v0.2.8 operator dashboard with per-call audit log + cost tracking

server/routes/transcribe.js

@@ -21,6 +21,10 @@
 //     result: { text: "[MM:SS] ...", segments: [], duration_seconds: 0 },
 //     credits_remaining, tier, credit_charged
 //   }
+//
+// Every outcome (success / quota-refused / backend-error) writes one
+// row to the audit log so the admin dashboard can compute cost,
+// margin, and speed metrics.
 
 import express from "express";
 import multer from "multer";
@@ -31,6 +35,8 @@ import { getConfigSnapshot, getTierQuotas } from "../config.js";
 import { createGeminiBackend } from "../backends/gemini.js";
 import { createHardwareBackend } from "../backends/hardware.js";
 import { envelope, errorEnvelope } from "./envelope.js";
+import { recordCall } from "../audit-log.js";
+import { calcGeminiCost } from "../pricing.js";
 
 const upload = multer({
   storage: multer.memoryStorage(),
@@ -41,6 +47,7 @@ export function transcribeRouter() {
   const router = express.Router();
 
   router.post("/transcribe", upload.single("audio"), async (req, res) => {
+    const t0 = Date.now();
     const installId = req.header("X-Recap-Install-Id");
     const jobId = req.header("X-Recap-Job-Id") || null;
     const auth = req.header("Authorization");
@@ -60,14 +67,9 @@ export function transcribeRouter() {
     const license = await resolveLicense(auth);
     const tier = license.tier;
 
-    // Persist tier on the row so the admin dashboard reflects the
-    // most recently seen tier for this install.
     const row = await getOrCreateRow(installId);
     row.tier_snapshot = tier;
 
-    // Job-id dedup. If we've already charged this job, skip the
-    // credit check entirely — the user is paying once for the whole
-    // summarize job.
     let reusedJob = false;
     let chosenBackend = null;
     const existingJob = lookupJob(installId, jobId);
@@ -82,6 +84,19 @@ export function transcribeRouter() {
         cfg.relay_transcribe_backend_preference || "gemini_first";
       const plan = planBackend(row, quota, { hasHardware, preference });
       if (!plan.allowed) {
+        await recordCall({
+          install_id: installId,
+          tier,
+          pipeline: "transcribe",
+          backend: null,
+          model: null,
+          status: "refused",
+          credit_charged: 0,
+          duration_ms: Date.now() - t0,
+          cost_usd: 0,
+          job_id: jobId,
+          error: plan.reason,
+        });
         const e = await errorEnvelope({
           error: plan.reason,
           installId,
@@ -93,7 +108,6 @@ export function transcribeRouter() {
       chosenBackend = plan.backend;
     }
 
-    // Build the backend client based on chosenBackend.
     const cfg = await getConfigSnapshot();
     let result;
     try {
@@ -126,10 +140,23 @@ export function transcribeRouter() {
         });
       }
     } catch (err) {
-      // If we'd charged this job already (rare — most refundable
-      // failures happen on the FIRST call), refund.
       if (reusedJob) refundJob(installId, jobId);
       console.error(`[relay/transcribe] backend error: ${err?.message}`);
+      await recordCall({
+        install_id: installId,
+        tier,
+        pipeline: "transcribe",
+        backend: chosenBackend,
+        model: chosenBackend === "gemini"
+          ? cfg.relay_gemini_transcription_model
+          : cfg.relay_parakeet_model,
+        status: "error",
+        credit_charged: 0,
+        duration_ms: Date.now() - t0,
+        cost_usd: 0,
+        job_id: jobId,
+        error: (err?.message || String(err)).slice(0, 200),
+      });
       const e = await errorEnvelope({
         error: err?.message || "backend_error",
         installId,
@@ -139,7 +166,6 @@ export function transcribeRouter() {
       return res.status(e.statusHint).json(e.body);
     }
 
-    // Commit the credit on success (unless this was a job-id reuse).
     let creditCharged = 0;
     if (!reusedJob) {
       await commitCredit(installId, { backend: chosenBackend, tier });
@@ -147,6 +173,32 @@ export function transcribeRouter() {
       creditCharged = 1;
     }
 
+    // Success — write the audit row with cost details. Gemini's usage
+    // metadata gives us token counts; calcGeminiCost translates that
+    // into USD. Hardware-served calls have no token data and we
+    // report cost_usd: 0 (operator's hardware is fixed-cost).
+    const costDetails =
+      chosenBackend === "gemini" && result.usage
+        ? calcGeminiCost(result.model, result.usage)
+        : {
+            input_tokens: 0,
+            output_tokens: 0,
+            thinking_tokens: 0,
+            cost_usd: 0,
+          };
+    await recordCall({
+      install_id: installId,
+      tier,
+      pipeline: "transcribe",
+      backend: chosenBackend,
+      model: result?.model || null,
+      status: "success",
+      credit_charged: creditCharged,
+      duration_ms: Date.now() - t0,
+      job_id: jobId,
+      ...costDetails,
+    });
+
     const body = await envelope({ result, installId, tier, creditCharged });
     res.json(body);
   });