and detail.get("priority") is True,
+              f"detail carries committed/pipeline_stage/priority (got committed={detail.get('committed')}, stage={detail.get('pipeline_stage')!r}, priority={detail.get('priority')!r})")
+        st, d = _req(port, "GET", f"/api/contacts/{vince['id']}", token)
+        vdetail = (d or {}).get("data") or {}
+        check(st == 200 and vdetail.get("committed") == 0 and vdetail.get("pipeline_stage") is None
+              and vdetail.get("priority") is False,
+              f"unlinked contact detail has committed 0 / stage None / priority False (got {vdetail.get('committed')}, {vdetail.get('pipeline_stage')!r}, {vdetail.get('priority')!r})")
+
+        # ── stage tracks the board: advancing the opp re-derives the contact's stage ──
+        print("\n[derived-live: advancing the board stage re-derives the contact's pill]")
+        opp_id = None
+        st, d = _req(port, "GET", "/api/fundraising/state", token)
+        for r in (d or {}).get("data", {}).get("grid", {}).get("rows", []):
+            if r.get("id") == "rowAcme":
+                opp_id = r.get("opportunity_id")
+        st, _ = _req(port, "PATCH", f"/api/opportunities/{opp_id}/stage", token, {"stage": "diligence"})
+        check(st == 200, f"advance Acme's opp -> diligence (got {st})")
+        st, d = _req(port, "GET", "/api/contacts?limit=500", token)
+        jane2 = _by_email((d or {}).get("data") or [], "jane@acme.com")
+        check((jane2 or {}).get("pipeline_stage") == "diligence",
+              f"Jane.pipeline_stage re-derives to 'diligence' (got {(jane2 or {}).get('pipeline_stage')!r})")
+
+        # ── dedup: a contact linked to two investors exposes the highest-committed one ──
+        print("\n[dedup: highest-committed linked investor wins for a multi-linked contact]")
+        c = _db()
+        # Link Jane's classic contact to a SECOND, richer investor (direct rows — the grid sync
+        # makes one link per pill; this exercises the multi-link branch in contact_grid_signals).
+        c.execute("INSERT INTO fundraising_investors (id, investor_name, source_row_id, total_invested) "
+                  "VALUES ('inv2','Mega Fund LP','rowMega',500000)")
+        c.execute("INSERT INTO fundraising_contacts (id, investor_id, full_name, contact_id) "
+                  "VALUES ('fc2','inv2','Jane Doe',?)", (jane['id'],))
+        c.commit()
+        c.close()
+        st, d = _req(port, "GET", f"/api/contacts/{jane['id']}", token)
+        jd = (d or {}).get("data") or {}
+        check(jd.get("committed") == 500000,
+              f"multi-linked contact exposes the higher committed (500000 > 250000) (got {jd.get('committed')})")
+        # The winning (higher-committed) link is Mega Fund LP, which is not flagged → priority follows it.
+        check(jd.get("priority") is False,
+              f"multi-linked contact's priority follows the higher-committed investor (Mega, unflagged) (got {jd.get('priority')!r})")
+        # The deep-link target also follows the winning link → Mega's grid row (rowMega), not rowAcme.
+        check(jd.get("source_row_id") == "rowMega",
+              f"multi-linked contact's source_row_id follows the higher-committed investor (rowMega) (got {jd.get('source_row_id')!r})")
+    finally:
+        httpd.shutdown()
+
+    print()
+    if FAILS:
+        print(f"FAILED ({len(FAILS)}):")
+        for m in FAILS:
+            print("  - " + m)
+        sys.exit(1)
+    print("All contacts-grid-signals tests passed.")
+
+
+if __name__ == "__main__":
+    main()

backend/test_dashboard_report.py

@@ -0,0 +1,135 @@
+#!/usr/bin/env python3
+"""Regression test for the dashboard KPI repoint + lp_profiles retirement (2026-06-16).
+
+"Total Committed" used to SUM lp_profiles.commitment_amount — an orphaned table with no
+reachable input path, so the dashboard read ~$0 while the real commitments lived in the
+fundraising grid. It now sums fundraising_investors.total_invested (the canonical grid
+rollup) with graveyarded (written-off) investors excluded, "Total Funded" is dropped
+(the grid has no funded-vs-committed concept), and the /api/lp-profiles* + lp-breakdown
+endpoints are gone.
+
+v0.1.0:106 repointed "Total LPs" / "Prospects" off the retired contacts.contact_type onto
+the canonical grid (investor entities): an LP = a grid investor with total_invested > 0
+(graveyard excluded); a prospect = a live grid row with $0 committed (graveyard + the
+'Untitled Investor' blank-row placeholder excluded).
+
+This boots the REAL server against a temp DB, seeds grid investors (live LP, graveyarded,
+live prospect, blank placeholder), and asserts: total_committed reflects the live grid
+rollup only, total_lps / total_prospects use the grid-entity definitions, the metrics no
+longer carry a total_funded key, and the retired routes 404. Synthetic only.
+
+Run: cd backend && python3 test_dashboard_report.py
+"""
+import http.client
+import json
+import os
+import sqlite3
+import sys
+import tempfile
+import threading
+from http.server import ThreadingHTTPServer
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import server  # noqa: E402
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Quiet(server.CRMHandler):
+    def log_message(self, *a):
+        pass
+
+
+def _get(port, path, token):
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    conn.request("GET", path, headers={"Authorization": "Bearer " + token})
+    resp = conn.getresponse()
+    body = resp.read().decode("utf-8", "replace")
+    conn.close()
+    data = None
+    if body:
+        try:
+            data = json.loads(body)
+        except ValueError:
+            pass
+    return resp.status, data
+
+
+def seed():
+    c = sqlite3.connect(os.environ["CRM_DB_PATH"])
+    c.execute("INSERT INTO users (id,username,email,password_hash,full_name,role,is_active) "
+              "VALUES ('u1','grant','grant@ten31.example','x','Grant','admin',1)")
+    # live investor committed 3,000,000; graveyarded investor committed 500,000 (must be excluded)
+    c.execute("INSERT INTO fundraising_investors (id,investor_name,source_row_id,total_invested,graveyard) "
+              "VALUES ('fiLive','Harbor LP','rowLive',3000000,0)")
+    c.execute("INSERT INTO fundraising_investors (id,investor_name,source_row_id,total_invested,graveyard) "
+              "VALUES ('fiDead','Passed LP','rowDead',500000,1)")
+    # a live prospect (in the grid, $0 committed) and a blank placeholder row — the prospect
+    # count includes the former and excludes the latter ('Untitled Investor' = a blank grid row)
+    c.execute("INSERT INTO fundraising_investors (id,investor_name,source_row_id,total_invested,graveyard) "
+              "VALUES ('fiProspect','Prospect Co','rowProspect',0,0)")
+    c.execute("INSERT INTO fundraising_investors (id,investor_name,source_row_id,total_invested,graveyard) "
+              "VALUES ('fiBlank','Untitled Investor','rowBlank',0,0)")
+    # one live + one soft-deleted contact: total_contacts must count only the live one
+    # (guards the deleted_at filter added alongside the contact_type repoint)
+    c.execute("INSERT INTO contacts (id,first_name,last_name) VALUES ('ctLive','Ann','Live')")
+    c.execute("INSERT INTO contacts (id,first_name,last_name,deleted_at) "
+              "VALUES ('ctGone','Bob','Gone','2026-06-01T00:00:00Z')")
+    c.commit()
+    c.close()
+
+
+def main():
+    server.init_db()
+    seed()
+    token = server.create_token("u1", "grant", "admin")
+
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), _Quiet)
+    port = httpd.server_address[1]
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        print("\n[dashboard total_committed comes from the grid, graveyard excluded]")
+        st, dash = _get(port, "/api/reports/dashboard", token)
+        check(st == 200, f"GET dashboard -> 200 (got {st})")
+        metrics = (dash or {}).get("data", {}).get("metrics", {})
+        check(metrics.get("total_committed") == 3000000,
+              f"total_committed = live grid rollup only (3,000,000; got {metrics.get('total_committed')})")
+        check("total_funded" not in metrics,
+              f"total_funded key dropped from metrics (got keys {sorted(metrics)})")
+
+        print("\n[Total LPs / Prospects derived from the grid, not the retired contacts.contact_type]")
+        check(metrics.get("total_lps") == 1,
+              f"total_lps = grid investors committed>0, graveyard excluded (1; got {metrics.get('total_lps')})")
+        check(metrics.get("total_prospects") == 1,
+              f"total_prospects = grid rows with $0 committed; graveyard + 'Untitled Investor' excluded (1; got {metrics.get('total_prospects')})")
+        check(metrics.get("total_contacts") == 1,
+              f"total_contacts excludes soft-deleted contacts (1; got {metrics.get('total_contacts')})")
+
+        print("\n[retired lp_profiles endpoints 404]")
+        for path in ("/api/lp-profiles", "/api/lp-profiles/anything", "/api/reports/lp-breakdown"):
+            st, _ = _get(port, path, token)
+            check(st == 404, f"GET {path} -> 404 (got {st})")
+    finally:
+        httpd.shutdown()
+
+    print()
+    if FAILS:
+        print(f"FAILED ({len(FAILS)}):")
+        for f in FAILS:
+            print(f"  - {f}")
+        sys.exit(1)
+    print("ALL PASS (dashboard KPI repoint + lp_profiles retirement)")
+
+
+if __name__ == "__main__":
+    main()

