Source provenance
Every completed record shows the source host, acquisition method, fetch time, segment counts, timed coverage, and raw source link.
Evidence-first platform
UVAI treats captions as evidence, keeps provenance visible, persists every workflow run, and puts a deliberate review boundary in front of external execution.

What is implemented
Every completed record shows the source host, acquisition method, fetch time, segment counts, timed coverage, and raw source link.
The transcript is built from backend-acquired caption segments with real timestamps. Unlabeled captions remain labeled as captions.
Export and success states stay blocked unless transcript evidence passes structural and provenance checks.
Each analysis receives a run ID, persists its result, and can resume polling after a browser refresh.
The model receives the verified transcript and returns summary, topics, and proposed actions without reconstructing missing source content.
Action preparation executes nothing. A second confirmation is required for the exact reviewed plan, and external writes are labeled.
The YouTube player, seek timeline, and caption timestamps stay connected so users can inspect the source while reviewing conclusions.
Exports are enabled only after the quality gate passes; failed acquisition never turns into a synthetic package.
AI endpoints use a tighter budget and fail closed in production when distributed rate limiting is unavailable.
These are the boundaries the current build enforces, not future roadmap promises.
Paste a public YouTube URL. UVAI will either return a verified analysis or show the exact stage that blocked it.
Open dashboard