Repository navigation
fix(video-bridge): validate structural segment sampling - #11381
Conversation
|
CI classification snapshot — 2026-08-24
This note does not claim the full CI is green. The PR remains OPEN/DRAFT; no rerun, waiver, ready-for-review, auto-merge, or merge was performed. |
|
Merged via consolidated batch validation (worktree |
…per-inference#11344/cheaper-inference#11381/cheaper-inference#11362/cheaper-inference#11382/cheaper-inference#11383 growth These entries were already validated in an earlier merge-batch worktree but never reached origin (worktree discarded before pushing). Re-adding them here since cheaper-inference#11355's test/route.ts growth (1215->1237) is now live on origin/release/v3.8.50 and fails the frozen cap otherwise.
Summary
segment_awareVideo Bridge sampler use one bounded, local-only FFmpeg structural pass for scene, freeze, blur, exposure, and SI/TI evidenceTDD and real-media evidence
Deliberate HOLDs
This PR does not claim semantic caption-model quality or a real monetary gain. No authorized caption endpoint, model credentials, frozen judge, or provider cost receipt was configured. CPU/RSS measurement also remains a HOLD when
/usr/bin/timeis unavailable. The harness reports those states explicitly instead of fabricating values.Validation
npm run typecheck:core— PASStypecheck:noimplicit:core— BASE-RED only in unchangedopen-sse/translator/response/openai-responses/pureHelpers.ts; no FU-07 diagnosticBase
Validated against exact release tip
dafb4ae808305c9287c751cf1ce16fe2234ba073.The PR remains draft for the broader release campaign; no merge, tag, release, or deployment is requested.