backend/test_digest_builder.py

@@ -0,0 +1,419 @@
+#!/usr/bin/env python3
+"""Tests for the daily activity digest (Phase B): the per-user + per-investor
+activity queries (soft-delete filtered), inbound dedup, the two-section body, the
+local-summary fallback, the DB-backed policy resolver, and the scheduler's
+once-per-day / send-hour / policy / force guards.
+
+The local Spark model and the mail transport are stubbed — no network. Synthetic
+data only (guardrail #9).
+Run: cd backend && python3 test_digest_builder.py
+"""
+import json
+import os
+import sqlite3
+import sys
+import tempfile
+from datetime import datetime, timezone
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+os.environ["CRM_DB_PATH"] = os.path.join(tempfile.mkdtemp(), "crm.db")
+os.environ.setdefault("CRM_DATA_DIR", os.path.dirname(os.environ["CRM_DB_PATH"]))
+os.environ["CRM_DIGEST_ENABLED"] = "1"   # so the non-force scheduler path is live
+
+import digest_builder  # noqa: E402
+from email_integration import digest_scheduler  # noqa: E402
+
+FAILS = []
+SINCE = "2026-06-17T00:00:00Z"
+UNTIL = "2026-06-18T00:00:00Z"
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+def _conn():
+    conn = sqlite3.connect(os.environ["CRM_DB_PATH"])
+    conn.row_factory = sqlite3.Row
+    return conn
+
+
+def setup():
+    conn = _conn()
+    conn.executescript("""
+        CREATE TABLE users (id TEXT PRIMARY KEY, username TEXT, full_name TEXT,
+            email TEXT, role TEXT, is_active INT DEFAULT 1);
+        CREATE TABLE email_accounts (id TEXT PRIMARY KEY, user_id TEXT, email_address TEXT);
+        CREATE TABLE emails (id TEXT PRIMARY KEY, subject TEXT, body_text TEXT, snippet TEXT,
+            from_email TEXT, sent_at TEXT, is_matched INT DEFAULT 1);
+        CREATE TABLE email_account_messages (id TEXT PRIMARY KEY, email_id TEXT, account_id TEXT,
+            gmail_message_id TEXT, gmail_thread_id TEXT, is_sent INT DEFAULT 0, deleted_at TEXT);
+        CREATE TABLE email_investor_links (id TEXT PRIMARY KEY, email_id TEXT,
+            fundraising_investor_id TEXT, organization_id TEXT, contact_id TEXT, matched_address TEXT);
+        CREATE TABLE fundraising_investors (id TEXT PRIMARY KEY, investor_name TEXT);
+        CREATE TABLE organizations (id TEXT PRIMARY KEY, name TEXT, deleted_at TEXT);
+        CREATE TABLE contacts (id TEXT PRIMARY KEY, first_name TEXT, last_name TEXT, deleted_at TEXT);
+        CREATE TABLE app_settings (key TEXT PRIMARY KEY, value_json TEXT, updated_at TEXT);
+    """)
+    conn.executemany("INSERT INTO users (id,username,full_name,email,role,is_active) VALUES (?,?,?,?,?,?)", [
+        ("u1", "grant", "Grant Gilliam", "grant@ten31.xyz", "admin", 1),
+        ("u2", "jk", "Jonathan K", "jk@ten31.xyz", "member", 1),
+        ("u3", "retired", "Old Admin", "old@ten31.xyz", "admin", 0),  # inactive -> excluded
+    ])
+    conn.executemany("INSERT INTO email_accounts (id,user_id,email_address) VALUES (?,?,?)", [
+        ("a1", "u1", "grant@ten31.xyz"), ("a2", "u2", "jk@ten31.xyz"), ("a3", "u3", "old@ten31.xyz"),
+    ])
+    conn.executemany("INSERT INTO fundraising_investors (id,investor_name) VALUES (?,?)", [
+        ("inv1", "Harbor & Vine"), ("inv2", "Brightwater Capital"), ("inv3", "Vela Partners"),
+    ])
+    conn.executemany("INSERT INTO organizations (id,name,deleted_at) VALUES (?,?,?)", [
+        ("org1", "Summit Fund", None), ("org2", "Deleted Org", "2026-06-01T00:00:00Z"),
+    ])
+    conn.executemany("INSERT INTO contacts (id,first_name,last_name,deleted_at) VALUES (?,?,?,?)", [
+        ("c1", "Jane", "Roe", None),
+    ])
+    # emails: id, subject, body, from_email, sent_at, is_matched. Outbound = from one
+    # of our own mailboxes (grant@/jk@/old@ ten31.xyz); inbound = from outside.
+    conn.executemany("INSERT INTO emails (id,subject,body_text,from_email,sent_at,is_matched) VALUES (?,?,?,?,?,?)", [
+        ("e1", "Fund III terms", "Discussing Fund III terms", "grant@ten31.xyz", "2026-06-17T14:00:00Z", 1),
+        ("e2", "Re: allocation", "Question about allocation", "lp@brightwater.example", "2026-06-17T09:00:00Z", 1),
+        ("e3", "Intro", "Summit intro thread", "jk@ten31.xyz", "2026-06-17T11:00:00Z", 1),
+        ("e4", "Coffee", "Catch up note", "jane@roe.example", "2026-06-17T16:00:00Z", 1),
+        ("e5", "Wire", "Wire instructions", "ir@summitcap.example", "2026-06-17T17:00:00Z", 1),
+        ("e6", "Old", "ancient", "grant@ten31.xyz", "2026-06-10T10:00:00Z", 1),    # out of window
+        ("e7", "Tombstoned", "deleted sighting", "lp@harborvine.example", "2026-06-17T08:00:00Z", 1),
+        ("e8", "Inactive", "from retired user", "old@ten31.xyz", "2026-06-17T12:00:00Z", 1),
+        ("e9", "Unmatched", "not matched", "lp@harborvine.example", "2026-06-17T13:00:00Z", 0),  # is_matched=0
+        ("e10", "Group update", "inbound to two of us", "lp@vela.example", "2026-06-17T15:00:00Z", 1),
+    ])
+    # sightings: id, email_id, account_id, is_sent, deleted_at
+    conn.executemany(
+        "INSERT INTO email_account_messages (id,email_id,account_id,gmail_message_id,gmail_thread_id,is_sent,deleted_at) "
+        "VALUES (?,?,?,?,?,?,?)", [
+            ("s1", "e1", "a1", "g1", "t1", 1, None),   # grant SENT
+            ("s2", "e2", "a1", "g2", "t2", 0, None),   # grant RECEIVED
+            ("s3", "e3", "a2", "g3", "t3", 1, None),   # jk SENT
+            ("s4", "e4", "a1", "g4", "t4", 0, None),   # grant RECEIVED (contact)
+            ("s5", "e5", "a1", "g5", "t5", 0, None),   # grant RECEIVED (deleted org)
+            ("s6", "e6", "a1", "g6", "t6", 0, None),   # out of window
+            ("s7", "e7", "a1", "g7", "t7", 0, "2026-06-17T09:00:00Z"),  # tombstoned
+            ("s8", "e8", "a3", "g8", "t8", 1, None),   # inactive user
+            ("s9", "e9", "a1", "g9", "t9", 0, None),   # unmatched email
+            ("s10a", "e10", "a1", "g10a", "t10", 0, None),  # e10 received by grant ...
+            ("s10b", "e10", "a2", "g10b", "t10", 0, None),  # ... and by jk (dedup target)
+        ])
+    # investor links: id, email_id, fr_investor, org, contact, matched_address
+    conn.executemany(
+        "INSERT INTO email_investor_links (id,email_id,fundraising_investor_id,organization_id,contact_id,matched_address) "
+        "VALUES (?,?,?,?,?,?)", [
+            ("l1", "e1", "inv1", None, None, "lp@harborvine.example"),
+            ("l2", "e2", "inv2", None, None, "lp@brightwater.example"),
+            ("l3", "e3", None, "org1", None, "ir@summitfund.example"),     # org name
+            ("l4", "e4", None, None, "c1", "jane@roe.example"),            # contact name
+            ("l5", "e5", None, "org2", None, "ir@summitcap.example"),      # deleted org -> address
+            ("l6", "e6", "inv1", None, None, "lp@harborvine.example"),
+            ("l7", "e7", "inv1", None, None, "lp@harborvine.example"),
+            ("l8", "e8", "inv1", None, None, "lp@harborvine.example"),
+            ("l9", "e9", "inv1", None, None, "lp@harborvine.example"),
+            ("l10", "e10", "inv3", None, None, "lp@vela.example"),
+        ])
+    conn.commit()
+    conn.close()
+
+
+def test_collect():
+    conn = _conn()
+    groups = digest_builder.collect_user_activity(conn, SINCE, UNTIL)
+    conn.close()
+    check(len(groups) == 2, f"two active users with activity (grant, jk), got {len(groups)}")
+    by_user = {g["user_id"]: g for g in groups}
+    check("u3" not in by_user, "inactive user (u3) excluded")
+
+    grant = by_user.get("u1")
+    if not grant:
+        FAILS.append("grant group missing"); return
+    ids = set(e["email_id"] for e in grant["emails"])
+    check(ids == {"e1", "e2", "e4", "e5", "e10"},
+          f"grant has e1,e2,e4,e5,e10 (e6 out-of-window, e7 tombstoned, e9 unmatched excluded), got {sorted(ids)}")
+    check(grant["sent"] == 1 and grant["received"] == 4, f"grant 1 sent / 4 received, got {grant['sent']}/{grant['received']}")
+    e1 = next(e for e in grant["emails"] if e["email_id"] == "e1")
+    check(e1["direction"] == "sent", "e1 direction sent")
+    check(e1["investors"] == ["Harbor & Vine"], f"e1 investor = grid name, got {e1['investors']}")
+    e4 = next(e for e in grant["emails"] if e["email_id"] == "e4")
+    check(e4["investors"] == ["Jane Roe"], f"e4 investor = contact fallback name, got {e4['investors']}")
+    e5 = next(e for e in grant["emails"] if e["email_id"] == "e5")
+    check(e5["investors"] == ["ir@summitcap.example"], f"e5 investor = address (deleted org skipped), got {e5['investors']}")
+
+    jk = by_user.get("u2")
+    check(jk and jk["emails"][0]["investors"] == ["Summit Fund"], "jk e3 investor = org name")
+
+
+def test_investor():
+    conn = _conn()
+    inv = digest_builder.collect_investor_activity(conn, SINCE, UNTIL)
+    conn.close()
+    by_name = {g["name"]: g for g in inv}
+    # Harbor & Vine, Brightwater, Vela Partners, Summit Fund, Jane Roe, ir@summitcap.example
+    check(len(inv) == 6, f"six investors with activity, got {len(inv)}: {sorted(by_name)}")
+
+    hv = by_name.get("Harbor & Vine")
+    check(hv and hv["outbound"] == 1 and hv["inbound"] == 0, f"Harbor & Vine = 1 out / 0 in, got {hv}")
+    check(hv and hv["emails"][0]["members"] == ["Grant Gilliam"], f"outbound attributed to sender, got {hv and hv['emails'][0]['members']}")
+
+    bw = by_name.get("Brightwater Capital")
+    check(bw and bw["inbound"] == 1 and bw["outbound"] == 0, f"Brightwater = 1 in / 0 out, got {bw}")
+
+    # e10 was received by TWO mailboxes (grant + jk) -> dedup to one inbound email
+    vela = by_name.get("Vela Partners")
+    check(vela and vela["total"] == 1 and vela["inbound"] == 1,
+          f"Vela inbound deduped across 2 sightings -> 1, got {vela}")
+
+
+def test_build_and_empty():
+    conn = _conn()
+    stub = lambda prompt, system=None, max_tokens=220: "Grant worked with Harbor & Vine on Fund III."
+    d = digest_builder.build_digest(conn, SINCE, UNTIL, chat_fn=stub)
+    check(d["has_activity"] is True, "build_digest has_activity True when there is activity")
+    check(d["user_count"] == 2 and d["email_count"] == 7 and d["investor_count"] == 6,
+          f"counts: 2 users / 7 emails / 6 investors, got {d['user_count']}/{d['email_count']}/{d['investor_count']}")
+    body = d["body"]
+    check("Daily Activity Digest" in body, "body has title")
+    check("BY TEAM MEMBER" in body and "BY INVESTOR" in body, "body has both sections")
+    check("Grant Gilliam" in body and "Jonathan K" in body, "body names both active users")
+    check("Harbor & Vine" in body and "Brightwater Capital" in body and "Vela Partners" in body,
+          "investor section lists investors")
+    check("Grant worked with Harbor & Vine on Fund III." in body, "body includes the local narrative")
+
+    empty = digest_builder.build_digest(conn, "2030-01-01T00:00:00Z", "2030-01-02T00:00:00Z", chat_fn=stub)
+    check(empty["has_activity"] is False, "empty window -> has_activity False")
+    check("No tracked email activity" in empty["body"], "empty window -> 'no activity' note (always-send)")
+    check("BY INVESTOR" not in empty["body"], "empty window -> no section headers")
+    conn.close()
+
+
+def test_reminders_due():
+    """The reminders-due section: overdue + due-today only (future / done / soft-deleted
+    excluded), rendered even on an empty email window. Creates + drops the reminders table
+    so the rest of the suite still exercises the table-absent path."""
+    from datetime import date, timedelta
+    conn = _conn()
+    conn.execute("""CREATE TABLE reminders (id TEXT PRIMARY KEY, investor_id TEXT,
+        investor_name TEXT, contact_id TEXT, title TEXT, details TEXT, due_date TEXT,
+        status TEXT DEFAULT 'open', snoozed_until TEXT, assignee_id TEXT, created_by TEXT,
+        source TEXT, completed_at TEXT, created_at TEXT, updated_at TEXT, deleted_at TEXT)""")
+    today = date.today().isoformat()
+    yest = (date.today() - timedelta(days=1)).isoformat()
+    future = (date.today() + timedelta(days=30)).isoformat()
+    conn.executemany(
+        "INSERT INTO reminders (id,investor_name,title,due_date,status,assignee_id,deleted_at) "
+        "VALUES (?,?,?,?,?,?,?)", [
+            ("r1", "Harbor & Vine", "Send wire instructions", yest, "open", "u1", None),       # overdue
+            ("r2", "Brightwater Capital", "Call about allocation", today, "open", None, None),  # due today
+            ("r3", "Vela Partners", "Quarterly touch", future, "open", "u1", None),            # future -> hidden
+            ("r4", "Gone LP", "Done already", yest, "done", "u1", None),                        # done -> hidden
+            ("r5", "Deleted LP", "Tombstoned", yest, "open", "u1", "2026-06-01T00:00:00Z"),     # deleted -> hidden
+        ])
+    conn.commit()
+
+    due = digest_builder.collect_due_reminders(conn, today)
+    titles = {r["title"] for r in due}
+    check(titles == {"Send wire instructions", "Call about allocation"},
+          f"due collector = overdue + due-today only (got {titles})")
+    overdue = [r for r in due if r["overdue"]]
+    check(len(overdue) == 1 and overdue[0]["title"] == "Send wire instructions", "overdue flagged")
+
+    stub = lambda prompt, system=None, max_tokens=220: "narrative"
+    d = digest_builder.build_digest(conn, SINCE, UNTIL, chat_fn=stub)
+    check(d["reminder_count"] == 2, f"reminder_count = 2 (got {d['reminder_count']})")
+    check("REMINDERS DUE (2)" in d["body"], "body has reminders section header")
+    check("Overdue (1):" in d["body"] and "Due today (1):" in d["body"], "body splits overdue / due today")
+    check("Harbor & Vine — Send wire instructions" in d["body"]
+          and "[Grant Gilliam]" in d["body"], "reminder line shows investor + title + resolved assignee")
+    check("Quarterly touch" not in d["body"], "future reminder excluded from due section")
+
+    empty = digest_builder.build_digest(conn, "2030-01-01T00:00:00Z", "2030-01-02T00:00:00Z", chat_fn=stub)
+    check("No tracked email activity" in empty["body"] and "REMINDERS DUE (2)" in empty["body"],
+          "reminders render even on an empty email window (current-state addendum)")
+
+    conn.execute("DROP TABLE reminders")
+    conn.commit()
+    conn.close()
+
+
+def test_policy():
+    conn = _conn()
+    # No DB row yet: CRM_DIGEST_ENABLED=1 (set at import) seeds enabled; hour defaults 18.
+    pol = digest_builder.load_digest_policy(conn)
+    check(pol["enabled"] is True and pol["send_hour"] == 18, f"env seed -> enabled, hour 18, got {pol}")
+    # A DB row wins over the env seed (the admin-panel control).
+    conn.execute("INSERT OR REPLACE INTO app_settings (key,value_json,updated_at) VALUES (?,?,?)",
+                 (digest_builder.DIGEST_POLICY_KEY, json.dumps({"enabled": False, "send_hour": 9}), "x"))
+    conn.commit()
+    pol2 = digest_builder.load_digest_policy(conn)
+    check(pol2["enabled"] is False and pol2["send_hour"] == 9, f"DB policy wins over env, got {pol2}")
+    conn.execute("DELETE FROM app_settings WHERE key = ?", (digest_builder.DIGEST_POLICY_KEY,))
+    conn.commit()
+    conn.close()
+
+
+def test_summary_fallback():
+    grp = {"user_id": "u1", "full_name": "Grant Gilliam", "username": "grant",
+           "emails": [{"direction": "sent", "subject": "x", "investors": ["Harbor & Vine"], "text": "hi"}],
+           "investors": ["Harbor & Vine"], "sent": 1, "received": 0, "total": 1}
+    def boom(*a, **k):
+        raise RuntimeError("spark down")
+    out = digest_builder.summarize_user_day(grp, chat_fn=boom)
+    check("Grant Gilliam" in out and "1 sent" in out and "unavailable" in out.lower(),
+          f"fallback narrative on chat error, got: {out}")
+
+
+def test_scheduler_guards():
+    sent_calls = []
+    build_fn = lambda conn, since, until: {"subject": "S", "body": "B",
+                                           "has_activity": True, "user_count": 1, "email_count": 2}
+    def send_fn(conn, to_addrs, subject, body, sender_email=None):
+        sent_calls.append(list(to_addrs))
+        return {"transport": "stub"}
+
+    factory = _conn
+    utc = datetime(2026, 6, 18, 1, 0, tzinfo=timezone.utc)
+
+    # Before the send hour (09:00 local < 18:00) -> no send
+    r = digest_scheduler.maybe_send_digest(factory, now_local=datetime(2026, 6, 18, 9, 0),
+                                           now_utc=utc, build_fn=build_fn, send_fn=send_fn)
+    check(r["status"] == "before_send_hour" and not sent_calls, f"before send hour -> skip, got {r}")
+
+    # At/after the send hour -> sends once, only to the active admin (grant)
+    r = digest_scheduler.maybe_send_digest(factory, now_local=datetime(2026, 6, 18, 19, 0),
+                                           now_utc=utc, build_fn=build_fn, send_fn=send_fn)
+    check(r["status"] == "sent" and len(sent_calls) == 1, f"after send hour -> sends, got {r}")
+    check(sent_calls and sent_calls[-1] == ["grant@ten31.xyz"], f"recipients = active admins only, got {sent_calls[-1]}")
+    # The window cursor must advance to the send time so a missed day rolls forward
+    # (since, now] — not be left unset/stale.
+    conn = _conn()
+    cursor_at = digest_scheduler._get_setting(conn, digest_scheduler._LAST_AT_KEY)
+    conn.close()
+    check(cursor_at == digest_scheduler._utc_iso(utc),
+          f"window cursor advanced to send time, got {cursor_at}")
+
+    # Same local day again -> suppressed (once-per-day guard)
+    r = digest_scheduler.maybe_send_digest(factory, now_local=datetime(2026, 6, 18, 20, 0),
+                                           now_utc=utc, build_fn=build_fn, send_fn=send_fn)
+    check(r["status"] == "already_sent_today" and len(sent_calls) == 1, f"second send same day -> skip, got {r}")
+
+    # force=True ignores the hour + once-per-day guard, and does NOT touch the cursor
+    conn = _conn()
+    before = digest_scheduler._get_setting(conn, digest_scheduler._LAST_DATE_KEY)
+    conn.close()
+    r = digest_scheduler.maybe_send_digest(factory, force=True, now_local=datetime(2026, 6, 18, 3, 0),
+                                           now_utc=utc, build_fn=build_fn, send_fn=send_fn)
+    check(r["status"] == "sent" and len(sent_calls) == 2, f"force sends regardless of guards, got {r}")
+    conn = _conn()
+    after = digest_scheduler._get_setting(conn, digest_scheduler._LAST_DATE_KEY)
+    conn.close()
+    check(before == after, "force send does not advance the daily cursor")
+
+    # DB policy disabled -> daily path skips even past the hour; force still sends.
+    conn = _conn()
+    digest_scheduler._set_setting(conn, digest_builder.DIGEST_POLICY_KEY, {"enabled": False, "send_hour": 18})
+    conn.commit()
+    conn.close()
+    r = digest_scheduler.maybe_send_digest(factory, now_local=datetime(2026, 6, 19, 19, 0),
+                                           now_utc=utc, build_fn=build_fn, send_fn=send_fn)
+    check(r["status"] == "disabled" and len(sent_calls) == 2, f"DB-disabled policy skips daily send, got {r}")
+    r = digest_scheduler.maybe_send_digest(factory, force=True, now_local=datetime(2026, 6, 19, 2, 0),
+                                           now_utc=utc, build_fn=build_fn, send_fn=send_fn)
+    check(r["status"] == "sent" and len(sent_calls) == 3, f"force overrides disabled policy, got {r}")
+    conn = _conn()
+    conn.execute("DELETE FROM app_settings WHERE key = ?", (digest_builder.DIGEST_POLICY_KEY,))
+    conn.commit()
+    conn.close()
+
+
+def test_window_resolver():
+    from datetime import timedelta
+    nu = datetime(2026, 6, 16, 15, 0, tzinfo=timezone.utc)
+    nl = datetime(2026, 6, 16, 8, 0, tzinfo=timezone(timedelta(hours=-7)))  # PDT
+
+    s, u = digest_builder.resolve_digest_window(now_utc=nu, now_local=nl)
+    check((s, u) == ("2026-06-15T15:00:00Z", "2026-06-16T15:00:00Z"), f"default = last 24h, got {(s,u)}")
+
+    s, u = digest_builder.resolve_digest_window(hours=48, now_utc=nu, now_local=nl)
+    check(s == "2026-06-14T15:00:00Z", f"hours=48 lookback, got {s}")
+
+    # since = a local calendar date -> that day's LOCAL midnight, expressed in UTC
+    s, u = digest_builder.resolve_digest_window(since="2026-06-01", now_utc=nu, now_local=nl)
+    check(s == "2026-06-01T07:00:00Z", f"since-date -> local midnight in UTC, got {s}")
+
+    # a since older than the cap clamps to MAX_WINDOW_DAYS (response echoes real window)
+    s, u = digest_builder.resolve_digest_window(since="2025-01-01", now_utc=nu, now_local=nl)
+    check(s == (nu - timedelta(days=digest_builder.MAX_WINDOW_DAYS)).strftime("%Y-%m-%dT%H:%M:%SZ"),
+          f"over-cap since clamps to {digest_builder.MAX_WINDOW_DAYS}d, got {s}")
+
+    # since wins over hours when both supplied
+    s, u = digest_builder.resolve_digest_window(hours=1, since="2026-06-10", now_utc=nu, now_local=nl)
+    check(s.startswith("2026-06-10"), f"since wins over hours, got {s}")
+
+    # same-day boundary: since = today's local date, now later in the day -> valid
+    # window (local midnight is strictly before now), not a "start must be before now" raise
+    s, u = digest_builder.resolve_digest_window(since="2026-06-16", now_utc=nu, now_local=nl)
+    check(s == "2026-06-16T07:00:00Z" and u == "2026-06-16T15:00:00Z",
+          f"since=today -> [local midnight, now], got {(s, u)}")
+
+    for bad in [dict(hours=0), dict(hours="abc"), dict(since="nope"), dict(since="2027-01-01")]:
+        try:
+            digest_builder.resolve_digest_window(now_utc=nu, now_local=nl, **bad)
+            check(False, f"bad input {bad} should raise")
+        except ValueError:
+            check(True, f"bad input rejected: {bad}")
+
+
+def test_send_digest_window():
+    sent = []
+    build_fn = lambda conn, since, until: {"subject": "S", "body": f"{since}|{until}",
+                                           "has_activity": True, "user_count": 1,
+                                           "email_count": 2, "investor_count": 1}
+    def send_fn(conn, to_addrs, subject, body, sender_email=None):
+        sent.append((list(to_addrs), body))
+        return {"transport": "stub"}
+
+    conn = _conn()
+    before = digest_scheduler._get_setting(conn, digest_scheduler._LAST_AT_KEY)
+    conn.close()
+
+    r = digest_scheduler.send_digest_window(_conn, since_iso="2026-05-01T00:00:00Z",
+                                            until_iso="2026-06-16T00:00:00Z",
+                                            build_fn=build_fn, send_fn=send_fn)
+    check(r["status"] == "sent" and r["window"] == ["2026-05-01T00:00:00Z", "2026-06-16T00:00:00Z"],
+          f"windowed send returns its window, got {r}")
+    check(sent and sent[-1][0] == ["grant@ten31.xyz"], f"windowed send -> active admins only, got {sent}")
+
+    conn = _conn()
+    after = digest_scheduler._get_setting(conn, digest_scheduler._LAST_AT_KEY)
+    conn.close()
+    check(before == after, "windowed manual send does not advance the daily cursor")
+
+
+def main():
+    setup()
+    print("collect_user_activity:");  test_collect()
+    print("collect_investor_activity:"); test_investor()
+    print("build_digest + empty:");   test_build_and_empty()
+    print("reminders due:");          test_reminders_due()
+    print("summary fallback:");       test_summary_fallback()
+    print("digest policy:");          test_policy()
+    print("window resolver:");        test_window_resolver()
+    print("windowed manual send:");   test_send_digest_window()
+    print("scheduler guards:");       test_scheduler_guards()
+    if FAILS:
+        print(f"\nFAILED ({len(FAILS)})")
+        for f in FAILS:
+            print("  - " + f)
+        sys.exit(1)
+    print("\nALL PASS (digest builder + scheduler)")
+
+
+if __name__ == "__main__":
+    main()

backend/test_email_activity.py

@@ -33,7 +33,7 @@ def setup():
     conn.executescript("""
         CREATE TABLE app_settings (key TEXT PRIMARY KEY, value_json TEXT, updated_at TEXT);
         CREATE TABLE email_accounts (id TEXT, email_address TEXT, sync_enabled INT DEFAULT 1, sync_status TEXT, backfill_complete INT);
-        CREATE TABLE emails (id TEXT PRIMARY KEY, subject TEXT, body_text TEXT, snippet TEXT, from_email TEXT, sent_at TEXT, is_matched INT, match_status TEXT);
+        CREATE TABLE emails (id TEXT PRIMARY KEY, subject TEXT, body_text TEXT, snippet TEXT, from_name TEXT, from_email TEXT, sent_at TEXT, is_matched INT, match_status TEXT);
         CREATE TABLE email_investor_links (id TEXT, email_id TEXT, fundraising_investor_id TEXT, organization_id TEXT, contact_id TEXT, match_confidence REAL);
         CREATE TABLE email_activity_proposals (id TEXT PRIMARY KEY, email_id TEXT UNIQUE, investor_id TEXT, investor_name TEXT,
             direction TEXT, summary TEXT, proposed_note TEXT, email_subject TEXT, email_date TEXT, status TEXT DEFAULT 'pending',
@@ -51,10 +51,10 @@ def setup():
     grid = {"columns": [], "rows": [{"id": "inv1", "investor_name": "Harbor & Vine", "notes": "existing note"}]}
     conn.execute("INSERT INTO fundraising_state (id,grid_json,views_json,version) VALUES ('main',?,?,1)", (json.dumps(grid), "[]"))
     # e1 sent (from us), e2 received, both after cutoff; e3 before cutoff (excluded)
-    conn.executemany("INSERT INTO emails (id,subject,body_text,from_email,sent_at,is_matched,match_status) VALUES (?,?,?,?,?,1,'matched')", [
+    conn.executemany("INSERT INTO emails (id,subject,body_text,from_name,from_email,sent_at,is_matched,match_status) VALUES (?,?,?,?,?,?,1,'matched')", [
-        ("e1", "Fund III", "Here is the update", "grant@ten31.xyz", "2026-06-01T10:00:00"),
+        ("e1", "Fund III", "Here is the update", "Grant", "grant@ten31.xyz", "2026-06-01T10:00:00"),
-        ("e2", "Re: Fund III", "Thanks, a question", "lp@harborvine.example", "2026-06-02T10:00:00"),
+        ("e2", "Re: Fund III", "Thanks, a question", "Harbor LP", "lp@harborvine.example", "2026-06-02T10:00:00"),
-        ("e3", "Old", "ancient", "lp@harborvine.example", "2025-01-01T10:00:00"),
+        ("e3", "Old", "ancient", "Harbor LP", "lp@harborvine.example", "2025-01-01T10:00:00"),
     ])
     conn.executemany("INSERT INTO email_investor_links (id,email_id,fundraising_investor_id,match_confidence) VALUES (?,?, 'inv1', 1.0)",
                      [("l1", "e1"), ("l2", "e2"), ("l3", "e3")])
@@ -77,7 +77,9 @@ def main():
     dirs = sorted(p["direction"] for p in props)
     check(dirs == ["received", "sent"], f"directions sent+received, got {dirs}")
     e1 = next(p for p in props if p["email_id"] == "e1")
-    check(e1["direction"] == "sent" and "Sent" in e1["proposed_note"], "e1 (from us) is 'sent'")
+    check(e1["direction"] == "sent" and "Grant emailed Harbor & Vine" in e1["proposed_note"], "e1 (from us) names sender + investor")
+    e2 = next(p for p in props if p["email_id"] == "e2")
+    check(e2["direction"] == "received" and "emailed the team" in e2["proposed_note"], "e2 (inbound) reads '<sender> emailed the team'")
     check("✉" in e1["proposed_note"] and "fundraising update" in e1["proposed_note"], "proposed note marked + has gist")
 
