Files

14 lines
2.3 KiB
TypeScript
Raw Permalink Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
import { VersionInfo } from '@start9labs/start-sdk'
export const v_0_2_67 = VersionInfo.of({
version: '0.2.67:0',
releaseNotes: {
en_US:
'Section-count target tuning methodology rebuilt around VIDEO duration instead of window duration. Old model (v0.2.62-v0.2.66): operator set three strings — short / medium / long — keyed by individual analyze window length. The relay picked one per window based on that window\'s transcript duration. Problem: a 30-min single-window podcast and a 3-hour 6-window podcast bucketed identically as "medium" but need totally different segmentation density. The model was producing too-many or too-few sections depending on which the operator hadn\'t calibrated for. New model: operator sets the target TOTAL sections per video for each of 7 duration buckets (under 30 / 30-60 / 60-90 / 90-120 / 120-150 / 150-180 / over 180 min). The relay reads the actual audio duration at request time, picks the matching bucket\'s target, then divides by the effective number of analyze windows (clamped >= 1 for single-shot) to get the average sections-per-window. Splices a formatted label ("around N sections" / "NM sections" / "1 section") into {{targetSections}} in the analyze prompt. Settings tab redesigned: a 7-row table with the duration label, an integer input for target-total, a live "Sections per window" preview column showing the computed result + the prompt label string, and the default value. The preview recomputes as the operator tweaks either the AN window body slider above (event-delegated input listener) or any target-total input directly (per-row oninput handler). Backend changes: chunked-analyze.js exports computePerWindowTarget(); buildWindowPrompt() takes targetTotalsByBucket + totalAudioSec parameters; runChunkedAnalysis threads them through. Both /relay/summarize-url and the admin test-run path supply the new config keys. The deprecated string settings (relay_analyze_target_short / medium / long) are no longer used and have been removed from the schema; old values saved in relay-config.json become orphans, harmless. Defaults: 6/8/9/10/11/12/12 — calibrated so a 90-min podcast at the default 18-min AN window body produces ~1-2 sections per window (about 9 total), matching what felt right in operator testing.',
},
migrations: {
up: async ({ effects }) => {},
down: async ({ effects }) => {},
},
})