Keysat
14 lines
1.5 KiB
TypeScript
14 lines
1.5 KiB
TypeScript
import { VersionInfo } from '@start9labs/start-sdk'
|
|
|
|
export const v_0_2_68 = VersionInfo.of({
|
|
version: '0.2.68:0',
|
|
releaseNotes: {
|
|
en_US:
|
|
'Two fixes. (1) Per-window analyze now retries ONCE on invalid-JSON responses (or on a thrown exception from the backend). The model occasionally returns a JSON-shaped string that fails safeParseSections — extra prose preamble, truncated closing brace, code fence the parser misses — and pre-0.2.68 those windows were dropped on the first failure. With a 10-window job, two such failures lost ~36-48 minutes of analyzed coverage. The retry runs against the same prompt; the upstream LLMs treat each call independently and almost always succeed on attempt 2. Audit row records the per-attempt error message and tags the final error with "(after 2 attempts)" so the operator can tell whether it\'s the model fundamentally struggling with that window or a one-shot blip. (2) Section-count targets preview now updates when you drag the AN window body slider, not just when you type in the number input next to it. Earlier the listener only matched `data-setting-key` — but the slider in a slider/number-input pair uses `data-setting-slave`, so drag-events were ignored. Symptom: changing AN window body from 18 → 24 left the "Sections / window" preview column (and the "current AN window body of N min" description line) stuck at 18. Fixed by matching both attributes in the input-event delegator.',
|
|
},
|
|
migrations: {
|
|
up: async ({ effects }) => {},
|
|
down: async ({ effects }) => {},
|
|
},
|
|
})
|