     # grid must be UNTOUCHED before approval

backend/test_email_proposal_matrix.py

@@ -0,0 +1,129 @@
+#!/usr/bin/env python3
+"""Test the Matrix review-bot bridge for email-activity proposals (Features 2/3):
+the bot work-lists (to_post / open / to_close), the Matrix side-row mark helpers, and an
+in-thread (source='matrix') decision that closes the thread — plus the bot-or-admin role gate.
+Synthetic data only (guardrail #9). The local model is stubbed.
+Run: cd backend && python3 test_email_proposal_matrix.py
+"""
+import json
+import os
+import sqlite3
+import sys
+import tempfile
+
+os.environ["CRM_DB_PATH"] = os.path.join(tempfile.mkdtemp(), "crm.db")
+os.environ.setdefault("CRM_DATA_DIR", os.path.dirname(os.environ["CRM_DB_PATH"]))
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import server  # noqa: E402
+
+server._summarize_email_gist = lambda subject, body: "fundraising update; proposed a call"
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+def setup():
+    conn = sqlite3.connect(os.environ["CRM_DB_PATH"])
+    conn.row_factory = sqlite3.Row
+    conn.executescript("""
+        CREATE TABLE app_settings (key TEXT PRIMARY KEY, value_json TEXT, updated_at TEXT);
+        CREATE TABLE email_accounts (id TEXT, email_address TEXT, sync_enabled INT DEFAULT 1, sync_status TEXT, backfill_complete INT);
+        CREATE TABLE emails (id TEXT PRIMARY KEY, subject TEXT, body_text TEXT, snippet TEXT, from_name TEXT, from_email TEXT, sent_at TEXT, is_matched INT, match_status TEXT);
+        CREATE TABLE email_investor_links (id TEXT, email_id TEXT, fundraising_investor_id TEXT, organization_id TEXT, contact_id TEXT, match_confidence REAL);
+        CREATE TABLE email_activity_proposals (id TEXT PRIMARY KEY, email_id TEXT UNIQUE, investor_id TEXT, investor_name TEXT,
+            direction TEXT, summary TEXT, proposed_note TEXT, email_subject TEXT, email_date TEXT, status TEXT DEFAULT 'pending',
+            decided_by TEXT, decided_at TEXT, final_note TEXT, created_at TEXT);
+        CREATE TABLE email_proposal_matrix (proposal_id TEXT PRIMARY KEY, event_id TEXT, posted_at TEXT, closed_at TEXT, created_at TEXT);
+        CREATE TABLE users (id TEXT PRIMARY KEY, username TEXT);
+        CREATE TABLE fundraising_investors (id TEXT PRIMARY KEY, investor_name TEXT, notes TEXT);
+        CREATE TABLE fundraising_state (id TEXT PRIMARY KEY, grid_json TEXT, views_json TEXT, version INT,
+            updated_by TEXT REFERENCES users(id), updated_at TEXT);
+        CREATE TABLE interaction_log (id TEXT PRIMARY KEY, ts TEXT, actor_type TEXT, actor_id TEXT, action TEXT, target_type TEXT, target_id TEXT, payload TEXT, source TEXT, created_at TEXT);
+    """)
+    conn.execute("INSERT INTO users (id,username) VALUES ('user-1','grant')")
+    conn.execute("INSERT INTO app_settings VALUES ('email_activity_since', ?, ?)", (json.dumps("2026-01-01T00:00:00"), "x"))
+    conn.execute("INSERT INTO email_accounts (id,email_address) VALUES ('a','grant@ten31.xyz')")
+    conn.execute("INSERT INTO fundraising_investors (id,investor_name,notes) VALUES ('inv1','Harbor & Vine','existing note')")
+    grid = {"columns": [], "rows": [{"id": "inv1", "investor_name": "Harbor & Vine", "notes": "existing note"}]}
+    conn.execute("INSERT INTO fundraising_state (id,grid_json,views_json,version) VALUES ('main',?,?,1)", (json.dumps(grid), "[]"))
+    conn.executemany("INSERT INTO emails (id,subject,body_text,snippet,from_name,from_email,sent_at,is_matched,match_status) VALUES (?,?,?,?,?,?,?,1,'matched')", [
+        ("e1", "Fund III", "Here is the update", "the quarterly update is attached", "Grant", "grant@ten31.xyz", "2026-06-01T10:00:00"),
+        ("e2", "Re: Fund III", "Thanks, a question", "thanks — one question on terms", "LP Contact", "lp@harborvine.example", "2026-06-02T10:00:00"),
+    ])
+    conn.executemany("INSERT INTO email_investor_links (id,email_id,fundraising_investor_id,match_confidence) VALUES (?,?, 'inv1', 1.0)",
+                     [("l1", "e1"), ("l2", "e2")])
+    conn.commit()
+    conn.close()
+
+
+def main():
+    setup()
+
+    # role gate: bot passes the agent gate but is NOT an admin; member passes neither.
+    check(server.require_bot_or_admin({"role": "bot"}), "bot passes require_bot_or_admin")
+    check(server.require_bot_or_admin({"role": "admin"}), "admin passes require_bot_or_admin")
+    check(not server.require_bot_or_admin({"role": "member"}), "member does NOT pass require_bot_or_admin")
+    check(not server.require_admin({"role": "bot"}), "bot is NOT an admin (no user-mgmt/settings reach)")
+
+    check(server.propose_email_activity_notes().get("proposed") == 2, "drafts 2 proposals")
+    conn = server.get_db()
+    props = server.list_email_activity_proposals(conn, status="pending")
+    by_email = {p["email_id"]: p for p in props}
+    p_a, p_b = by_email["e1"], by_email["e2"]
+
+    # Both are pending + un-posted → both in to_post; card carries from/snippet/note context.
+    lists = server.list_bot_email_proposals(conn)
+    check(len(lists["to_post"]) == 2 and not lists["open"] and not lists["to_close"], "both proposals queued to_post")
+    card = next(it for it in lists["to_post"] if it["id"] == p_a["id"])
+    check(card.get("from_name") == "Grant" and "quarterly update" in (card.get("snippet") or ""), "card carries from_name + snippet")
+    check("✉" in (card.get("proposed_note") or ""), "card carries the drafted note")
+
+    # Post p_a to Matrix → it leaves to_post and becomes an open thread (event id recorded).
+    server.mark_proposal_matrix_posted(conn, p_a["id"], "evtA")
+    lists = server.list_bot_email_proposals(conn)
+    check(len(lists["to_post"]) == 1 and lists["to_post"][0]["id"] == p_b["id"], "posting p_a leaves only p_b to_post")
+    check(len(lists["open"]) == 1 and lists["open"][0]["id"] == p_a["id"] and lists["open"][0]["event_id"] == "evtA",
+          "posted p_a is an open thread carrying its event id")
+
+    # Decide p_a IN-THREAD on Matrix (approve + close in one transaction).
+    r = server.decide_email_activity_proposal(conn, p_a["id"], "approve", "user-1", source="matrix", close_matrix=True)
+    check(r.get("status") == "approved" and r.get("placed_in_grid") is True, "matrix approve appends to the grid")
+    lists = server.list_bot_email_proposals(conn)
+    check(not any(it["id"] == p_a["id"] for it in lists["open"] + lists["to_close"]),
+          "matrix-decided proposal is closed (not re-announced via to_close)")
+    src = conn.execute("SELECT source FROM interaction_log WHERE action='email.activity_approved'").fetchone()["source"]
+    check(src == "matrix", "matrix decision is audited source='matrix'")
+
+    # Web-decide path: post p_b, then dismiss it on the WEB (default source, no close) → the bot
+    # must see it in to_close so it can announce the web decision in-thread, then close.
+    server.mark_proposal_matrix_posted(conn, p_b["id"], "evtB")
+    server.decide_email_activity_proposal(conn, p_b["id"], "dismiss", "user-1")  # web path: source crm_ui, no close
+    lists = server.list_bot_email_proposals(conn)
+    check(len(lists["to_close"]) == 1 and lists["to_close"][0]["id"] == p_b["id"] and lists["to_close"][0]["status"] == "dismissed",
+          "web-decided open thread surfaces in to_close")
+    src2 = conn.execute("SELECT source FROM interaction_log WHERE action='email.activity_dismissed'").fetchone()["source"]
+    check(src2 == "crm_ui", "web decision is audited source='crm_ui'")
+
+    server.mark_proposal_matrix_closed(conn, p_b["id"])
+    lists = server.list_bot_email_proposals(conn)
+    check(not lists["to_close"] and not lists["open"], "closing the thread clears the work-lists")
+
+    # Marking a non-existent proposal is a clean not_found, not a crash.
+    check(server.mark_proposal_matrix_posted(conn, "nope", "evtX").get("error") == "not_found", "mark posted on unknown id -> not_found")
+    conn.close()
+
+    if FAILS:
+        print(f"\nFAILED ({len(FAILS)})")
+        for f in FAILS:
+            print("  - " + f)
+        sys.exit(1)
+    print("\nALL PASS (email-proposal Matrix bridge)")
+
+
+if __name__ == "__main__":
+    main()

backend/test_fundraising_update_row.py

@@ -0,0 +1,227 @@
+#!/usr/bin/env python3
+"""Integration test for the version-safe single-row update endpoint (P3b).
+
+POST /api/fundraising/update-row edits ONE investor row's name and/or contact pills by
+reading the canonical grid blob fresh server-side and mutating only the target row — never
+accepting a whole-grid payload (BRIEF §3a), so it can't clobber concurrent edits to other
+rows. This boots the REAL server against a throwaway DB, seeds a two-row grid, then drives
+the live HTTP endpoint to assert:
+  * rename + pill add/edit persist into the blob and bump the version;
+  * removing a pill drops it from the row + fundraising_contacts, but the classic contacts
+    directory entry is NOT hard-deleted (soft-delete-only convention);
+  * preserved pill fields (title/city/linkedin) survive an edit that only touches the name;
+  * the OTHER grid row is untouched (no whole-grid clobber);
+  * guards: missing row_id -> 400, unknown row_id -> 404, blank name -> 400,
+    no-op body (neither field) -> 400.
+Synthetic data only (guardrail #9).
+
+Run:  cd backend && python3 test_fundraising_update_row.py
+"""
+import http.client
+import json
+import os
+import sqlite3
+import sys
+import tempfile
+import threading
+from http.server import ThreadingHTTPServer
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import server  # noqa: E402
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Quiet(server.CRMHandler):
+    def log_message(self, *a):
+        pass
+
+
+def _post(port, path, token, payload):
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    conn.request("POST", path, body=json.dumps(payload),
+                 headers={"Authorization": "Bearer " + token, "Content-Type": "application/json"})
+    resp = conn.getresponse()
+    raw = resp.read().decode("utf-8", "replace")
+    conn.close()
+    data = json.loads(raw) if raw else None
+    return resp.status, data
+
+
+GRID = {
+    "columns": [
+        {"id": "investor_name", "label": "Investor Name", "type": "text"},
+        {"id": "contacts", "label": "Contacts", "type": "contacts"},
+        {"id": "notes", "label": "Notes", "type": "text"},
+    ],
+    "rows": [
+        {"id": "row-1", "investor_name": "Acme Capital", "notes": "",
+         "contacts": [
+             {"name": "Jane Doe", "email": "jane@acme.com", "title": "Partner",
+              "city": "Austin", "state": "TX", "country": "USA", "location_query": "Austin",
+              "linkedin_url": "https://linkedin.com/in/janedoe"},
+             {"name": "Bob Roe", "email": "bob@acme.com", "title": ""},
+         ]},
+        {"id": "row-2", "investor_name": "Beacon Fund", "notes": "untouched",
+         "contacts": [{"name": "Carl Vane", "email": "carl@beacon.com"}]},
+    ],
+}
+
+
+def seed():
+    c = sqlite3.connect(os.environ["CRM_DB_PATH"])
+    c.row_factory = sqlite3.Row
+    c.execute("INSERT INTO users (id,username,email,password_hash,full_name,role,is_active) "
+              "VALUES ('u1','grant','grant@ten31.example','x','Grant','admin',1)")
+    c.execute("INSERT INTO fundraising_state (id, grid_json, views_json, version, updated_by) "
+              "VALUES ('main', ?, '[]', 1, 'u1') "
+              "ON CONFLICT(id) DO UPDATE SET grid_json=excluded.grid_json, views_json='[]', version=1",
+              (json.dumps(GRID),))
+    server.sync_fundraising_relational(c, server.sanitize_fundraising_grid(GRID), [], actor_user_id="u1")
+    c.commit()
+    c.close()
+
+
+def blob_rows():
+    c = sqlite3.connect(os.environ["CRM_DB_PATH"])
+    c.row_factory = sqlite3.Row
+    row = c.execute("SELECT grid_json, version FROM fundraising_state WHERE id='main'").fetchone()
+    c.close()
+    grid = json.loads(row["grid_json"])
+    by_id = {r["id"]: r for r in grid["rows"]}
+    return by_id, int(row["version"])
+
+
+def main():
+    server.init_db()
+    seed()
+    token = server.create_token("u1", "grant", "admin")
+
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), _Quiet)
+    port = httpd.server_address[1]
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        # ── rename + add a third contact, edit Bob's title, keep Jane unchanged ──
+        print("\n[rename + pill add/edit]")
+        new_contacts = [
+            # Jane: only her name is "re-sent"; the client preserves her other fields by spread.
+            {"name": "Jane Doe", "email": "jane@acme.com", "title": "Partner",
+             "city": "Austin", "state": "TX", "country": "USA", "location_query": "Austin",
+             "linkedin_url": "https://linkedin.com/in/janedoe"},
+            {"name": "Bob Roe", "email": "bob@acme.com", "title": "Principal"},   # edited title
+            {"name": "Dana Fox", "email": "dana@acme.com", "title": "Analyst"},   # added
+        ]
+        st, data = _post(port, "/api/fundraising/update-row", token,
+                         {"row_id": "row-1", "investor_name": "Acme Capital LLC", "contacts": new_contacts})
+        check(st == 200, f"update-row -> 200 (got {st})")
+        by_id, version = blob_rows()
+        check(version == 2, f"version bumped 1 -> 2 (got {version})")
+        r1 = by_id.get("row-1", {})
+        check(r1.get("investor_name") == "Acme Capital LLC", f"name renamed in blob (got {r1.get('investor_name')!r})")
+        names = [c.get("name") for c in r1.get("contacts", [])]
+        check(names == ["Jane Doe", "Bob Roe", "Dana Fox"], f"three pills in order (got {names})")
+        bob = next((c for c in r1["contacts"] if c["name"] == "Bob Roe"), {})
+        check(bob.get("title") == "Principal", f"Bob's title edited (got {bob.get('title')!r})")
+        jane = next((c for c in r1["contacts"] if c["name"] == "Jane Doe"), {})
+        check(jane.get("linkedin_url") == "https://linkedin.com/in/janedoe" and jane.get("city") == "Austin",
+              f"Jane's preserved fields survived (got {jane})")
+
+        # ── the OTHER row is byte-for-byte untouched (no whole-grid clobber) ──
+        print("\n[other row untouched]")
+        r2 = by_id.get("row-2", {})
+        check(r2.get("investor_name") == "Beacon Fund" and r2.get("notes") == "untouched"
+              and [c.get("name") for c in r2.get("contacts", [])] == ["Carl Vane"],
+              f"row-2 unchanged (got {r2})")
+
+        # ── relational sync: classic contacts directory now has Dana; Dana also in fundraising_contacts ──
+        print("\n[relational + classic-contacts sync]")
+        c = sqlite3.connect(os.environ["CRM_DB_PATH"])
+        c.row_factory = sqlite3.Row
+        dana = c.execute("SELECT id, deleted_at FROM contacts WHERE lower(email)='dana@acme.com'").fetchone()
+        check(bool(dana), "added contact Dana propagated to classic contacts directory")
+        inv = c.execute("SELECT id FROM fundraising_investors WHERE source_row_id='row-1'").fetchone()
+        fc_names = {r["full_name"] for r in c.execute(
+            "SELECT full_name FROM fundraising_contacts WHERE investor_id=?", (inv["id"],)).fetchall()}
+        check(fc_names == {"Jane Doe", "Bob Roe", "Dana Fox"}, f"fundraising_contacts mirrors the three pills (got {fc_names})")
+        c.close()
+
+        # ── remove Bob: pill drops + fundraising_contacts drops, but classic contact NOT hard-deleted ──
+        print("\n[remove pill is soft on the classic directory]")
+        st, _ = _post(port, "/api/fundraising/update-row", token, {"row_id": "row-1", "contacts": [
+            {"name": "Jane Doe", "email": "jane@acme.com", "title": "Partner"},
+            {"name": "Dana Fox", "email": "dana@acme.com", "title": "Analyst"},
+        ]})
+        check(st == 200, f"remove-pill update -> 200 (got {st})")
+        by_id, version = blob_rows()
+        names = [c.get("name") for c in by_id["row-1"].get("contacts", [])]
+        check(names == ["Jane Doe", "Dana Fox"], f"Bob removed from the row (got {names})")
+        check(version == 3, f"version bumped 2 -> 3 (got {version})")
+        c = sqlite3.connect(os.environ["CRM_DB_PATH"])
+        c.row_factory = sqlite3.Row
+        inv = c.execute("SELECT id FROM fundraising_investors WHERE source_row_id='row-1'").fetchone()
+        fc_names = {r["full_name"] for r in c.execute(
+            "SELECT full_name FROM fundraising_contacts WHERE investor_id=?", (inv["id"],)).fetchall()}
+        check("Bob Roe" not in fc_names, f"Bob dropped from fundraising_contacts (got {fc_names})")
+        bob_classic = c.execute("SELECT id FROM contacts WHERE lower(email)='bob@acme.com'").fetchone()
+        check(bool(bob_classic), "removing a pill does NOT hard-delete the classic contacts row (soft-delete only)")
+        c.close()
+
+        # ── name-only update (no contacts key) leaves the pill list intact ──
+        print("\n[name-only update preserves contacts]")
+        st, _ = _post(port, "/api/fundraising/update-row", token, {"row_id": "row-1", "investor_name": "Acme Capital"})
+        check(st == 200, f"name-only update -> 200 (got {st})")
+        by_id, _ = blob_rows()
+        names = [c.get("name") for c in by_id["row-1"].get("contacts", [])]
+        check(by_id["row-1"].get("investor_name") == "Acme Capital", "name updated again")
+        check(names == ["Jane Doe", "Dana Fox"], f"contacts untouched by a name-only edit (got {names})")
+
+        # ── a name-only pill (no email) is KEPT; a fully-blank pill is dropped ──
+        # (locks _sanitize_fundraising_contacts's emptiness rule = name OR email, not AND.)
+        print("\n[name-only pill kept, blank pill dropped]")
+        st, _ = _post(port, "/api/fundraising/update-row", token, {"row_id": "row-1", "contacts": [
+            {"name": "Jane Doe", "email": "jane@acme.com"},
+            {"name": "Erin Pope", "email": ""},          # name only -> kept
+            {"name": "", "email": ""},                    # fully blank -> dropped
+            {"name": "  ", "title": "Ghost"},             # whitespace-only name, no email -> dropped
+        ]})
+        check(st == 200, f"name-only-pill update -> 200 (got {st})")
+        by_id, _ = blob_rows()
+        names = [c.get("name") for c in by_id["row-1"].get("contacts", [])]
+        check(names == ["Jane Doe", "Erin Pope"],
+              f"name-only pill kept, blank/whitespace pills dropped (got {names})")
+
+        # ── guards ──
+        print("\n[validation guards]")
+        st, _ = _post(port, "/api/fundraising/update-row", token, {"investor_name": "No Id"})
+        check(st == 400, f"missing row_id -> 400 (got {st})")
+        st, _ = _post(port, "/api/fundraising/update-row", token, {"row_id": "row-nope", "investor_name": "Ghost"})
+        check(st == 404, f"unknown row_id -> 404 (got {st})")
+        st, _ = _post(port, "/api/fundraising/update-row", token, {"row_id": "row-1", "investor_name": "   "})
+        check(st == 400, f"blank name -> 400 (got {st})")
+        st, _ = _post(port, "/api/fundraising/update-row", token, {"row_id": "row-1"})
+        check(st == 400, f"no-op body (neither name nor contacts) -> 400 (got {st})")
+        st, _ = _post(port, "/api/fundraising/update-row", token, {"row_id": "row-1", "contacts": "nope"})
+        check(st == 400, f"contacts wrong type -> 400 (got {st})")
+    finally:
+        httpd.shutdown()
+
+    print()
+    if FAILS:
+        print(f"FAILED ({len(FAILS)}):")
+        for f in FAILS:
+            print(f"  - {f}")
+        sys.exit(1)
+    print("ALL PASS (fundraising update-row: version-safe single-row name/pill edit)")
+
+
+if __name__ == "__main__":
+    main()

backend/test_gmail_send.py

@@ -0,0 +1,230 @@
+#!/usr/bin/env python3
+"""Tests for gmail_send (DWD Gmail-API send) + digest_mailer (transport routing).
+
+No network: the credential provider and urllib.request.urlopen are monkeypatched.
+Run directly or via backend/run_tests.py.
+"""
+import base64
+import io
+import json
+import os
+import sys
+import urllib.error
+import urllib.request
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+from email_integration import gmail_send
+import digest_mailer
+import smtp_send
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  ok  " if cond else "  XX  ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Tok:
+    def __init__(self, t):
+        self.token = t
+
+
+class _Provider:
+    def __init__(self):
+        self.calls = []
+
+    def access_token_for(self, email, scope):
+        self.calls.append((email, scope))
+        return _Tok("tok-" + email)
+
+
+class _Resp:
+    def __init__(self, payload):
+        self._p = payload
+
+    def read(self):
+        return json.dumps(self._p).encode()
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, *a):
+        return False
+
+
+def main():
+    # 1. _build_raw round-trips
+    raw = gmail_send._build_raw("a@x", ["b@x", "c@x"], "Subj", "Hello body")
+    decoded = base64.urlsafe_b64decode(raw).decode()
+    check("From: a@x" in decoded and "Subject: Subj" in decoded
+          and "b@x, c@x" in decoded and "Hello body" in decoded,
+          "_build_raw round-trips From/To/Subject/body")
+
+    # 2. send_via_gmail posts correctly (fake provider + fake urlopen)
+    captured = {}
+    orig_open = urllib.request.urlopen
+    orig_build = gmail_send._creds.build_provider
+    prov = _Provider()
+    gmail_send._creds.build_provider = lambda factory: prov
+
+    def fake_open(req, timeout=None):
+        captured["url"] = req.full_url
+        captured["auth"] = req.headers.get("Authorization")
+        captured["body"] = json.loads(req.data.decode())
+        return _Resp({"id": "msg123", "threadId": "thr123"})
+
+    urllib.request.urlopen = fake_open
+    try:
+        res = gmail_send.send_via_gmail("grant@ten31.xyz",
+                                        ["a@ten31.xyz", "b@ten31.xyz"], "S", "B")
+    finally:
+        urllib.request.urlopen = orig_open
+        gmail_send._creds.build_provider = orig_build
+
+    check(res["message_id"] == "msg123", "send returns message_id")
+    check(res["from"] == "grant@ten31.xyz", "send reports from")
+    check(res["sent_to"] == ["a@ten31.xyz", "b@ten31.xyz"], "send reports recipients")
+    check(captured["url"].endswith("/users/grant%40ten31.xyz/messages/send"),
+          "posts to messages/send for the impersonated user")
+    check(captured["auth"] == "Bearer tok-grant@ten31.xyz", "uses the compose-scoped token")
+    check("raw" in captured["body"] and "message" not in captured["body"],
+          "send body is {raw:...} (not nested under message, unlike drafts)")
+    check(prov.calls and prov.calls[0][1] == gmail_send._creds.GMAIL_COMPOSE_SCOPE,
+          "requests the gmail.compose scope")
+
+    # 3. validation
+    try:
+        gmail_send.send_via_gmail("x@x", [], "s", "b"); ok = False
+    except ValueError:
+        ok = True
+    check(ok, "empty recipients -> ValueError")
+    try:
+        gmail_send.send_via_gmail("", ["a@x"], "s", "b"); ok = False
+    except ValueError:
+        ok = True
+    check(ok, "missing sender -> ValueError")
+
+    # 4. transport() selection
+    orig_avail = gmail_send.gmail_available
+    orig_smtp_cfg = smtp_send.smtp_configured
+    try:
+        gmail_send.gmail_available = lambda: True
+        check(digest_mailer.transport() == "gmail-dwd", "transport prefers gmail-dwd")
+        gmail_send.gmail_available = lambda: False
+        smtp_send.smtp_configured = lambda: True
+        check(digest_mailer.transport() == "smtp", "transport falls back to smtp")
+        smtp_send.smtp_configured = lambda: False
+        check(digest_mailer.transport() is None, "transport None when neither configured")
+    finally:
+        gmail_send.gmail_available = orig_avail
+        smtp_send.smtp_configured = orig_smtp_cfg
+
+    # 5. send_digest routes to gmail + honors CRM_DIGEST_SENDER
+    orig_send = gmail_send.send_via_gmail
+    sent = {}
+    try:
+        gmail_send.gmail_available = lambda: True
+
+        def fake_send(sender, to, subj, body, conn=None):
+            sent.update(sender=sender, to=to)
+            return {"sent_to": to, "from": sender, "message_id": "m"}
+
+        gmail_send.send_via_gmail = fake_send
+        os.environ["CRM_DIGEST_SENDER"] = "digest@ten31.xyz"
+        res = digest_mailer.send_digest(None, ["a@ten31.xyz"], "S", "B")
+        check(res["transport"] == "gmail-dwd", "send_digest tags transport gmail-dwd")
+        check(sent["sender"] == "digest@ten31.xyz", "send_digest uses CRM_DIGEST_SENDER")
+    finally:
+        gmail_send.send_via_gmail = orig_send
+        gmail_send.gmail_available = orig_avail
+        os.environ.pop("CRM_DIGEST_SENDER", None)
+
+    # 6. send_digest raises NoTransport when neither is available
+    try:
+        gmail_send.gmail_available = lambda: False
+        smtp_send.smtp_configured = lambda: False
+        try:
+            digest_mailer.send_digest(None, ["a@x"], "S", "B"); ok = False
+        except digest_mailer.NoTransport:
+            ok = True
+        check(ok, "send_digest raises NoTransport when no transport")
+    finally:
+        gmail_send.gmail_available = orig_avail
+        smtp_send.smtp_configured = orig_smtp_cfg
+
+    # 7. HTTPError from Gmail is wrapped as RuntimeError (not raised raw)
+    orig_build = gmail_send._creds.build_provider
+    gmail_send._creds.build_provider = lambda factory: _Provider()
+    orig_open2 = urllib.request.urlopen
+
+    def raise_http(req, timeout=None):
+        raise urllib.error.HTTPError(req.full_url, 403, "Forbidden", {},
+                                     io.BytesIO(b'{"error":"denied"}'))
+
+    urllib.request.urlopen = raise_http
+    try:
+        try:
+            gmail_send.send_via_gmail("g@x", ["a@x"], "s", "b"); ok = False
+        except RuntimeError as e:
+            ok = "403" in str(e)
+        except urllib.error.HTTPError:
+            ok = False  # should have been wrapped, not raised raw
+    finally:
+        urllib.request.urlopen = orig_open2
+        gmail_send._creds.build_provider = orig_build
+    check(ok, "Gmail HTTPError wrapped as RuntimeError carrying the status")
+
+    # 8. default_sender: env wins; else DB first-admin; else None
+    class _FakeCur:
+        def __init__(self, row):
+            self._row = row
+
+        def fetchone(self):
+            return self._row
+
+    class _Conn:
+        def __init__(self, row):
+            self._row = row
+
+        def execute(self, *a, **k):
+            return _FakeCur(self._row)
+
+    os.environ.pop("CRM_DIGEST_SENDER", None)
+    check(digest_mailer.default_sender(_Conn({"email": "first@ten31.xyz"})) == "first@ten31.xyz",
+          "default_sender falls back to first active admin")
+    check(digest_mailer.default_sender(_Conn(None)) is None,
+          "default_sender returns None when no admin has an email")
+    os.environ["CRM_DIGEST_SENDER"] = "env@ten31.xyz"
+    check(digest_mailer.default_sender(_Conn({"email": "first@ten31.xyz"})) == "env@ten31.xyz",
+          "CRM_DIGEST_SENDER overrides the DB lookup")
+    os.environ.pop("CRM_DIGEST_SENDER", None)
+
+    # 9. send_digest tags transport 'smtp' on the fallback path
+    orig_avail2 = gmail_send.gmail_available
+    orig_smtp_cfg2 = smtp_send.smtp_configured
+    orig_smtp_send = smtp_send.send_email
+    try:
+        gmail_send.gmail_available = lambda: False
+        smtp_send.smtp_configured = lambda: True
+        smtp_send.send_email = lambda to, subj, body, **k: {"sent_to": to, "from": "f@x"}
+        res = digest_mailer.send_digest(None, ["a@x"], "S", "B")
+        check(res["transport"] == "smtp", "send_digest tags transport smtp on fallback")
+    finally:
+        gmail_send.gmail_available = orig_avail2
+        smtp_send.smtp_configured = orig_smtp_cfg2
+        smtp_send.send_email = orig_smtp_send
+
+    print()
+    if FAILS:
+        print(f"FAILED ({len(FAILS)}):")
+        for f in FAILS:
+            print("  -", f)
+        raise SystemExit(1)
+    print("ALL PASS (gmail_send + digest_mailer)")
+
+
+if __name__ == "__main__":
+    main()

backend/test_grid_add_investor.py

@@ -0,0 +1,202 @@
+#!/usr/bin/env python3
+"""Tests for the mobile add-investor flow (Phase 8g).
+
+Boots the REAL server against a temp DB and exercises the create path the mobile
+"New investor" sheet drives:
+  - POST /api/fundraising/log-communication with create_investor_if_missing honors an
+    optional initial `priority` flag on the NEW row (and defaults it to False when omitted);
+  - the brand-new row's source_row_id resolves immediately for the follow-on
+    POST /api/fundraising/pipeline/link (the relational sync runs inside the create), so the
+    create -> link-at-stage handshake the UI does works end to end;
+  - a follow-on POST /api/reminders with the new row's source_row_id resolves to the synced
+    investor (the create -> reminder handshake).
+Synthetic data only.
+
+Run: cd backend && python3 test_grid_add_investor.py
+"""
+import http.client
+import json
+import os
+import sqlite3
+import sys
+import tempfile
+import threading
+from http.server import ThreadingHTTPServer
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import server  # noqa: E402
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Quiet(server.CRMHandler):
+    def log_message(self, *a):
+        pass
+
+
+def _req(port, method, path, token=None, body=None):
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    headers = {}
+    if token:
+        headers["Authorization"] = "Bearer " + token
+    payload = None
+    if body is not None:
+        payload = json.dumps(body)
+        headers["Content-Type"] = "application/json"
+    conn.request(method, path, body=payload, headers=headers)
+    resp = conn.getresponse()
+    raw = resp.read().decode("utf-8", "replace")
+    conn.close()
+    data = None
+    if raw:
+        try:
+            data = json.loads(raw)
+        except ValueError:
+            pass
+    return resp.status, data
+
+
+def _db():
+    return sqlite3.connect(os.environ["CRM_DB_PATH"])
+
+
+def seed():
+    c = _db()
+    c.execute("INSERT INTO users (id,username,email,password_hash,full_name,role,is_active) "
+              "VALUES ('u1','grant','grant@ten31.example','x','Grant','admin',1)")
+    c.commit()
+    c.close()
+
+
+def _create(port, token, name, contact_name, **extra):
+    body = {
+        "investor_name": name, "create_investor_if_missing": True,
+        "contact": {"name": contact_name, "email": contact_name.split(" ")[0].lower() + "@firm.com"},
+        "type": "note", "body": extra.pop("note", ""), "append_note": bool(extra.pop("note_append", False)),
+    }
+    body.update(extra)
+    return _req(port, "POST", "/api/fundraising/log-communication", token, body)
+
+
+def _grid_rows(port, token):
+    st, d = _req(port, "GET", "/api/fundraising/state", token)
+    return {r["id"]: r for r in (d or {}).get("data", {}).get("grid", {}).get("rows", [])}
+
+
+def main():
+    server.init_db()
+    seed()
+    token = server.create_token("u1", "grant", "admin")
+
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), _Quiet)
+    port = httpd.server_address[1]
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        # ── create with priority:true seeds the row's Priority flag ──
+        print("\n[create: optional initial priority flag honored]")
+        st, d = _create(port, token, "Acme Capital", "Jane Doe", priority=True, note="Intro call", note_append=True)
+        row = (d or {}).get("data", {}).get("row") or {}
+        acme_id = row.get("id")
+        check(st == 201, f"create -> 201 (got {st})")
+        check(row.get("priority") is True, f"returned row carries priority=true (got {row.get('priority')!r})")
+        rows = _grid_rows(port, token)
+        check(rows.get(acme_id, {}).get("priority") is True,
+              f"GET /state shows the new row priority=true (got {rows.get(acme_id, {}).get('priority')!r})")
+        check(len(rows.get(acme_id, {}).get("contacts", [])) == 1,
+              f"new row has its first contact (got {rows.get(acme_id, {}).get('contacts')})")
+
+        # ── create without priority defaults to False (no accidental flag) ──
+        print("\n[create: priority defaults False when omitted]")
+        st, d = _create(port, token, "Beta Partners", "Pat Roe")  # no note, no priority
+        beta = (d or {}).get("data", {}).get("row") or {}
+        beta_id = beta.get("id")
+        check(st == 201, f"no-note create -> 201 (got {st})")
+        check(beta.get("priority") is False, f"omitted priority -> False (got {beta.get('priority')!r})")
+
+        # ── priority is honored ONLY on the create branch: logging against an EXISTING row
+        #    with priority:true must not flip its flag (Beta was created without priority) ──
+        print("\n[invariant: priority on an existing-row log does NOT change its flag]")
+        st, _ = _req(port, "POST", "/api/fundraising/log-communication", token, {
+            "row_id": beta_id, "type": "note", "body": "follow-up", "append_note": True, "priority": True,
+        })
+        check(st in (200, 201), f"log against existing Beta -> ok (got {st})")
+        rows = _grid_rows(port, token)
+        check(rows.get(beta_id, {}).get("priority") is False,
+              f"existing-row priority untouched by the log's priority flag (got {rows.get(beta_id, {}).get('priority')!r})")
+
+        # ── create -> link handshake: the brand-new row links at the chosen stage ──
+        print("\n[create -> link: freshly-created row resolves for pipeline link at stage]")
+        st, d = _req(port, "POST", "/api/fundraising/pipeline/link", token, {
+            "source_row_id": acme_id, "contact_index": 0, "name": "Acme Capital — Pipeline",
+            "stage": "engaged", "expected_amount": 0, "probability": 55, "fund_name": "",
+        })
+        opp = (d or {}).get("data") or {}
+        check(st == 201 and opp.get("stage") == "engaged",
+              f"link new row @engaged -> 201 (got {st}, stage={opp.get('stage')})")
+        rows = _grid_rows(port, token)
+        check(rows.get(acme_id, {}).get("pipeline") is True
+              and rows.get(acme_id, {}).get("pipeline_stage") == "engaged",
+              f"new row now in pipeline @engaged (got {rows.get(acme_id, {}).get('pipeline')}, "
+              f"{rows.get(acme_id, {}).get('pipeline_stage')})")
+
+        # ── create -> reminder handshake: source_row_id resolves to the synced investor ──
+        print("\n[create -> reminder: source_row_id resolves to the new investor]")
+        st, d = _req(port, "POST", "/api/reminders", token, {
+            "source_row_id": acme_id, "investor_name": "Acme Capital",
+            "title": "Send Fund III deck", "due_date": "2026-07-01", "details": "",
+        })
+        rem = (d or {}).get("data") or {}
+        check(st == 201, f"reminder create -> 201 (got {st})")
+        check(bool(rem.get("investor_id")) and rem.get("investor_name") == "Acme Capital",
+              f"reminder linked to the new investor (got id={rem.get('investor_id')!r}, "
+              f"name={rem.get('investor_name')!r})")
+
+        # ── card-intake contact fields land on the canonical contact (phone/city/linkedin) ──
+        #    The Matrix card flow sends these on the contact dict; the upsert must persist them.
+        print("\n[contact fields: phone + city + linkedin persist on the contact]")
+        st, d = _req(port, "POST", "/api/fundraising/log-communication", token, {
+            "investor_name": "Fortitude Investment Group", "create_investor_if_missing": True,
+            "contact": {"name": "Daniel Raupp", "email": "draupp@fortitude.example",
+                        "phone": "631-474-5610", "mobile": "631-922-1195", "city": "Setauket, NY",
+                        "linkedin_url": "linkedin.com/in/danielraupp"},
+            "type": "note", "body": "from a business card", "append_note": True,
+        })
+        check(st == 201, f"create with contact fields -> 201 (got {st})")
+        c = _db()
+        crow = c.execute("SELECT phone, mobile, city, linkedin_url FROM contacts WHERE lower(email) = ?",
+                         ("draupp@fortitude.example",)).fetchone()
+        c.close()
+        check(crow is not None, "contact row exists")
+        check(bool(crow) and crow[0] == "631-474-5610", f"phone (office) persisted (got {crow[0] if crow else None!r})")
+        check(bool(crow) and crow[1] == "631-922-1195", f"mobile (cell) persisted (got {crow[1] if crow else None!r})")
+        check(bool(crow) and crow[2] == "Setauket, NY", f"city persisted (got {crow[2] if crow else None!r})")
+        check(bool(crow) and crow[3] == "linkedin.com/in/danielraupp",
+              f"linkedin persisted (got {crow[3] if crow else None!r})")
+
+        # ── unknown source_row_id is refused (guard) ──
+        print("\n[guard: reminder on an unknown source_row_id -> 404]")
+        st, _ = _req(port, "POST", "/api/reminders", token, {
+            "source_row_id": "nope", "title": "x",
+        })
+        check(st == 404, f"unknown source_row_id -> 404 (got {st})")
+    finally:
+        httpd.shutdown()
+
+    print("\n" + ("ALL PASS" if not FAILS else f"{len(FAILS)} FAILURE(S):"))
+    for f in FAILS:
+        print("  - " + f)
+    sys.exit(1 if FAILS else 0)
+
+
+if __name__ == "__main__":
+    main()

backend/test_grid_comm_timeline.py

@@ -0,0 +1,164 @@
+#!/usr/bin/env python3
+"""Regression test for the Grid-detail communications timeline filter (Phase 8c, G6).
+
+The mobile Grid detail's notes timeline pulls an investor-level communication stream
+via GET /api/communications?source_row_id=<grid row id>. That filter (added to
+handle_list_communications) maps the grid JSON row id → fundraising_investors.source_row_id
+→ fundraising_contacts.contact_id → communications, so it must:
+  - return every communication across ALL the investor's contacts,
+  - stay isolated (one investor's row id never returns another's comms),
+  - respect soft-delete (cm.deleted_at IS NULL) through the join.
+
+Boots the REAL server, seeds investors by driving the one-row log path (which creates the
+grid row + contact + communication AND syncs the relational mirror the filter joins on),
+then drives the live read path with a real token. Synthetic only (guardrail #9).
+
+Run: cd backend && python3 test_grid_comm_timeline.py
+"""
+import http.client
+import json
+import os
+import sqlite3
+import sys
+import tempfile
+import threading
+from http.server import ThreadingHTTPServer
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import server  # noqa: E402
+
+FAILS = []
+DEL = "2026-06-01T00:00:00"
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Quiet(server.CRMHandler):
+    def log_message(self, *a):
+        pass
+
+
+def _req(method, port, path, token, body=None):
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    headers = {"Authorization": "Bearer " + token}
+    payload = None
+    if body is not None:
+        payload = json.dumps(body)
+        headers["Content-Type"] = "application/json"
+    conn.request(method, path, body=payload, headers=headers)
+    resp = conn.getresponse()
+    raw = resp.read().decode("utf-8", "replace")
+    conn.close()
+    data = None
+    if raw:
+        try:
+            data = json.loads(raw)
+        except ValueError:
+            pass
+    return resp.status, data
+
+
+def _log_comm(port, token, investor_name, contact, subject, create=False):
+    """Drive the one-row log path; returns (status, grid_row_id)."""
+    st, data = _req("POST", port, "/api/fundraising/log-communication", token, {
+        "investor_name": investor_name,
+        "create_investor_if_missing": create,
+        "contact": contact,
+        "type": "note",
+        "subject": subject,
+        "body": subject,
+        "append_note": True,
+    })
+    row_id = ((data or {}).get("data", {}).get("row") or {}).get("id")
+    return st, row_id
+
+
+def main():
+    server.init_db()
+    conn = sqlite3.connect(os.environ["CRM_DB_PATH"])
+    # password_hash is intentionally a non-bcrypt placeholder — we mint the token directly via
+    # create_token(), so the password-verify path is never exercised.
+    conn.execute("INSERT INTO users (id,username,email,password_hash,full_name,role,is_active) "
+                 "VALUES ('u1','grant','grant@ten31.example','x','Grant','admin',1)")
+    conn.commit()
+    conn.close()
+    token = server.create_token("u1", "grant", "admin")
+
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), _Quiet)
+    port = httpd.server_address[1]
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        # Investor A: two contacts, one communication per contact. Create seeds the row with
+        # Jane + logs "Intro call"; update-row adds John as a second pill (so the relational mirror
+        # links BOTH contacts to A's row); then a comm is logged against John. The timeline must
+        # aggregate across both contacts — the point of the source_row_id join over a single contact.
+        st, rowA = _log_comm(port, token, "Acme Capital",
+                             {"name": "Jane Doe", "email": "jane@acme.example"}, "Intro call", create=True)
+        check(st == 201 and bool(rowA), f"create investor A via log path -> 201 + row id (got {st}, {rowA})")
+        st, _ = _req("POST", port, "/api/fundraising/update-row", token, {
+            "row_id": rowA, "investor_name": "Acme Capital",
+            "contacts": [{"name": "Jane Doe", "email": "jane@acme.example"},
+                         {"name": "John Roe", "email": "john@acme.example"}],
+        })
+        check(st == 200, f"add John as a second contact on A via update-row (got {st})")
+        st, _ = _log_comm(port, token, "Acme Capital",
+                          {"name": "John Roe", "email": "john@acme.example"}, "Follow-up email")
+        check(st == 201, f"second contact's comm logged on A (got {st})")
+
+        # Investor B: a separate investor, one communication (isolation control).
+        st, rowB = _log_comm(port, token, "Beacon Ventures",
+                             {"name": "Sam Poe", "email": "sam@beacon.example"}, "Beacon note", create=True)
+        check(st == 201 and bool(rowB), f"create investor B via log path -> 201 + row id (got {st}, {rowB})")
+
+        # ── source_row_id returns the whole investor (across contacts) ──
+        print("\n[source_row_id timeline]")
+        st, data = _req("GET", port, f"/api/communications?source_row_id={rowA}", token)
+        subsA = {c.get("subject") for c in (data or {}).get("data", [])}
+        check(st == 200, f"GET timeline for A -> 200 (got {st})")
+        check(subsA == {"Intro call", "Follow-up email"},
+              f"A's timeline spans both contacts' comms (got {subsA})")
+
+        # ── isolation: A's row id never returns B's comms ──
+        print("\n[isolation]")
+        check("Beacon note" not in subsA, "A's timeline excludes investor B's comm")
+        st, dataB = _req("GET", port, f"/api/communications?source_row_id={rowB}", token)
+        subsB = {c.get("subject") for c in (dataB or {}).get("data", [])}
+        check(subsB == {"Beacon note"}, f"B's timeline is its own comm only (got {subsB})")
+
+        # ── soft-delete respected through the join ──
+        print("\n[soft-delete]")
+        c = sqlite3.connect(os.environ["CRM_DB_PATH"])
+        c.execute("UPDATE communications SET deleted_at=? WHERE subject='Intro call'", (DEL,))
+        c.commit()
+        c.close()
+        st, data2 = _req("GET", port, f"/api/communications?source_row_id={rowA}", token)
+        subsA2 = {c.get("subject") for c in (data2 or {}).get("data", [])}
+        check(subsA2 == {"Follow-up email"},
+              f"soft-deleted comm filtered from A's timeline (got {subsA2})")
+
+        # ── unknown row id returns empty, not an error ──
+        st, data3 = _req("GET", port, "/api/communications?source_row_id=does-not-exist", token)
+        check(st == 200 and (data3 or {}).get("data") == [],
+              f"unknown source_row_id -> 200 + empty (got {st}, {(data3 or {}).get('data')})")
+    finally:
+        httpd.shutdown()
+
+    print()
+    if FAILS:
+        print(f"FAILED ({len(FAILS)}):")
+        for f in FAILS:
+            print(f"  - {f}")
+        sys.exit(1)
+    print("ALL PASS (grid comm timeline source_row_id filter)")
+
+
+if __name__ == "__main__":
+    main()

backend/test_grid_email_heal.py

@@ -0,0 +1,145 @@
+#!/usr/bin/env python3
+"""Regression: GET /api/fundraising/state heals blank grid-pill emails from the relational mirror.
+
+The grid blob is canonical for the mobile "Edit investor" sheet, but an email can reach a linked
+classic contact (email capture / a contact edit) without ever being written back into the blob pill
+— so the edit form showed an empty email for a contact the directory clearly had (Grant, 2026-06-20).
+The state handler now fills a blank pill email from fundraising_contacts.email, else the linked
+contacts.email, matched by pill order then name. This asserts:
+  - a blank pill whose linked contact has an email is HEALED on read;
+  - a blank pill whose linked contact is also blank stays blank;
+  - a pill that already carries an email in the blob is NEVER overwritten (fill-only).
+Synthetic data only.
+
+Run: cd backend && python3 test_grid_email_heal.py
+"""
+import http.client
+import json
+import os
+import sqlite3
+import sys
+import tempfile
+import threading
+from http.server import ThreadingHTTPServer
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import server  # noqa: E402
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Quiet(server.CRMHandler):
+    def log_message(self, *a):
+        pass
+
+
+def _get_state(port, token):
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    conn.request("GET", "/api/fundraising/state", headers={"Authorization": "Bearer " + token})
+    resp = conn.getresponse()
+    raw = resp.read().decode("utf-8", "replace")
+    conn.close()
+    return resp.status, (json.loads(raw) if raw else None)
+
+
+GRID = {
+    "columns": [{"id": "investor_name", "label": "Investor", "type": "text"},
+                {"id": "contacts", "label": "Contacts", "type": "contacts"}],
+    "rows": [
+        {"id": "rowW", "investor_name": "Wyoming", "notes": "",
+         "contacts": [{"name": "Philip Treick", "email": "", "title": ""},
+                      {"name": "Jose Briones", "email": "", "title": ""}]},
+        {"id": "rowA", "investor_name": "Acme Capital", "notes": "",
+         "contacts": [{"name": "Jane Doe", "email": "keep@acme.com", "title": ""}]},
+        {"id": "rowO", "investor_name": "Orphan LP", "notes": "",
+         "contacts": [{"name": "No Link", "email": "", "title": ""}]},
+    ],
+}
+
+
+def seed():
+    c = sqlite3.connect(os.environ["CRM_DB_PATH"])
+    c.execute("INSERT INTO users (id,username,email,password_hash,full_name,role,is_active) "
+              "VALUES ('u1','grant','grant@ten31.example','x','Grant','admin',1)")
+    c.execute("INSERT INTO fundraising_state (id, grid_json, views_json, version) "
+              "VALUES ('main', ?, '[]', 1) "
+              "ON CONFLICT(id) DO UPDATE SET grid_json = excluded.grid_json", (json.dumps(GRID),))
+    # Classic contacts directory: Jose has the captured email the blob never got; Philip is blank.
+    c.execute("INSERT INTO contacts (id,first_name,last_name,email) VALUES "
+              "('c-phil','Philip','Treick',''),"
+              "('c-jose','Jose','Briones','jbriones@uwyo.edu'),"
+              "('c-jane','Jane','Doe','other@acme.com')")  # differs from the blob's keep@acme.com
+    # Relational mirror (what sync_fundraising_relational would build): blank fc.email, linked contact_id.
+    c.execute("INSERT INTO fundraising_investors (id,investor_name,source_row_id,total_invested) VALUES "
+              "('inv-w','Wyoming','rowW',0),('inv-a','Acme Capital','rowA',0),('inv-o','Orphan LP','rowO',0)")
+    # fc-orphan has contact_id NULL (pre-0004 orphan) and blank email — nothing to heal from.
+    c.execute("INSERT INTO fundraising_contacts (id,investor_id,full_name,email,sort_order,contact_id) VALUES "
+              "('fc-phil','inv-w','Philip Treick','',0,'c-phil'),"
+              "('fc-jose','inv-w','Jose Briones','',1,'c-jose'),"
+              "('fc-jane','inv-a','Jane Doe','',0,'c-jane'),"
+              "('fc-orphan','inv-o','No Link','',0,NULL)")
+    c.commit()
+    c.close()
+
+
+def main():
+    server.init_db()
+    seed()
+    token = server.create_token("u1", "grant", "admin")
+
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), _Quiet)
+    port = httpd.server_address[1]
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        st, d = _get_state(port, token)
+        rows = ((d or {}).get("data", {}).get("grid", {}) or {}).get("rows", [])
+        by_id = {r.get("id"): r for r in rows}
+        w = by_id.get("rowW", {})
+        a = by_id.get("rowA", {})
+        wc = w.get("contacts", [])
+        ac = a.get("contacts", [])
+
+        print("\n[heal: blank pill email filled from the linked contact (Jose)]")
+        jose = next((c for c in wc if c.get("name") == "Jose Briones"), {})
+        check(st == 200 and jose.get("email") == "jbriones@uwyo.edu",
+              f"Jose pill healed to jbriones@uwyo.edu (got {jose.get('email')!r})")
+
+        print("\n[heal: blank pill whose contact is also blank stays blank (Philip)]")
+        phil = next((c for c in wc if c.get("name") == "Philip Treick"), {})
+        check(phil.get("email", "") == "",
+              f"Philip pill stays blank (got {phil.get('email')!r})")
+
+        print("\n[heal: a pill that already has an email is never overwritten (Jane)]")
+        jane = next((c for c in ac if c.get("name") == "Jane Doe"), {})
+        check(jane.get("email") == "keep@acme.com",
+              f"Jane pill keeps its blob email, not the contact's (got {jane.get('email')!r})")
+
+        print("\n[heal: a pill whose fundraising_contacts row has contact_id NULL stays blank (orphan)]")
+        o = by_id.get("rowO", {})
+        orphan = next((c for c in o.get("contacts", []) if c.get("name") == "No Link"), {})
+        check(orphan.get("email", "") == "",
+              f"orphan pill (no contact_id, no email source) stays blank (got {orphan.get('email')!r})")
+    finally:
+        httpd.shutdown()
+
+    print()
+    if FAILS:
+        print(f"FAILED ({len(FAILS)}):")
+        for f in FAILS:
+            print(f"  - {f}")
+        sys.exit(1)
+    print("ALL PASS (grid email heal)")
+
+
+if __name__ == "__main__":
+    main()

backend/test_grid_pipeline_link.py

@@ -0,0 +1,312 @@
+#!/usr/bin/env python3
+"""Tests for the grid → Pipeline link ("Adopt the Pipeline", v0.1.0:87).
+
+Boots the REAL server against a temp DB and exercises the new endpoints end-to-end:
+  - POST /api/fundraising/pipeline/link creates exactly ONE opportunity, linked via
+    opportunities.fundraising_investor_id, reusing the grid's synced contact (no
+    POST /api/contacts side-door) and mapping the grid 'lead' -> owner;
+  - the link is idempotent: a re-link returns the existing opp and NEVER reseeds its
+    Pipeline-owned funnel fields (stage/probability) — the board owns those;
+  - GET /api/fundraising/state injects read-only pipeline / pipeline_stage row values
+    derived from the live opp;
+  - linking a contactless row, or an unknown row, is refused;
+  - POST .../unlink soft-deletes the opp (off the board, recoverable) while leaving the
+    grid investor row fully intact;
+  - deleting an investor from the grid archives its orphaned opp on the next save;
+  - the pipeline report + dashboard aggregates exclude archived (soft-deleted) opps.
+Synthetic data only.
+
+Run: cd backend && python3 test_grid_pipeline_link.py
+"""
+import http.client
+import json
+import os
+import sqlite3
+import sys
+import tempfile
+import threading
+from http.server import ThreadingHTTPServer
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import server  # noqa: E402
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Quiet(server.CRMHandler):
+    def log_message(self, *a):
+        pass
+
+
+def _req(port, method, path, token=None, body=None):
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    headers = {}
+    if token:
+        headers["Authorization"] = "Bearer " + token
+    payload = None
+    if body is not None:
+        payload = json.dumps(body)
+        headers["Content-Type"] = "application/json"
+    conn.request(method, path, body=payload, headers=headers)
+    resp = conn.getresponse()
+    raw = resp.read().decode("utf-8", "replace")
+    conn.close()
+    data = None
+    if raw:
+        try:
+            data = json.loads(raw)
+        except ValueError:
+            pass
+    return resp.status, data
+
+
+def _put_grid(port, token, rows):
+    return _req(port, "PUT", "/api/fundraising/state", token,
+                {"grid": {"columns": [], "rows": rows}, "views": []})
+
+
+ROW_ACME = {"id": "rowAcme", "investor_name": "Acme Capital", "notes": "", "lead": "Grant",
+            "contacts": [{"name": "Jane Doe", "email": "jane@acme.com", "title": "GP"}]}
+ROW_BETA = {"id": "rowBeta", "investor_name": "Beta Capital LLC", "notes": "", "lead": "",
+            "contacts": [{"name": "Pat Roe", "email": "pat@beta.com", "title": ""}]}
+ROW_EMPTY = {"id": "rowEmpty", "investor_name": "Empty LP", "notes": "", "contacts": []}
+
+
+def _db():
+    return sqlite3.connect(os.environ["CRM_DB_PATH"])
+
+
+def seed():
+    c = _db()
+    c.execute("INSERT INTO users (id,username,email,password_hash,full_name,role,is_active) "
+              "VALUES ('u1','grant','grant@ten31.example','x','Grant','admin',1)")
+    c.commit()
+    c.close()
+
+
+def _opp_count_live(fr_investor_id=None):
+    c = _db()
+    if fr_investor_id:
+        n = c.execute("SELECT COUNT(*) FROM opportunities WHERE fundraising_investor_id = ? "
+                      "AND deleted_at IS NULL", (fr_investor_id,)).fetchone()[0]
+    else:
+        n = c.execute("SELECT COUNT(*) FROM opportunities WHERE deleted_at IS NULL").fetchone()[0]
+    c.close()
+    return n
+
+
+def main():
+    server.init_db()
+    seed()
+    token = server.create_token("u1", "grant", "admin")
+
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), _Quiet)
+    port = httpd.server_address[1]
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        st, _ = _put_grid(port, token, [ROW_ACME, ROW_BETA, ROW_EMPTY])
+        check(st == 200, f"seed grid via PUT /state (got {st})")
+
+        # ── link creates one linked opp with the seeds + resolved contact + mapped owner ──
+        print("\n[link: creates one linked opportunity with seeds]")
+        st, d = _req(port, "POST", "/api/fundraising/pipeline/link", token, {
+            "source_row_id": "rowAcme", "fund_name": "Fund III",
+            "expected_amount": 250000, "probability": 40, "stage": "engaged",
+        })
+        opp = (d or {}).get("data") or {}
+        check(st == 201 and (d or {}).get("already_linked") is False, f"link -> 201 new (got {st}, {d})")
+        check(opp.get("stage") == "engaged" and opp.get("expected_amount") == 250000
+              and opp.get("probability") == 40 and opp.get("fund_name") == "Fund III",
+              f"seeds applied (got {{stage:{opp.get('stage')}, amt:{opp.get('expected_amount')}, "
+              f"prob:{opp.get('probability')}, fund:{opp.get('fund_name')}}})")
+        check(opp.get("first_name") == "Jane", f"reused synced contact Jane Doe (got {opp.get('first_name')})")
+        check(opp.get("owner_name") == "Grant", f"grid lead 'Grant' -> owner Grant (got {opp.get('owner_name')})")
+        fr_id = opp.get("fundraising_investor_id")
+        check(bool(fr_id), f"opportunity carries fundraising_investor_id (got {fr_id})")
+        check(_opp_count_live(fr_id) == 1, "exactly one live opp linked to the investor")
+        opp_id = opp.get("id")
+        jane_contact_id = opp.get("contact_id")
+
+        # ── idempotent re-link: returns existing, board-owned stage NOT reseeded ──
+        print("\n[idempotent: re-link returns existing opp without reseeding funnel fields]")
+        st, _ = _req(port, "PATCH", f"/api/opportunities/{opp_id}/stage", token, {"stage": "diligence"})
+        check(st == 200, f"advance stage on the board -> diligence (got {st})")
+        st, d = _req(port, "POST", "/api/fundraising/pipeline/link", token, {
+            "source_row_id": "rowAcme", "stage": "lead", "expected_amount": 999, "probability": 5,
+        })
+        opp2 = (d or {}).get("data") or {}
+        check(st == 200 and (d or {}).get("already_linked") is True, f"re-link -> already_linked (got {st}, {d})")
+        check(opp2.get("stage") == "diligence" and opp2.get("expected_amount") == 250000,
+              f"funnel fields preserved, not reseeded (got stage={opp2.get('stage')}, amt={opp2.get('expected_amount')})")
+        check(_opp_count_live(fr_id) == 1, "still exactly one live opp (no duplicate)")
+
+        # ── stage validation: legacy/invalid values rejected (4-stage enum guard) ──
+        # The stage check precedes the contact lookup in handle_create_opportunity, so a fake
+        # contact_id still surfaces the stage error first.
+        print("\n[validation: legacy stage values rejected by stage + create endpoints]")
+        st, _ = _req(port, "PATCH", f"/api/opportunities/{opp_id}/stage", token, {"stage": "outreach"})
+        check(st >= 400, f"PATCH legacy stage 'outreach' rejected (got {st})")
+        st, _ = _req(port, "POST", "/api/opportunities", token,
+                     {"name": "X", "contact_id": "x", "stage": "due_diligence"})
+        check(st >= 400, f"POST opportunity with legacy stage 'due_diligence' rejected (got {st})")
+
+        # ── read-injection: GET state shows pipeline flag + stage, derived live ──
+        print("\n[read-injection: GET /state exposes read-only pipeline + pipeline_stage]")
+        st, d = _req(port, "GET", "/api/fundraising/state", token)
+        rows = {r["id"]: r for r in (d or {}).get("data", {}).get("grid", {}).get("rows", [])}
+        check(rows.get("rowAcme", {}).get("pipeline") is True
+              and rows.get("rowAcme", {}).get("pipeline_stage") == "diligence",
+              f"rowAcme pipeline true @diligence (got {rows.get('rowAcme', {}).get('pipeline')}, "
+              f"{rows.get('rowAcme', {}).get('pipeline_stage')})")
+        # Read-only opportunity_id is injected for a linked row so the mobile grid detail can
+        # PATCH the stage on the opportunities endpoint (the grid row carries no opp id otherwise).
+        check(rows.get("rowAcme", {}).get("opportunity_id") == opp_id,
+              f"rowAcme carries the live opportunity_id (got {rows.get('rowAcme', {}).get('opportunity_id')}, want {opp_id})")
+        # Phase 0 derived signals are injected read-only on every row (values depend on seed;
+        # assert the keys are present so the strip/inject round-trip below is meaningful).
+        check(all(k in rows.get("rowAcme", {}) for k in ("existing_investor", "staleness", "last_activity_at")),
+              f"rowAcme carries derived existing_investor/staleness/last_activity (keys: {sorted(rows.get('rowAcme', {}).keys())})")
+        check(rows.get("rowBeta", {}).get("pipeline") is False
+              and rows.get("rowBeta", {}).get("pipeline_stage") == ""
+              and rows.get("rowBeta", {}).get("opportunity_id") == "",
+              f"rowBeta not in pipeline (got {rows.get('rowBeta', {}).get('pipeline')}, "
+              f"opp_id={rows.get('rowBeta', {}).get('opportunity_id')!r})")
+
+        # ── round-trip: a save echoing the injected read-only values is lossless ──
+        print("\n[round-trip: PUT carrying injected pipeline values strips them, link intact]")
+        st, d = _req(port, "GET", "/api/fundraising/state", token)
+        echoed = (d or {}).get("data", {}).get("grid", {}).get("rows", [])
+        st, _ = _put_grid(port, token, echoed)  # as the frontend autosave would, rows still carry pipeline*
+        check(st == 200, f"echo-back save -> 200 (got {st})")
+        check(_opp_count_live(fr_id) == 1, "link survives the round-trip (no dup, not archived)")
+        c = _db()
+        blob = json.loads(c.execute("SELECT grid_json FROM fundraising_state WHERE id='main'").fetchone()[0])
+        c.close()
+        stored_acme = {r["id"]: r for r in blob.get("rows", [])}.get("rowAcme", {})
+        check(not any(k in stored_acme for k in ("pipeline", "pipeline_stage", "opportunity_id",
+                                                  "existing_investor", "staleness", "last_activity_at")),
+              "computed keys (pipeline + opportunity_id + existing_investor/staleness/last_activity) NOT persisted into the grid blob")
+        st, d = _req(port, "GET", "/api/fundraising/state", token)
+        rt = {r["id"]: r for r in (d or {}).get("data", {}).get("grid", {}).get("rows", [])}.get("rowAcme", {})
+        check(rt.get("pipeline") is True and rt.get("pipeline_stage") == "diligence"
+              and rt.get("opportunity_id") == opp_id,
+              f"pipeline values re-injected after round-trip (got {rt.get('pipeline')}, {rt.get('pipeline_stage')}, opp_id={rt.get('opportunity_id')!r})")
+        check(all(k in rt for k in ("existing_investor", "staleness", "last_activity_at")),
+              "derived signals re-injected after round-trip")
+
+        # ── guards ──
+        print("\n[guard: a contactless row cannot be added to the pipeline]")
+        st, d = _req(port, "POST", "/api/fundraising/pipeline/link", token, {"source_row_id": "rowEmpty"})
+        check(st == 400, f"no contact -> 400 (got {st}, {d})")
+        check(_opp_count_live() == 1, "no stray opp created for the contactless row")
+
+        print("\n[guard: unknown grid row -> 404]")
+        st, _ = _req(port, "POST", "/api/fundraising/pipeline/link", token, {"source_row_id": "nope"})
+        check(st == 404, f"unknown row -> 404 (got {st})")
+
+        print("\n[guard: unauthenticated -> 401]")
+        st, _ = _req(port, "POST", "/api/fundraising/pipeline/link", None, {"source_row_id": "rowAcme"})
+        check(st == 401, f"no token -> 401 (got {st})")
+
+        # ── the opp loads on the board + counts in the dashboard while live ──
+        print("\n[board + dashboard count the live opp]")
+        st, d = _req(port, "GET", "/api/opportunities?limit=1000", token)
+        ids = [o["id"] for o in (d or {}).get("data", [])]
+        check(opp_id in ids, "linked opp appears on the board")
+        # 8h deep-link: the opp list injects source_row_id from the durable fundraising_investor_id
+        # (the mobile Pipeline detail's "Open investor in Grid" target).
+        acme_opp = next((o for o in (d or {}).get("data", []) if o["id"] == opp_id), {})
+        check(acme_opp.get("source_row_id") == "rowAcme",
+              f"opp list injects source_row_id == 'rowAcme' (got {acme_opp.get('source_row_id')!r})")
+        st, d = _req(port, "GET", "/api/reports/dashboard", token)
+        active = (d or {}).get("data", {}).get("metrics", {}).get("active_opportunities")
+        check(active == 1, f"dashboard active_opportunities == 1 (got {active})")
+
+        # ── unlink soft-deletes the opp; the GRID ROW stays fully intact ──
+        print("\n[unlink: archives the opp, leaves the grid investor intact]")
+        st, d = _req(port, "POST", "/api/fundraising/pipeline/unlink", token, {"source_row_id": "rowAcme"})
+        check(st == 200 and (d or {}).get("data", {}).get("archived") == 1, f"unlink -> archived 1 (got {st}, {d})")
+        check(_opp_count_live(fr_id) == 0, "opp is no longer live (soft-deleted)")
+        c = _db()
+        gone = c.execute("SELECT deleted_at FROM opportunities WHERE id = ?", (opp_id,)).fetchone()[0]
+        inv_still = c.execute("SELECT investor_name FROM fundraising_investors WHERE source_row_id = 'rowAcme'").fetchone()
+        contact_still = c.execute("SELECT COUNT(*) FROM fundraising_contacts WHERE investor_id = ?", (fr_id,)).fetchone()[0]
+        c.close()
+        check(gone is not None, "opp row tombstoned (deleted_at set), not hard-deleted")
+        check(inv_still and inv_still[0] == "Acme Capital", "grid investor row untouched by unlink")
+        check(contact_still >= 1, "grid investor's contacts untouched by unlink")
+        st, d = _req(port, "GET", "/api/opportunities?limit=1000", token)
+        check(opp_id not in [o["id"] for o in (d or {}).get("data", [])], "archived opp left the board")
+        st, d = _req(port, "GET", "/api/fundraising/state", token)
+        rows = {r["id"]: r for r in (d or {}).get("data", {}).get("grid", {}).get("rows", [])}
+        check(rows.get("rowAcme", {}).get("pipeline") is False, "grid no longer flags rowAcme as in-pipeline")
+
+        # ── aggregates exclude the archived opp ──
+        print("\n[aggregates exclude archived opps]")
+        st, d = _req(port, "GET", "/api/reports/dashboard", token)
+        active = (d or {}).get("data", {}).get("metrics", {}).get("active_opportunities")
+        check(active == 0, f"dashboard active_opportunities back to 0 (got {active})")
+        st, d = _req(port, "GET", "/api/reports/pipeline", token)
+        by_stage = (d or {}).get("data", {}).get("by_stage", [])
+        total = sum(s.get("count", 0) for s in by_stage)
+        check(total == 0, f"pipeline report by_stage excludes archived (got total {total})")
+
+        # ── re-link after unlink: a fresh opp is created (the archived one stays archived) ──
+        print("\n[re-link after unlink: creates a new opp, flag reappears]")
+        st, d = _req(port, "POST", "/api/fundraising/pipeline/link", token, {
+            "source_row_id": "rowAcme", "stage": "engaged", "expected_amount": 50000,
+        })
+        relinked = (d or {}).get("data") or {}
+        check(st == 201 and (d or {}).get("already_linked") is False and relinked.get("id") != opp_id,
+              f"re-link -> a NEW opp distinct from the archived one (got {st}, {relinked.get('id')} vs {opp_id})")
+        check(_opp_count_live(fr_id) == 1, "exactly one live opp again after re-link")
+        st, _ = _req(port, "POST", "/api/fundraising/pipeline/unlink", token, {"source_row_id": "rowAcme"})
+        check(st == 200, "reset: unlink the re-linked opp")
+
+        # ── orphan reconciler: deleting the investor from the grid archives its opp ──
+        print("\n[orphan: deleting the grid investor archives its linked opp on next save]")
+        st, d = _req(port, "POST", "/api/fundraising/pipeline/link", token, {
+            "source_row_id": "rowBeta", "stage": "lead", "expected_amount": 100000,
+        })
+        beta = (d or {}).get("data") or {}
+        beta_opp_id, beta_fr = beta.get("id"), beta.get("fundraising_investor_id")
+        check(st == 201 and _opp_count_live(beta_fr) == 1, f"beta linked (got {st})")
+        # drop rowBeta from the grid (keep the others)
+        st, _ = _put_grid(port, token, [ROW_ACME, ROW_EMPTY])
+        check(st == 200, f"save grid without rowBeta (got {st})")
+        check(_opp_count_live(beta_fr) == 0, "beta's orphaned opp archived by the reconciler")
+        st, d = _req(port, "GET", "/api/opportunities?limit=1000", token)
+        check(beta_opp_id not in [o["id"] for o in (d or {}).get("data", [])], "orphaned opp left the board")
+
+        # ── 8h: a manually-created deal (no fundraising_investor_id) has a null source_row_id, so
+        #        the mobile Pipeline detail hides "Open investor in Grid" for it ──
+        print("\n[8h: opp source_row_id is null for a deal with no grid link]")
+        st, d = _req(port, "POST", "/api/opportunities", token,
+                     {"name": "Manual deal", "contact_id": jane_contact_id, "stage": "lead"})
+        manual_id = (d or {}).get("data", {}).get("id")
+        check(st in (200, 201) and bool(manual_id), f"create a manual (non-grid) opp (got {st})")
+        st, d = _req(port, "GET", "/api/opportunities?limit=1000", token)
+        manual = next((o for o in (d or {}).get("data", []) if o["id"] == manual_id), {})
+        check(manual.get("source_row_id") in (None, ""),
+              f"manual opp has null source_row_id (got {manual.get('source_row_id')!r})")
+    finally:
+        httpd.shutdown()
+
+    print("\n" + ("ALL PASS" if not FAILS else f"{len(FAILS)} FAILURE(S):"))
+    for f in FAILS:
+        print("  - " + f)
+    sys.exit(1 if FAILS else 0)
+
+
+if __name__ == "__main__":
+    main()

backend/test_intake_card.py

@@ -0,0 +1,275 @@
+#!/usr/bin/env python3
+"""Tests for the in-app business-card intake endpoint (#7): POST /api/intake/card.
+
+The endpoint reuses the Matrix card flow's nio-free core — vision-transcribe (spark) -> text
+parse (parse) -> the same fuzzy matcher (find_intake_match / find_intake_candidates) — minus
+Matrix, surfaced for a mobile sheet. The real vision/OCR path is live-smoke only (same as the
+Matrix M3 path), so here we STUB the two network legs and assert the wiring + contract:
+  - happy path: transcribe -> parse -> proposal + match/candidates, status 200 ok:true;
+  - the email-integrity rule rides along (a model-minted address NOT in the transcription is
+    dropped in favor of the one literally present), exactly as on the text/Matrix path;
+  - new-vs-existing: an exact firm name returns `match`; a near-spelling returns `candidates`;
+  - soft-fails: an unreadable image -> ok:false/unreadable; vision down -> 502/vision_unavailable;
+  - guards: missing/invalid image -> 400; unauthenticated -> 401;
+  - provenance: the approve write reuses log-communication tagged source="app_card".
+Synthetic data only.
+
+Run: cd backend && python3 test_intake_card.py
+"""
+import base64
+import http.client
+import json
+import os
+import sqlite3
+import sys
+import tempfile
+import threading
+from http.server import ThreadingHTTPServer
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+
+_BACKEND = os.path.dirname(os.path.abspath(__file__))
+sys.path.insert(0, _BACKEND)
+sys.path.insert(0, os.path.join(_BACKEND, "ingest"))         # llm
+sys.path.insert(0, os.path.join(_BACKEND, "matrix_intake"))  # spark, parse
+
+import server  # noqa: E402
+import llm     # noqa: E402  (ingest/llm.py — patched so spark.parse_json hits no network)
+import spark   # noqa: E402  (matrix_intake/spark.py — transcribe_card stubbed)
+import parse   # noqa: E402  (matrix_intake/parse.py — parse_message defaults to spark.parse_json)
+
+FAILS = []
+
+# The handler imports `spark`/`parse` lazily and looks up transcribe_card on the module at call
+# time, so patching the module attribute here takes effect. parse.parse_message binds its default
+# parse_fn=spark.parse_json at import, and spark.parse_json calls llm.chat_json dynamically — so
+# patching llm.chat_json (not spark.parse_json) is what reaches the parse leg.
+_STATE = {"transcription": "", "raw": {}, "boom": False}
+
+
+def _fake_transcribe(image_b64, mime="image/jpeg", chat_fn=None):
+    if _STATE["boom"]:
+        raise RuntimeError("spark control unreachable")
+    return _STATE["transcription"]
+
+
+def _fake_chat_json(prompt, system=None, max_tokens=200):
+    return dict(_STATE["raw"])
+
+
+spark.transcribe_card = _fake_transcribe
+llm.chat_json = _fake_chat_json
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Quiet(server.CRMHandler):
+    def log_message(self, *a):
+        pass
+
+
+def _req(port, method, path, token=None, body=None):
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    headers = {}
+    if token:
+        headers["Authorization"] = "Bearer " + token
+    payload = None
+    if body is not None:
+        payload = json.dumps(body)
+        headers["Content-Type"] = "application/json"
+    conn.request(method, path, body=payload, headers=headers)
+    resp = conn.getresponse()
+    raw = resp.read().decode("utf-8", "replace")
+    conn.close()
+    data = None
+    if raw:
+        try:
+            data = json.loads(raw)
+        except ValueError:
+            pass
+    return resp.status, data
+
+
+GRID = {
+    "columns": [],
+    "rows": [
+        {"id": "rowAcme", "investor_name": "Acme Capital", "notes": "",
+         "contacts": [{"name": "Jane Doe", "email": "jane@acme.com", "title": "GP"}]},
+    ],
+}
+
+_IMG = base64.b64encode(b"not-a-real-image-just-valid-base64").decode()
+
+
+def seed():
+    c = sqlite3.connect(os.environ["CRM_DB_PATH"])
+    c.execute("INSERT INTO users (id,username,email,password_hash,full_name,role,is_active) "
+              "VALUES ('u1','grant','grant@ten31.example','x','Grant','admin',1)")
+    c.execute("INSERT INTO fundraising_state (id, grid_json, views_json, version) "
+              "VALUES ('main', ?, '[]', 1) "
+              "ON CONFLICT(id) DO UPDATE SET grid_json = excluded.grid_json", (json.dumps(GRID),))
+    c.commit()
+    c.close()
+
+
+def main():
+    server.init_db()
+    seed()
+    token = server.create_token("u1", "grant", "admin")
+
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), _Quiet)
+    port = httpd.server_address[1]
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        print("\n[happy path: transcribe -> parse -> proposal, new investor, no match]")
+        _STATE["transcription"] = ("Sam Lee\nPartner\nBeacon Ventures\n"
+                                   "sam@beacon.vc\nMobile: +1 555 987 6543")
+        _STATE["raw"] = {"intent": "new_investor", "investor_name": "Beacon Ventures",
+                         "contact_name": "Sam Lee", "contact_title": "Partner",
+                         "mobile": "+1 555 987 6543", "contact_email": "sam@beacon.vc"}
+        _STATE["boom"] = False
+        st, d = _req(port, "POST", "/api/intake/card", token, {"image_b64": _IMG})
+        data = (d or {}).get("data", {})
+        p = data.get("proposal", {})
+        check(st == 200 and data.get("ok") is True, f"200 ok:true (got {st}, {data})")
+        check(p.get("investor_name") == "Beacon Ventures" and p.get("contact_name") == "Sam Lee",
+              f"proposal carries firm + person (got {p})")
+        check(p.get("contact_email") == "sam@beacon.vc", f"email kept (got {p.get('contact_email')})")
+        check(p.get("mobile") == "+1 555 987 6543", f"mobile kept (got {p.get('mobile')})")
+        check("transcription" in data and data["match"] is None and data["candidates"] == [],
+              f"transcription returned, unknown firm -> no match/candidates (got {data})")
+        check(not any(k.startswith("_") for k in p), f"internal control keys stripped (got {list(p)})")
+
+        print("\n[email integrity: a model-minted address NOT in the card is dropped]")
+        _STATE["transcription"] = "Ann Roe\nDir\nOmega LP\nann@omega.fund"  # the only address present
+        _STATE["raw"] = {"intent": "new_investor", "investor_name": "Omega LP",
+                         "contact_name": "Ann Roe", "contact_email": "evil@phish.example"}
+        st, d = _req(port, "POST", "/api/intake/card", token, {"image_b64": _IMG})
+        p = (d or {}).get("data", {}).get("proposal", {})
+        check(p.get("contact_email") == "ann@omega.fund",
+              f"source address wins over the minted one (got {p.get('contact_email')})")
+
+        print("\n[match: exact firm name returns the grid row id]")
+        _STATE["transcription"] = "Jane Doe\nGP\nAcme Capital"  # no email -> match on name
+        _STATE["raw"] = {"intent": "new_investor", "investor_name": "Acme Capital",
+                         "contact_name": "Jane Doe", "contact_title": "GP"}
+        st, d = _req(port, "POST", "/api/intake/card", token, {"image_b64": _IMG})
+        m = (d or {}).get("data", {}).get("match")
+        check(m and m.get("id") == "rowAcme", f"exact firm -> match rowAcme (got {m})")
+
+        print("\n[match by card email: exact contact email returns the grid row id]")
+        _STATE["transcription"] = "Jane Doe\nGP\nAcme Capital Group\njane@acme.com"
+        _STATE["raw"] = {"intent": "new_investor", "investor_name": "Acme Capital Group",
+                         "contact_name": "Jane Doe"}
+        st, d = _req(port, "POST", "/api/intake/card", token, {"image_b64": _IMG})
+        m = (d or {}).get("data", {}).get("match")
+        check(m and m.get("id") == "rowAcme" and m.get("matched_on") == "email",
+              f"card email -> exact match rowAcme on email (got {m})")
+
+        print("\n[fuzzy: a near-spelling returns a candidate, no exact match]")
+        # Typo in the DISTINCTIVE token ('Acme'->'Acne') so the fuzzy matcher surfaces it; a typo
+        # in a generic descriptor (e.g. 'Capitol') wouldn't, since those are stripped first.
+        _STATE["transcription"] = "Jane Doe\nGP\nAcne Capital"  # no email -> name-only fuzzy
+        _STATE["raw"] = {"intent": "new_investor", "investor_name": "Acne Capital",
+                         "contact_name": "Jane Doe"}
+        st, d = _req(port, "POST", "/api/intake/card", token, {"image_b64": _IMG})
+        data = (d or {}).get("data", {})
+        cids = [c["id"] for c in data.get("candidates", [])]
+        check(data.get("match") is None and "rowAcme" in cids,
+              f"near-spelling -> candidate rowAcme, no exact (got {data})")
+
+        print("\n[no firm and no person: readable but unactionable -> ok:true, no DB lookup, no 500]")
+        _STATE["transcription"] = "some faded scribbles, no usable fields"  # >=5 chars, no email/firm
+        _STATE["raw"] = {"intent": "unclear"}
+        st, d = _req(port, "POST", "/api/intake/card", token, {"image_b64": _IMG})
+        data = (d or {}).get("data", {})
+        check(st == 200 and data.get("ok") is True
+              and data.get("match") is None and data.get("candidates") == [],
+              f"unclear proposal -> ok:true, no match/candidates, not 500 (got {st}, {data})")
+
+        print("\n[parse leg down: parse_message raises -> 502/vision_unavailable]")
+        _orig_pm = parse.parse_message
+        parse.parse_message = lambda *a, **k: (_ for _ in ()).throw(RuntimeError("qwen down"))
+        try:
+            _STATE["transcription"] = "Jane Doe\nGP\nAcme Capital"
+            st, d = _req(port, "POST", "/api/intake/card", token, {"image_b64": _IMG})
+            data = (d or {}).get("data", {})
+            check(st == 502 and data.get("reason") == "vision_unavailable",
+                  f"parse error -> 502 vision_unavailable (got {st}, {data})")
+        finally:
+            parse.parse_message = _orig_pm
+
+        print("\n[unreadable: model saw no card -> ok:false/unreadable, 200]")
+        _STATE["transcription"] = ""  # transcribe_card returns '' on the NONE sentinel
+        st, d = _req(port, "POST", "/api/intake/card", token, {"image_b64": _IMG})
+        data = (d or {}).get("data", {})
+        check(st == 200 and data.get("ok") is False and data.get("reason") == "unreadable",
+              f"empty transcription -> unreadable (got {st}, {data})")
+
+        print("\n[vision down: transcribe raises -> 502/vision_unavailable]")
+        _STATE["boom"] = True
+        st, d = _req(port, "POST", "/api/intake/card", token, {"image_b64": _IMG})
+        data = (d or {}).get("data", {})
+        check(st == 502 and data.get("reason") == "vision_unavailable",
+              f"spark error -> 502 vision_unavailable (got {st}, {data})")
+        _STATE["boom"] = False
+
+        print("\n[data-URI tolerated: a full data: prefix is stripped to raw base64]")
+        _STATE["transcription"] = "Sam Lee\nPartner\nBeacon Ventures"
+        _STATE["raw"] = {"intent": "new_investor", "investor_name": "Beacon Ventures",
+                         "contact_name": "Sam Lee"}
+        st, d = _req(port, "POST", "/api/intake/card", token,
+                     {"image_b64": "data:image/jpeg;base64," + _IMG})
+        check(st == 200 and (d or {}).get("data", {}).get("ok") is True,
+              f"data-URI accepted (got {st})")
+
+        print("\n[guard: missing image -> 400]")
+        st, _ = _req(port, "POST", "/api/intake/card", token, {})
+        check(st == 400, f"no image_b64 -> 400 (got {st})")
+
+        print("\n[guard: malformed base64 -> 400]")
+        st, _ = _req(port, "POST", "/api/intake/card", token, {"image_b64": "%%%not base64%%%"})
+        check(st == 400, f"invalid base64 -> 400 (got {st})")
+
+        print("\n[guard: oversized image -> 413 (size check runs before decode)]")
+        st, _ = _req(port, "POST", "/api/intake/card", token, {"image_b64": "A" * 12_000_001})
+        check(st == 413, f"over the 12 MB b64 cap -> 413 (got {st})")
+
+        print("\n[guard: unauthenticated -> 401]")
+        st, _ = _req(port, "POST", "/api/intake/card", None, {"image_b64": _IMG})
+        check(st == 401, f"no token -> 401 (got {st})")
+
+        print("\n[provenance: the approve write reuses log-communication tagged source=app_card]")
+        st, d = _req(port, "POST", "/api/fundraising/log-communication", token, {
+            "investor_name": "Beacon Ventures",
+            "contact": {"name": "Sam Lee", "email": "sam@beacon.vc", "title": "Partner"},
+            "create_investor_if_missing": True,
+            "type": "note", "subject": "", "body": "scanned business card",
+            "source": "app_card",
+        })
+        check(st in (200, 201), f"app_card create -> 201 (got {st})")
+        c = sqlite3.connect(os.environ["CRM_DB_PATH"])
+        rows = c.execute("SELECT changes FROM audit_log WHERE entity_type='communication' AND action='create'").fetchall()
+        c.close()
+        sources = [json.loads(r[0]).get("source") for r in rows if r[0]]
+        check("app_card" in sources, f"audit carries source=app_card (got {sources})")
+    finally:
+        httpd.shutdown()
+
+    print()
+    if FAILS:
+        print(f"FAILED ({len(FAILS)}):")
+        for f in FAILS:
+            print(f"  - {f}")
+        sys.exit(1)
+    print("ALL PASS (in-app card intake endpoint)")
+
+
+if __name__ == "__main__":
+    main()

backend/test_intake_endpoints.py

@@ -0,0 +1,259 @@
+#!/usr/bin/env python3
+"""Tests for the Matrix-intake CRM surface (v0.1.0 Matrix-intake M2).
+
+The bot adds no parallel write path — it reuses /api/fundraising/log-communication and adds
+one read-only lookup, GET /api/intake/match. This boots the REAL server against a temp DB and
+asserts:
+  - match by normalized name and by contact email, returning the GRID ROW id;
+  - the new-vs-existing contract: a bot-style create (log-communication +
+    create_investor_if_missing) then matches by name — so an approved note lands on that same
+    investor instead of duplicating it;
+  - provenance: an intake-sourced communication is audited with source="matrix_intake";
+  - guards: missing q/email -> 400, unauthenticated -> 401.
+Synthetic data only.
+
+Run: cd backend && python3 test_intake_endpoints.py
+"""
+import http.client
+import json
+import os
+import sqlite3
+import sys
+import tempfile
+import threading
+from http.server import ThreadingHTTPServer
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import server  # noqa: E402
+
+FAILS = []
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Quiet(server.CRMHandler):
+    def log_message(self, *a):
+        pass
+
+
+def _req(port, method, path, token=None, body=None):
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    headers = {}
+    if token:
+        headers["Authorization"] = "Bearer " + token
+    payload = None
+    if body is not None:
+        payload = json.dumps(body)
+        headers["Content-Type"] = "application/json"
+    conn.request(method, path, body=payload, headers=headers)
+    resp = conn.getresponse()
+    raw = resp.read().decode("utf-8", "replace")
+    conn.close()
+    data = None
+    if raw:
+        try:
+            data = json.loads(raw)
+        except ValueError:
+            pass
+    return resp.status, data
+
+
+GRID = {
+    "columns": [],
+    "rows": [
+        {"id": "rowAcme", "investor_name": "Acme Capital", "notes": "",
+         "contacts": [{"name": "Jane Doe", "email": "jane@acme.com", "title": "GP"}]},
+        {"id": "rowCharlie", "investor_name": "Charlie Brown", "notes": "",
+         "contacts": [{"name": "Charlie Brown", "email": "cb@brown.fund", "title": ""}]},
+        {"id": "rowBeta", "investor_name": "Beta Capital LLC", "notes": "",
+         "contacts": [{"name": "Pat Roe", "email": "pat@beta.com", "title": ""}]},
+        # Generic-descriptor decoys: share only "investment group" / "investments" with the
+        # Fortitude card below — must NOT surface as look-alikes (the 2026-06-20 false-positive fix).
+        {"id": "rowAether", "investor_name": "Aether Investment Group", "notes": "",
+         "contacts": [{"name": "Ada Ng", "email": "ada@aether.com", "title": ""}]},
+        {"id": "rowRussell", "investor_name": "Russell Investments", "notes": "",
+         "contacts": [{"name": "Russ Lee", "email": "russ@russell.com", "title": ""}]},
+    ],
+}
+
+
+def seed():
+    c = sqlite3.connect(os.environ["CRM_DB_PATH"])
+    c.execute("INSERT INTO users (id,username,email,password_hash,full_name,role,is_active) "
+              "VALUES ('u1','grant','grant@ten31.example','x','Grant','admin',1)")
+    # init_db doesn't create the 'main' state row (it's created lazily on first write), so
+    # upsert rather than UPDATE — a plain UPDATE would silently match zero rows.
+    c.execute("INSERT INTO fundraising_state (id, grid_json, views_json, version) "
+              "VALUES ('main', ?, '[]', 1) "
+              "ON CONFLICT(id) DO UPDATE SET grid_json = excluded.grid_json", (json.dumps(GRID),))
+    c.commit()
+    c.close()
+
+
+def main():
+    server.init_db()
+    seed()
+    token = server.create_token("u1", "grant", "admin")
+
+    # Unit: the distinctive-token similarity edges (the all-generic fallback path the endpoint
+    # seed can't naturally reach — no real investor is named purely with generic descriptors).
+    print("\n[unit: _name_similarity distinctive-token edges]")
+    sim = server._name_similarity
+    check(sim("Fortitude Investment Group", "Aether Investment Group") < 0.62,
+          f"generic-only overlap stays below threshold (got {sim('Fortitude Investment Group', 'Aether Investment Group'):.2f})")
+    check(sim("Aether Capital", "Aether Capital Partners") == 1.0,
+          f"distinctive 'aether' (generic descriptors stripped) scores 1.0 (got {sim('Aether Capital', 'Aether Capital Partners'):.2f})")
+    # Both sides all-generic → fallback compares full tokens on BOTH sides; shared generic word
+    # alone must not clear the bar.
+    check(sim("Capital Group", "Global Capital") < 0.62,
+          f"all-generic both sides stays below threshold (got {sim('Capital Group', 'Global Capital'):.2f})")
+    check(sim("Family Office", "Family Office") == 1.0,
+          "identical all-generic names still score 1.0 (early-out)")
+
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), _Quiet)
+    port = httpd.server_address[1]
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        print("\n[match: existing investor by name returns the grid row id]")
+        st, d = _req(port, "GET", "/api/intake/match?q=Acme%20Capital", token)
+        m = (d or {}).get("data", {}).get("match")
+        check(st == 200 and m and m["id"] == "rowAcme" and m["matched_on"] == "name",
+              f"name match -> rowAcme (got {st}, {m})")
+
+        print("\n[match: case-insensitive name]")
+        st, d = _req(port, "GET", "/api/intake/match?q=acme%20capital", token)
+        m = (d or {}).get("data", {}).get("match")
+        check(m and m["id"] == "rowAcme", f"normalized name match (got {m})")
+
+        print("\n[match: by contact email]")
+        st, d = _req(port, "GET", "/api/intake/match?email=jane@acme.com", token)
+        m = (d or {}).get("data", {}).get("match")
+        check(m and m["id"] == "rowAcme" and m["matched_on"] == "email",
+              f"email match -> rowAcme (got {m})")
+
+        print("\n[match: unknown -> null]")
+        st, d = _req(port, "GET", "/api/intake/match?q=Nobody%20LP", token)
+        check(st == 200 and (d or {}).get("data", {}).get("match") is None,
+              f"no match -> null (got {st}, {d})")
+
+        print("\n[fuzzy: exact match returns no candidates (bot auto-attaches)]")
+        st, d = _req(port, "GET", "/api/intake/match?q=Acme%20Capital", token)
+        data = (d or {}).get("data", {})
+        check(st == 200 and data.get("match") and data.get("candidates") == [],
+              f"exact match -> match set, candidates empty (got {data})")
+
+        print("\n[fuzzy: near-spelling surfaces a candidate (Charles Brown ~ Charlie Brown)]")
+        st, d = _req(port, "GET", "/api/intake/match?q=Charles%20Brown", token)
+        data = (d or {}).get("data", {})
+        cids = [c["id"] for c in data.get("candidates", [])]
+        check(data.get("match") is None and "rowCharlie" in cids,
+              f"near-spelling -> candidate rowCharlie, no exact (got {data})")
+
+        print("\n[fuzzy: legal-suffix difference surfaces a candidate (Beta Capital ~ Beta Capital LLC)]")
+        st, d = _req(port, "GET", "/api/intake/match?q=Beta%20Capital", token)
+        data = (d or {}).get("data", {})
+        cids = [c["id"] for c in data.get("candidates", [])]
+        check(data.get("match") is None and "rowBeta" in cids,
+              f"legal-suffix -> candidate rowBeta, no exact (got {data})")
+
+        print("\n[fuzzy: legal-suffix-only difference ranks as a top candidate (Acme Capital LLC ~ Acme Capital)]")
+        st, d = _req(port, "GET", "/api/intake/match?q=Acme%20Capital%20LLC", token)
+        data = (d or {}).get("data", {})
+        top = (data.get("candidates") or [None])[0]
+        check(data.get("match") is None and top and top["id"] == "rowAcme" and top["score"] == 1.0,
+              f"legal-suffix-only -> rowAcme top candidate @1.0, no exact (got {data})")
+
+        print("\n[fuzzy: one-character email typo surfaces a candidate by email]")
+        st, d = _req(port, "GET", "/api/intake/match?email=jhane@acme.com", token)
+        data = (d or {}).get("data", {})
+        cands = data.get("candidates", [])
+        hit = next((c for c in cands if c["id"] == "rowAcme"), None)
+        check(data.get("match") is None and hit and hit["matched_on"] == "email",
+              f"email typo -> candidate rowAcme matched_on email (got {data})")
+
+        print("\n[fuzzy: two-character email typo (distance 2) still surfaces]")
+        st, d = _req(port, "GET", "/api/intake/match?email=jane@acne.con", token)  # acme->acne, com->con
+        data = (d or {}).get("data", {})
+        hit = next((c for c in data.get("candidates", []) if c["id"] == "rowAcme"), None)
+        check(data.get("match") is None and hit and hit["matched_on"] == "email" and hit["score"] == 0.8,
+              f"dist-2 email -> rowAcme @0.8 (got {data})")
+
+        print("\n[fuzzy: a row matching on BOTH name and email appears once (deduped)]")
+        st, d = _req(port, "GET", "/api/intake/match?q=Acme%20Capitol&email=jhane@acme.com", token)
+        data = (d or {}).get("data", {})
+        acme_hits = [c for c in data.get("candidates", []) if c["id"] == "rowAcme"]
+        check(data.get("match") is None and len(acme_hits) == 1,
+              f"name+email both match rowAcme -> single deduped entry (got {data})")
+
+        print("\n[fuzzy: nothing close -> empty candidates]")
+        st, d = _req(port, "GET", "/api/intake/match?q=Zphq%20Nobody%20LP", token)
+        data = (d or {}).get("data", {})
+        check(st == 200 and data.get("match") is None and data.get("candidates") == [],
+              f"unrelated query -> no match, no candidates (got {data})")
+
+        print("\n[fuzzy: shared generic words alone do NOT surface look-alikes (Fortitude vs Aether/Russell)]")
+        st, d = _req(port, "GET", "/api/intake/match?q=Fortitude%20Investment%20Group", token)
+        data = (d or {}).get("data", {})
+        cids = [c["id"] for c in data.get("candidates", [])]
+        check(data.get("match") is None and "rowAether" not in cids and "rowRussell" not in cids,
+              f"generic-only overlap -> no decoy candidates (got {data})")
+
+        print("\n[fuzzy: a shared DISTINCTIVE word still surfaces (Aether Capital ~ Aether Investment Group)]")
+        st, d = _req(port, "GET", "/api/intake/match?q=Aether%20Capital", token)
+        data = (d or {}).get("data", {})
+        cids = [c["id"] for c in data.get("candidates", [])]
+        check(data.get("match") is None and "rowAether" in cids,
+              f"distinctive overlap -> rowAether candidate (got {data})")
+
+        print("\n[match: missing q and email -> 400]")
+        st, _ = _req(port, "GET", "/api/intake/match", token)
+        check(st == 400, f"no params -> 400 (got {st})")
+
+        print("\n[match: unauthenticated -> 401]")
+        st, _ = _req(port, "GET", "/api/intake/match?q=Acme", None)
+        check(st == 401, f"no token -> 401 (got {st})")
+
+        print("\n[bot create: log-communication + create_investor_if_missing, source tagged]")
+        st, d = _req(port, "POST", "/api/fundraising/log-communication", token, {
+            "investor_name": "Beacon Ventures",
+            "contact": {"name": "Sam Lee", "email": "sam@beacon.vc", "title": "Partner"},
+            "create_investor_if_missing": True,
+            "type": "note", "subject": "Intake (Matrix)", "body": "met at the Austin conf",
+            "source": "matrix_intake",
+        })
+        check(st in (200, 201), f"create new investor -> 201 (got {st})")
+
+        print("\n[new-vs-existing contract: the just-created investor now matches by name]")
+        st, d = _req(port, "GET", "/api/intake/match?q=Beacon%20Ventures", token)
+        m = (d or {}).get("data", {}).get("match")
+        check(m and m.get("investor_name") == "Beacon Ventures",
+              f"created investor is matchable (no duplicate on next note) (got {m})")
+
+        print("\n[provenance: the intake communication is audited as source=matrix_intake]")
+        c = sqlite3.connect(os.environ["CRM_DB_PATH"])
+        rows = c.execute("SELECT changes FROM audit_log WHERE entity_type='communication' AND action='create'").fetchall()
+        c.close()
+        sources = [json.loads(r[0]).get("source") for r in rows if r[0]]
+        check("matrix_intake" in sources, f"audit carries source=matrix_intake (got {sources})")
+    finally:
+        httpd.shutdown()
+
+    print()
+    if FAILS:
+        print(f"FAILED ({len(FAILS)}):")
+        for f in FAILS:
+            print(f"  - {f}")
+        sys.exit(1)
+    print("ALL PASS (matrix-intake endpoints)")
+
+
+if __name__ == "__main__":
+    main()

backend/test_pipeline_stages_v2.py

@@ -0,0 +1,149 @@
+#!/usr/bin/env python3
+"""Regression test for Phase 0 — the 4-stage pipeline funnel + the derived grid signals
+(ROADMAP "Pipeline stages + investor flags/labels -- LOCKED SPEC", 2026-06-19).
+
+Covers the parts the round-trip test in test_grid_pipeline_link.py only checks structurally:
+
+  * migration 0007 stage remap — outreach/meeting -> engaged, due_diligence -> diligence,
+    committed/funded -> commitment, and the stray 'lost' value archived (soft-deleted), with
+    'lead' left untouched;
+  * existing_investor_by_source_row — total_invested > 0 is the auto-derived "Existing Investor";
+  * staleness_by_source_row — last-contact age maps to '' / 'aging' (>= 30d) / 'stale' (>= 60d),
+    boundaries inclusive, and NO recorded activity -> '' (no false "stale" on a fresh lead).
+
+Synthetic only (guardrail #9). Run: cd backend && python3 test_pipeline_stages_v2.py
+"""
+import os
+import sqlite3
+import sys
+import tempfile
+from datetime import datetime, timedelta
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import server  # noqa: E402
+
+FAILS = []
+_HERE = os.path.dirname(os.path.abspath(__file__))
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+def _days_ago(n):
+    return (datetime.utcnow() - timedelta(days=n)).strftime("%Y-%m-%d")
+
+
+def test_migration_remap(conn):
+    print("\n[migration 0007: stage remap + lost archived]")
+    c = conn.cursor()
+    olds = [("m_out", "outreach"), ("m_meet", "meeting"), ("m_dd", "due_diligence"),
+            ("m_comm", "committed"), ("m_fund", "funded"), ("m_lost", "lost"),
+            ("m_lead", "lead")]
+    for oid, stage in olds:
+        c.execute("INSERT INTO opportunities (id, name, stage, contact_id, owner_id) VALUES (?,?,?,?,?)",
+                  (oid, oid, stage, "c0", "u1"))
+    conn.commit()
+    # init_db already applied 0007; re-run its SQL against the rows we just inserted with
+    # legacy values (the UPDATEs target by old stage value, so this is exactly the remap).
+    with open(os.path.join(_HERE, "migrations", "0007_pipeline_stages_v2.sql")) as fh:
+        conn.executescript(fh.read())
+    conn.commit()
+
+    def stage_of(oid):
+        return c.execute("SELECT stage FROM opportunities WHERE id=?", (oid,)).fetchone()[0]
+
+    def archived(oid):
+        return c.execute("SELECT deleted_at FROM opportunities WHERE id=?", (oid,)).fetchone()[0] is not None
+
+    check(stage_of("m_out") == "engaged" and stage_of("m_meet") == "engaged",
+          "outreach + meeting -> engaged")
+    check(stage_of("m_dd") == "diligence", "due_diligence -> diligence")
+    check(stage_of("m_comm") == "commitment" and stage_of("m_fund") == "commitment",
+          "committed + funded -> commitment")
+    check(stage_of("m_lead") == "lead", "lead unchanged")
+    check(archived("m_lost"), "lost opp archived (deleted_at set)")
+    check(not archived("m_lead") and not archived("m_comm"),
+          "non-lost opps NOT archived by the migration")
+    # Cleanup so these rows don't perturb the derivation seed below.
+    c.execute("DELETE FROM opportunities WHERE id IN ('m_out','m_meet','m_dd','m_comm','m_fund','m_lost','m_lead')")
+    conn.commit()
+
+
+def _investor(conn, srid, total, contact_id=None, comm_days_ago=None):
+    """Seed a grid investor (+ optional linked contact & dated communication)."""
+    iid = "i_" + srid
+    conn.execute("INSERT INTO fundraising_investors (id, investor_name, source_row_id, total_invested) "
+                 "VALUES (?,?,?,?)", (iid, srid, srid, total))
+    if contact_id:
+        conn.execute("INSERT INTO contacts (id, first_name, last_name) VALUES (?,?,?)",
+                     (contact_id, srid, "Person"))
+        conn.execute("INSERT INTO fundraising_contacts (id, investor_id, full_name, contact_id) "
+                     "VALUES (?,?,?,?)", ("fc_" + srid, iid, srid + " Person", contact_id))
+        if comm_days_ago is not None:
+            conn.execute("INSERT INTO communications (id, contact_id, communication_date, created_by, subject) "
+                         "VALUES (?,?,?,?,?)",
+                         ("cm_" + srid, contact_id, _days_ago(comm_days_ago), "u1", "note"))
+
+
+def test_derivations(conn):
+    print("\n[existing_investor + staleness derivations]")
+    # Existing flag: only total_invested > 0.
+    _investor(conn, "rowExist", 5_000_000, contact_id="c_exist", comm_days_ago=100)
+    _investor(conn, "rowProspect", 0)
+    # Staleness ramp + boundaries (>=30 aging, >=60 stale; inclusive).
+    _investor(conn, "rowStale", 0, contact_id="c_stale", comm_days_ago=70)
+    _investor(conn, "rowAging", 0, contact_id="c_aging", comm_days_ago=45)
+    _investor(conn, "rowFresh", 0, contact_id="c_fresh", comm_days_ago=5)
+    _investor(conn, "rowNoAct", 0)
+    _investor(conn, "rowB60", 0, contact_id="c_b60", comm_days_ago=60)   # boundary -> stale
+    _investor(conn, "rowB59", 0, contact_id="c_b59", comm_days_ago=59)   # -> aging
+    _investor(conn, "rowB30", 0, contact_id="c_b30", comm_days_ago=30)   # boundary -> aging
+    _investor(conn, "rowB29", 0, contact_id="c_b29", comm_days_ago=29)   # -> fresh
+    # Empty source_row_id with committed capital — must be EXCLUDED by the `if not srid` guard
+    # (would otherwise key the injection under '' and clobber a real row).
+    _investor(conn, "", 9_999, contact_id="c_empty", comm_days_ago=100)
+    conn.commit()
+
+    existing = server.existing_investor_by_source_row(conn)
+    check(existing == {"rowExist"},
+          f"existing_investor = total_invested>0 with a non-empty source_row_id only (got {sorted(existing)})")
+
+    st = server.staleness_by_source_row(conn)
+    level = lambda srid: st.get(srid, (None, "MISSING"))[1]
+    check(level("rowStale") == "stale", f"70d -> stale (got {level('rowStale')})")
+    check(level("rowAging") == "aging", f"45d -> aging (got {level('rowAging')})")
+    check(level("rowFresh") == "", f"5d -> fresh/'' (got {level('rowFresh')!r})")
+    check(level("rowNoAct") == "", f"no activity -> '' (got {level('rowNoAct')!r})")
+    check(level("rowExist") == "stale", "existing + stale coexist (orthogonal axes)")
+    check(level("rowB60") == "stale" and level("rowB59") == "aging",
+          f"60d boundary inclusive -> stale; 59d -> aging (got {level('rowB60')}, {level('rowB59')})")
+    check(level("rowB30") == "aging" and level("rowB29") == "",
+          f"30d boundary inclusive -> aging; 29d -> '' (got {level('rowB30')}, {level('rowB29')!r})")
+    # last_activity_at is carried through alongside the level for the recency display.
+    check(st.get("rowStale", (None, ""))[0] is not None, "stale row carries a last_activity_at value")
+    check(st.get("rowNoAct", ("X", ""))[0] is None, "no-activity row has last_activity_at None")
+
+
+def main():
+    server.init_db()
+    conn = server.get_db()
+    conn.execute("INSERT INTO users (id,username,email,password_hash,full_name,role,is_active) "
+                 "VALUES ('u1','grant','grant@ten31.example','x','Grant','admin',1)")
+    conn.execute("INSERT INTO contacts (id, first_name, last_name) VALUES ('c0','Seed','Contact')")
+    conn.commit()
+    test_migration_remap(conn)
+    test_derivations(conn)
+    conn.close()
+    print("\n" + ("ALL PASS (pipeline stages v2)" if not FAILS else f"{len(FAILS)} FAILED"))
+    sys.exit(1 if FAILS else 0)
+
+
+if __name__ == "__main__":
+    main()

backend/test_purge_soft_deleted.py

@@ -0,0 +1,157 @@
+#!/usr/bin/env python3
+"""Test for the admin soft-deleted purge (v0.1.0:104).
+
+The purge is a deliberate, admin-only, type-to-confirm exception to never-hard-delete, for
+clearing dummy/test data. It must be SAFE: only ever touch a soft-deleted row, and never
+remove or mutate LIVE data via a cascade/SET-NULL. This boots the real server, seeds live +
+soft-deleted graphs, and drives /api/admin/soft-deleted[/purge] over HTTP. Synthetic only.
+
+Run: cd backend && python3 test_purge_soft_deleted.py
+"""
+import http.client
+import json
+import os
+import sqlite3
+import sys
+import tempfile
+import threading
+from http.server import ThreadingHTTPServer
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import server  # noqa: E402
+
+FAILS = []
+DEL = "2026-06-01T00:00:00"
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Quiet(server.CRMHandler):
+    def log_message(self, *a):
+        pass
+
+
+def _post(port, path, token, payload):
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    conn.request("POST", path, body=json.dumps(payload),
+                 headers={"Authorization": "Bearer " + token, "Content-Type": "application/json"})
+    resp = conn.getresponse()
+    body = resp.read().decode("utf-8", "replace")
+    conn.close()
+    try:
+        return resp.status, (json.loads(body) if body else None)
+    except ValueError:
+        return resp.status, None
+
+
+def _get(port, path, token):
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    conn.request("GET", path, headers={"Authorization": "Bearer " + token})
+    resp = conn.getresponse()
+    body = resp.read().decode("utf-8", "replace")
+    conn.close()
+    try:
+        return resp.status, (json.loads(body) if body else None)
+    except ValueError:
+        return resp.status, None
+
+
+def exists(table, rid):
+    c = sqlite3.connect(os.environ["CRM_DB_PATH"])
+    n = c.execute(f"SELECT COUNT(*) FROM {table} WHERE id = ?", (rid,)).fetchone()[0]
+    c.close()
+    return n > 0
+
+
+def seed():
+    c = sqlite3.connect(os.environ["CRM_DB_PATH"])
+    c.execute("INSERT INTO users (id,username,email,password_hash,full_name,role,is_active) "
+              "VALUES ('u1','grant','grant@ten31.example','x','Grant','admin',1)")
+    # Soft-deleted contact with ONLY soft-deleted children -> purgeable; cascade should remove them.
+    c.execute("INSERT INTO contacts (id,first_name,last_name,deleted_at) VALUES ('cClean','Dummy','Clean',?)", (DEL,))
+    c.execute("INSERT INTO opportunities (id,name,contact_id,owner_id,deleted_at) VALUES ('opC','Opp','cClean','u1',?)", (DEL,))
+    c.execute("INSERT INTO communications (id,contact_id,communication_date,created_by,subject,deleted_at) VALUES ('cmC','cClean','2026-05-01','u1','note',?)", (DEL,))
+    # A reminder pointing at the purge target (reminders.contact_id is a bare logical FK, no ON DELETE):
+    # the purge must NULL it, not leave it dangling and not delete the reminder.
+    c.execute("INSERT INTO reminders (id,contact_id,investor_id,title) VALUES ('remC','cClean','inv-x','Follow up dummy')")
+    # Soft-deleted contact WITH a live child -> must refuse (cascade would kill live data).
+    c.execute("INSERT INTO contacts (id,first_name,last_name,deleted_at) VALUES ('cLiveKid','Has','Livekid',?)", (DEL,))
+    c.execute("INSERT INTO communications (id,contact_id,communication_date,created_by,subject) VALUES ('cmLive','cLiveKid','2026-05-02','u1','live note')")
+    # A live contact -> must refuse (not soft-deleted).
+    c.execute("INSERT INTO contacts (id,first_name,last_name) VALUES ('cLive','Real','Person')")
+    # Soft-deleted org with no live refs -> purgeable.
+    c.execute("INSERT INTO organizations (id,name,deleted_at) VALUES ('orgClean','Dead Org',?)", (DEL,))
+    # Soft-deleted org referenced by a LIVE contact -> must refuse (SET NULL would mutate live data).
+    c.execute("INSERT INTO organizations (id,name,deleted_at) VALUES ('orgRef','Ref Org',?)", (DEL,))
+    c.execute("INSERT INTO contacts (id,first_name,last_name,organization_id) VALUES ('cRef','Org','Member','orgRef')")
+    c.commit()
+    c.close()
+
+
+def main():
+    server.init_db()
+    seed()
+    token = server.create_token("u1", "grant", "admin")
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), _Quiet)
+    port = httpd.server_address[1]
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        print("\n[list soft-deleted]")
+        st, body = _get(port, "/api/admin/soft-deleted", token)
+        groups = (body or {}).get("groups", {})
+        cids = {x["id"] for x in groups.get("contacts", [])}
+        oids = {x["id"] for x in groups.get("organizations", [])}
+        check(st == 200, f"GET soft-deleted -> 200 (got {st})")
+        check({"cClean", "cLiveKid"} <= cids and "cLive" not in cids, f"lists soft-deleted contacts only (got {cids})")
+        check({"orgClean", "orgRef"} <= oids, f"lists soft-deleted orgs (got {oids})")
+        check("opC" in {x["id"] for x in groups.get("opportunities", [])}, "lists the soft-deleted opportunity")
+
+        print("\n[purge guards]")
+        st, _ = _post(port, "/api/admin/soft-deleted/purge", token, {"table": "contacts", "id": "cLive"})
+        check(st == 400, f"purge a LIVE contact -> 400 (got {st})")
+        check(exists("contacts", "cLive"), "live contact still present after refused purge")
+        st, _ = _post(port, "/api/admin/soft-deleted/purge", token, {"table": "contacts", "id": "cLiveKid"})
+        check(st == 409, f"purge contact with a LIVE child -> 409 (got {st})")
+        check(exists("contacts", "cLiveKid") and exists("communications", "cmLive"), "contact + its live child preserved")
+        st, _ = _post(port, "/api/admin/soft-deleted/purge", token, {"table": "organizations", "id": "orgRef"})
+        check(st == 409, f"purge org referenced by a LIVE contact -> 409 (got {st})")
+        check(exists("organizations", "orgRef") and exists("contacts", "cRef"), "org + its live referencing contact preserved")
+        st, _ = _post(port, "/api/admin/soft-deleted/purge", token, {"table": "bogus", "id": "x"})
+        check(st == 400, f"unknown table -> 400 (got {st})")
+        st, _ = _post(port, "/api/admin/soft-deleted/purge", token, {"table": "contacts", "id": "nope"})
+        check(st == 404, f"missing id -> 404 (got {st})")
+
+        print("\n[purge happy path + cascade]")
+        st, _ = _post(port, "/api/admin/soft-deleted/purge", token, {"table": "contacts", "id": "cClean"})
+        check(st == 200, f"purge a clean soft-deleted contact -> 200 (got {st})")
+        check(not exists("contacts", "cClean"), "purged contact is gone")
+        check(not exists("opportunities", "opC") and not exists("communications", "cmC"),
+              "its soft-deleted children were cascade-removed")
+        _rc = sqlite3.connect(os.environ["CRM_DB_PATH"])
+        _rem = _rc.execute("SELECT contact_id FROM reminders WHERE id = 'remC'").fetchone()
+        _rc.close()
+        check(_rem is not None and _rem[0] is None,
+              "a reminder on the purged contact is KEPT but its contact_id is NULL'd (no dangling ref)")
+        st, _ = _post(port, "/api/admin/soft-deleted/purge", token, {"table": "organizations", "id": "orgClean"})
+        check(st == 200, f"purge a clean soft-deleted org -> 200 (got {st})")
+        check(not exists("organizations", "orgClean"), "purged org is gone")
+    finally:
+        httpd.shutdown()
+
+    print()
+    if FAILS:
+        print(f"{len(FAILS)} FAILED")
+        sys.exit(1)
+    print("ALL PASS (soft-deleted purge)")
+
+
+if __name__ == "__main__":
+    main()

backend/test_reminders.py

@@ -0,0 +1,303 @@
+#!/usr/bin/env python3
+"""Tests for reminders / follow-ups (W1).
+
+Boots the REAL server against a temp DB and exercises the new endpoints end-to-end:
+  - POST /api/reminders creates an open reminder tied to a grid investor (denormalized
+    investor_name resolved from the grid), or a standalone task (no investor_id);
+  - GET /api/reminders lists + filters by status (active/open/done/...), overdue, investor_id,
+    assignee=me; every read is soft-delete filtered;
+  - PATCH completes (stamps completed_at) / snoozes / edits a reminder; status is validated;
+  - DELETE soft-deletes (gone from every list, never hard-deleted);
+  - GET /api/fundraising/state injects a read-only reminder_status (overdue/due_soon/open/'')
+    derived live from open reminders, and strips it on save (never persisted to the blob);
+  - deleting an investor from the grid cancels its orphaned reminders (reconcile twin),
+    while a standalone reminder is untouched;
+  - the usual guards (missing title -> 400, bad status -> 400, unknown investor -> 404,
+    unauthenticated -> 401).
+Synthetic data only.
+
+Run: cd backend && python3 test_reminders.py
+"""
+import http.client
+import json
+import os
+import sqlite3
+import sys
+import tempfile
+import threading
+from datetime import datetime, timedelta
+from http.server import ThreadingHTTPServer
+
+_DATA = tempfile.mkdtemp()
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import server  # noqa: E402
+
+FAILS = []
+
+TODAY = datetime.utcnow().date()
+TOMORROW = (TODAY + timedelta(days=1)).isoformat()
+YESTERDAY = (TODAY - timedelta(days=1)).isoformat()
+FAR = (TODAY + timedelta(days=30)).isoformat()
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Quiet(server.CRMHandler):
+    def log_message(self, *a):
+        pass
+
+
+def _req(port, method, path, token=None, body=None):
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    headers = {}
+    if token:
+        headers["Authorization"] = "Bearer " + token
+    payload = None
+    if body is not None:
+        payload = json.dumps(body)
+        headers["Content-Type"] = "application/json"
+    conn.request(method, path, body=payload, headers=headers)
+    resp = conn.getresponse()
+    raw = resp.read().decode("utf-8", "replace")
+    conn.close()
+    data = None
+    if raw:
+        try:
+            data = json.loads(raw)
+        except ValueError:
+            pass
+    return resp.status, data
+
+
+def _put_grid(port, token, rows):
+    return _req(port, "PUT", "/api/fundraising/state", token,
+                {"grid": {"columns": [], "rows": rows}, "views": []})
+
+
+ROW_ACME = {"id": "rowAcme", "investor_name": "Acme Capital", "notes": "", "lead": "Grant",
+            "contacts": [{"name": "Jane Doe", "email": "jane@acme.com", "title": "GP"}]}
+ROW_BETA = {"id": "rowBeta", "investor_name": "Beta Capital", "notes": "",
+            "contacts": [{"name": "Pat Roe", "email": "pat@beta.com"}]}
+ROW_GAMMA = {"id": "rowGamma", "investor_name": "Gamma Partners", "notes": "",
+             "contacts": [{"name": "Sam Lee", "email": "sam@gamma.com"}]}
+
+
+def _db():
+    return sqlite3.connect(os.environ["CRM_DB_PATH"])
+
+
+def seed():
+    c = _db()
+    c.execute("INSERT INTO users (id,username,email,password_hash,full_name,role,is_active) "
+              "VALUES ('u1','grant','grant@ten31.example','x','Grant','admin',1)")
+    c.commit()
+    c.close()
+
+
+def _investor_id(source_row_id):
+    c = _db()
+    r = c.execute("SELECT id FROM fundraising_investors WHERE source_row_id = ?", (source_row_id,)).fetchone()
+    c.close()
+    return r[0] if r else None
+
+
+def _grid_reminder_status(port, token):
+    st, d = _req(port, "GET", "/api/fundraising/state", token)
+    rows = (d or {}).get("data", {}).get("grid", {}).get("rows", [])
+    return {r["id"]: r.get("reminder_status") for r in rows}
+
+
+def main():
+    server.init_db()
+    seed()
+    token = server.create_token("u1", "grant", "admin")
+
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), _Quiet)
+    port = httpd.server_address[1]
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        st, _ = _put_grid(port, token, [ROW_ACME, ROW_BETA, ROW_GAMMA])
+        check(st == 200, f"seed grid via PUT /state (got {st})")
+        acme_id = _investor_id("rowAcme")
+        beta_id = _investor_id("rowBeta")
+        gamma_id = _investor_id("rowGamma")
+        check(bool(acme_id and beta_id and gamma_id), "investor ids resolved from the grid")
+
+        # ── create: investor-linked + denormalized name resolved from the grid ──
+        print("\n[create: investor-linked reminder resolves the denormalized name]")
+        st, d = _req(port, "POST", "/api/reminders", token,
+                     {"investor_id": acme_id, "title": "Send Fund III deck", "due_date": TOMORROW})
+        rem = (d or {}).get("data") or {}
+        acme_rem_id = rem.get("id")
+        check(st == 201 and rem.get("status") == "open", f"create -> 201 open (got {st}, {d})")
+        check(rem.get("investor_name") == "Acme Capital", f"name denormalized from grid (got {rem.get('investor_name')})")
+
+        # overdue reminder on Beta; far-future + standalone for filter coverage
+        st, d = _req(port, "POST", "/api/reminders", token,
+                     {"investor_id": beta_id, "title": "Call Pat", "due_date": YESTERDAY})
+        beta_rem_id = (d or {}).get("data", {}).get("id")
+        check(st == 201, f"create overdue beta reminder (got {st})")
+        st, d = _req(port, "POST", "/api/reminders", token,
+                     {"investor_id": gamma_id, "title": "Quarterly touch", "due_date": FAR})
+        gamma_rem_id = (d or {}).get("data", {}).get("id")
+        st, d = _req(port, "POST", "/api/reminders", token, {"title": "Team: refresh pipeline view"})
+        standalone_id = (d or {}).get("data", {}).get("id")
+        check(st == 201 and (d or {}).get("data", {}).get("investor_id") in (None, ""),
+              f"standalone reminder (no investor) created (got {st})")
+
+        # ── list + filters ──
+        print("\n[list + filters]")
+        st, d = _req(port, "GET", "/api/reminders", token)
+        items = (d or {}).get("data", [])
+        check(st == 200 and len(items) == 4, f"list returns all 4 open (got {st}, {len(items)})")
+        check(all("last_activity_at" in it for it in items), "each row carries last_activity_at")
+        # 8h deep-link: each row injects source_row_id (joined via investor_id) — the grid row id
+        # the mobile Reminders edit sheet uses for "Open investor in Grid"; null for a team task.
+        by_id = {it.get("id"): it for it in items}
+        check(by_id.get(acme_rem_id, {}).get("source_row_id") == "rowAcme",
+              f"investor reminder injects source_row_id == 'rowAcme' (got {by_id.get(acme_rem_id, {}).get('source_row_id')!r})")
+        check(by_id.get(standalone_id, {}).get("source_row_id") in (None, ""),
+              f"team task has null source_row_id -> Open-in-Grid hidden (got {by_id.get(standalone_id, {}).get('source_row_id')!r})")
+        # dated reminders sort before undated, soonest first -> YESTERDAY (beta) leads
+        check(items and items[0].get("id") == beta_rem_id, f"overdue sorts first (got {items[0].get('id') if items else None})")
+
+        st, d = _req(port, "GET", "/api/reminders?overdue=1", token)
+        ids = [it["id"] for it in (d or {}).get("data", [])]
+        check(ids == [beta_rem_id], f"overdue=1 -> only the past-due one (got {ids})")
+        # overdue owns the status constraint: a conflicting status= must not silently zero out
+        st, d = _req(port, "GET", "/api/reminders?overdue=1&status=done", token)
+        ids = [it["id"] for it in (d or {}).get("data", [])]
+        check(ids == [beta_rem_id], f"overdue wins over a conflicting status= (got {ids})")
+
+        st, d = _req(port, "GET", f"/api/reminders?investor_id={acme_id}", token)
+        ids = [it["id"] for it in (d or {}).get("data", [])]
+        check(ids == [acme_rem_id], f"investor_id filter (got {ids})")
+
+        st, d = _req(port, "GET", "/api/reminders?assignee=me", token)
+        check(len(d.get("data", [])) == 0, "assignee=me -> none (reminders created unassigned)")
+
+        # ── grid injection: reminder_status derived live, never persisted ──
+        print("\n[grid injection: read-only reminder_status]")
+        s = _grid_reminder_status(port, token)
+        check(s.get("rowAcme") == "due_soon", f"due-tomorrow -> due_soon (got {s.get('rowAcme')})")
+        check(s.get("rowBeta") == "overdue", f"past-due -> overdue (got {s.get('rowBeta')})")
+        check(s.get("rowGamma") == "open", f"far-future -> open (got {s.get('rowGamma')})")
+        # echo the injected value back on save; it must NOT persist into the blob
+        st, d = _req(port, "GET", "/api/fundraising/state", token)
+        echoed = (d or {}).get("data", {}).get("grid", {}).get("rows", [])
+        st, _ = _put_grid(port, token, echoed)
+        c = _db()
+        blob = json.loads(c.execute("SELECT grid_json FROM fundraising_state WHERE id='main'").fetchone()[0])
+        c.close()
+        acme_stored = {r["id"]: r for r in blob.get("rows", [])}.get("rowAcme", {})
+        check("reminder_status" not in acme_stored, "reminder_status not persisted into the grid blob")
+
+        # ── complete / reopen stamps completed_at ──
+        print("\n[complete + reopen]")
+        st, d = _req(port, "PATCH", f"/api/reminders/{acme_rem_id}", token, {"status": "done"})
+        check(st == 200 and (d or {}).get("data", {}).get("status") == "done"
+              and (d or {}).get("data", {}).get("completed_at"), f"done stamps completed_at (got {d})")
+        st, d = _req(port, "GET", "/api/reminders?status=open", token)
+        check(acme_rem_id not in [it["id"] for it in (d or {}).get("data", [])], "done reminder drops from status=open")
+        st, d = _req(port, "GET", "/api/reminders?status=done", token)
+        check(acme_rem_id in [it["id"] for it in (d or {}).get("data", [])], "done reminder shows under status=done")
+        check(_grid_reminder_status(port, token).get("rowAcme") in (None, ""), "completed reminder clears the grid chip")
+        st, d = _req(port, "PATCH", f"/api/reminders/{acme_rem_id}", token, {"status": "open"})
+        check((d or {}).get("data", {}).get("completed_at") in (None, ""), "reopen clears completed_at")
+
+        # ── snooze: out of 'open', still in 'active' ──
+        print("\n[snooze]")
+        st, d = _req(port, "PATCH", f"/api/reminders/{beta_rem_id}", token,
+                     {"status": "snoozed", "snoozed_until": FAR})
+        check(st == 200 and (d or {}).get("data", {}).get("status") == "snoozed", f"snooze (got {st})")
+        st, d = _req(port, "GET", "/api/reminders?status=open", token)
+        check(beta_rem_id not in [it["id"] for it in (d or {}).get("data", [])], "snoozed drops from status=open")
+        st, d = _req(port, "GET", "/api/reminders?status=active", token)
+        check(beta_rem_id in [it["id"] for it in (d or {}).get("data", [])], "snoozed stays in status=active")
+
+        # ── edit title + due date ──
+        print("\n[edit]")
+        st, d = _req(port, "PATCH", f"/api/reminders/{gamma_rem_id}", token,
+                     {"title": "Quarterly touch — Q3", "due_date": TOMORROW})
+        check((d or {}).get("data", {}).get("title") == "Quarterly touch — Q3"
+              and (d or {}).get("data", {}).get("due_date") == TOMORROW, f"title+due edited (got {d})")
+
+        # ── soft-delete: gone from every list, tombstoned not hard-deleted ──
+        print("\n[soft-delete]")
+        st, d = _req(port, "DELETE", f"/api/reminders/{standalone_id}", token)
+        check(st == 200 and (d or {}).get("data", {}).get("deleted") is True, f"delete -> 200 (got {st})")
+        st, d = _req(port, "GET", "/api/reminders", token)
+        check(standalone_id not in [it["id"] for it in (d or {}).get("data", [])], "deleted reminder hidden from list")
+        c = _db()
+        gone = c.execute("SELECT deleted_at FROM reminders WHERE id = ?", (standalone_id,)).fetchone()[0]
+        c.close()
+        check(gone is not None, "deleted reminder tombstoned (deleted_at set), not hard-deleted")
+
+        # ── orphan reconcile: drop the investor from the grid -> its reminders cancelled ──
+        print("\n[orphan reconcile: deleting the grid investor cancels its reminders]")
+        # create a fresh standalone task to confirm it is NOT cancelled by the reconciler
+        st, d = _req(port, "POST", "/api/reminders", token, {"title": "Standalone keeps living"})
+        keep_id = (d or {}).get("data", {}).get("id")
+        st, _ = _put_grid(port, token, [ROW_ACME, ROW_BETA])  # drop rowGamma
+        c = _db()
+        gamma_status = c.execute("SELECT status FROM reminders WHERE id = ?", (gamma_rem_id,)).fetchone()[0]
+        keep_status = c.execute("SELECT status FROM reminders WHERE id = ?", (keep_id,)).fetchone()[0]
+        gamma_name = c.execute("SELECT investor_name FROM reminders WHERE id = ?", (gamma_rem_id,)).fetchone()[0]
+        c.close()
+        check(gamma_status == "cancelled", f"orphaned investor's reminder cancelled (got {gamma_status})")
+        check(gamma_name == "Gamma Partners", "cancelled reminder keeps denormalized investor_name for history")
+        check(keep_status == "open", f"standalone reminder untouched by reconcile (got {keep_status})")
+
+        # ── recency rollup: a tombstoned email sighting must not inflate last_activity_at ──
+        print("\n[recency: soft-deleted email sighting excluded from last_activity_at]")
+        c = _db()
+        c.execute("INSERT INTO emails (id, rfc_message_id, from_email, sent_at, is_matched, match_status) "
+                  "VALUES ('em1','<em1@x>','lp@acme.com','2026-06-10T00:00:00Z',1,'matched')")
+        c.execute("INSERT INTO email_investor_links (id, email_id, fundraising_investor_id, match_kind, match_confidence, matched_address) "
+                  "VALUES ('eil1','em1',?, 'exact_email',1.0,'lp@acme.com')", (acme_id,))
+        c.execute("INSERT INTO email_account_messages (id, email_id, account_id, gmail_message_id, gmail_thread_id, deleted_at) "
+                  "VALUES ('eam1','em1','acct1','g1','t1','2026-06-11T00:00:00Z')")  # tombstoned sighting only
+        c.commit(); c.close()
+        st, d = _req(port, "GET", f"/api/reminders?investor_id={acme_id}", token)
+        items = (d or {}).get("data", [])
+        la = items[0].get("last_activity_at") if items else "MISSING"
+        check(bool(items) and la is None, f"tombstoned-only email -> no last_activity (got {la})")
+        c = _db()
+        c.execute("INSERT INTO email_account_messages (id, email_id, account_id, gmail_message_id, gmail_thread_id, deleted_at) "
+                  "VALUES ('eam2','em1','acct2','g2','t2',NULL)")
+        c.commit(); c.close()
+        st, d = _req(port, "GET", f"/api/reminders?investor_id={acme_id}", token)
+        items = (d or {}).get("data", [])
+        la = items[0].get("last_activity_at") if items else "MISSING"
+        check(bool(items) and la == '2026-06-10T00:00:00Z', f"live sighting -> last_activity set (got {la})")
+
+        # ── guards ──
+        print("\n[guards]")
+        st, _ = _req(port, "POST", "/api/reminders", token, {"investor_id": acme_id})
+        check(st == 400, f"missing title -> 400 (got {st})")
+        st, _ = _req(port, "POST", "/api/reminders", token, {"title": "x", "investor_id": "nope"})
+        check(st == 404, f"unknown investor_id -> 404 (got {st})")
+        st, _ = _req(port, "PATCH", f"/api/reminders/{gamma_rem_id}", token, {"status": "bogus"})
+        check(st == 400, f"invalid status -> 400 (got {st})")
+        st, _ = _req(port, "DELETE", "/api/reminders/doesnotexist", token)
+        check(st == 404, f"delete unknown -> 404 (got {st})")
+        st, _ = _req(port, "GET", "/api/reminders", None)
+        check(st == 401, f"unauthenticated list -> 401 (got {st})")
+    finally:
+        httpd.shutdown()
+
+    print("\n" + ("ALL PASS" if not FAILS else f"{len(FAILS)} FAILURE(S):"))
+    for f in FAILS:
+        print("  - " + f)
+    sys.exit(1 if FAILS else 0)
+
+
+if __name__ == "__main__":
+    main()

backend/test_smtp_endpoint.py

@@ -0,0 +1,171 @@
+#!/usr/bin/env python3
+"""Endpoint tests for POST /api/admin/digest/test-email (handle_admin_send_test_email).
+
+Boots the REAL server in-process against a throwaway DB, monkeypatches the SMTP
+sender (no network), and proves the security-relevant contract:
+  * admin-gated: 401 without a token, 403 for a non-admin.
+  * recipients are restricted to the active-admin set — an arbitrary `to` is
+    rejected (the endpoint is NOT an open relay), an admin `to` is allowed,
+    and the default audience is every active admin.
+  * a send failure does not echo the exception (which could carry the SMTP
+    credential) back to the caller.
+
+Run: cd backend && python3 test_smtp_endpoint.py
+"""
+import http.client
+import json
+import os
+import sys
+import tempfile
+import threading
+from http.server import ThreadingHTTPServer
+
+_BASE = tempfile.mkdtemp()
+_FRONTEND = os.path.join(_BASE, "frontend")
+os.makedirs(os.path.join(_FRONTEND, "assets"))
+_DATA = os.path.join(_BASE, "data")
+os.makedirs(_DATA)
+with open(os.path.join(_FRONTEND, "index.html"), "w") as f:
+    f.write("<!doctype html><title>crm</title>")
+os.environ["CRM_FRONTEND_DIR"] = _FRONTEND
+os.environ["CRM_DATA_DIR"] = _DATA
+os.environ["CRM_DB_PATH"] = os.path.join(_DATA, "crm.db")
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import server  # noqa: E402
+import smtp_send  # noqa: E402
+
+FAILS = []
+
+# ── fake sender: record calls, optionally raise (no network) ──
+SENT = []
+RAISE = {"exc": None}
+
+
+def fake_send(to_addrs, subject, body, **kw):
+    if RAISE["exc"] is not None:
+        raise RAISE["exc"]
+    SENT.append({"to": list(to_addrs), "subject": subject})
+    return {"sent_to": list(to_addrs), "from": "digest@ten31.test"}
+
+
+smtp_send.send_email = fake_send
+smtp_send.smtp_configured = lambda: True
+
+
+def check(cond, msg):
+    print(("  PASS " if cond else "  FAIL ") + msg)
+    if not cond:
+        FAILS.append(msg)
+
+
+class _Quiet(server.CRMHandler):
+    def log_message(self, *a):
+        pass
+
+
+def _post(port, path, token=None, body=None):
+    conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
+    headers = {"Content-Type": "application/json"}
+    if token:
+        headers["Authorization"] = "Bearer " + token
+    conn.request("POST", path, body=json.dumps(body or {}), headers=headers)
+    resp = conn.getresponse()
+    raw = resp.read().decode("utf-8", "replace")
+    conn.close()
+    try:
+        parsed = json.loads(raw)
+    except Exception:
+        parsed = None
+    return resp.status, raw, parsed
+
+
+def main():
+    server.init_db()
+
+    # Admin = first registered user. A member is inserted directly (self-register
+    # is disabled after the first user) so get_user resolves a real active row.
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), _Quiet)
+    port = httpd.server_address[1]
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        st, raw, j = _post(port, "/api/auth/register", body={
+            "username": "admin", "email": "admin@ten31.test",
+            "password": "password123", "full_name": "Admin User"})
+        check(st == 201 and j and j.get("token"), f"register first user as admin (got {st})")
+        admin_token = j["token"]
+        admin_email = j["user"]["email"]
+
+        conn = server.get_db()
+        conn.execute(
+            "INSERT INTO users (id, username, email, password_hash, full_name, role, is_active) "
+            "VALUES (?,?,?,?,?,?,1)",
+            ("member-1", "member1", "member1@ten31.test",
+             server.hash_password("password123"), "Member One", "member"))
+        conn.commit()
+        conn.close()
+        member_token = server.create_token("member-1", "member1", "member")
+
+        path = "/api/admin/digest/test-email"
+
+        # 1. unauthenticated -> 401, sender untouched
+        SENT.clear()
+        st, raw, j = _post(port, path)
+        check(st == 401, f"no token -> 401 (got {st})")
+        check(not SENT, "no token: sender not called")
+
+        # 2. non-admin -> 403
+        SENT.clear()
+        st, raw, j = _post(port, path, token=member_token)
+        check(st == 403, f"member -> 403 (got {st})")
+        check(not SENT, "member: sender not called")
+
+        # 3. admin, no `to` -> 200, default audience = the admin set
+        SENT.clear()
+        st, raw, j = _post(port, path, token=admin_token)
+        check(st == 200, f"admin default -> 200 (got {st})")
+        check(len(SENT) == 1 and SENT[0]["to"] == [admin_email],
+              f"default recipients = active admins ({admin_email}); got {SENT}")
+
+        # 4. admin, arbitrary outside `to` -> 400, NOT an open relay
+        SENT.clear()
+        st, raw, j = _post(port, path, token=admin_token, body={"to": "attacker@evil.com"})
+        check(st == 400, f"outside `to` -> 400 (got {st})")
+        check(not SENT, "outside `to`: sender not called (no relay)")
+
+        # 5. admin, `to` an admin address (case-insensitive) -> 200
+        SENT.clear()
+        st, raw, j = _post(port, path, token=admin_token, body={"to": admin_email.upper()})
+        check(st == 200, f"admin `to` -> 200 (got {st})")
+        check(len(SENT) == 1 and SENT[0]["to"] == [admin_email.upper()],
+              "admin `to` is delivered as given")
+
+        # 6. mixed list with one outside address -> 400, all rejected
+        SENT.clear()
+        st, raw, j = _post(port, path, token=admin_token,
+                           body={"to": [admin_email, "outsider@evil.com"]})
+        check(st == 400, f"mixed list with outsider -> 400 (got {st})")
+        check(not SENT, "mixed list: sender not called")
+
+        # 7. send failure must NOT leak the exception text (could carry the credential)
+        SENT.clear()
+        secret = "PWLEAK_SsHh99"
+        RAISE["exc"] = Exception(f"535 auth failed for user with password {secret}")
+        st, raw, j = _post(port, path, token=admin_token)
+        RAISE["exc"] = None
+        check(st == 502, f"send failure -> 502 (got {st})")
+        check(secret not in raw, "send-failure response does not leak the exception/credential text")
+    finally:
+        httpd.shutdown()
+
+    print()
+    if FAILS:
+        print(f"FAILED ({len(FAILS)}):")
+        for f in FAILS:
+            print(f"  - {f}")
+        sys.exit(1)
+    print("ALL PASS (digest test-email endpoint)")
+
+
+if __name__ == "__main__":
+    main()