# Hybrid system overview

One evidence token moving through signals, fallback, review, and public proof.

## Transcript

- **00:00:00.000–00:00:05.000** What makes an AI quality report trustworthy when one signal fails? Follow one evidence token.
- **00:00:05.000–00:00:10.000** It enters inspectable heuristics, temporal context, and semantic continuity checks.
- **00:00:10.000–00:00:13.000** Each route keeps its source, score, and limit.
- **00:00:13.000–00:00:17.000** If the model service fails, the token does not disappear.
- **00:00:17.000–00:00:24.000** The report marks the missing signal, keeps the safe baseline, and routes uncertainty to a human reviewer.
- **00:00:24.000–00:00:27.000** The reviewer sees evidence, not just an alarm.
- **00:00:27.000–00:00:32.000** Public proof publishes architecture, benchmark, and limitations while private clips stay protected.
- **00:00:32.000–00:00:35.000** Trustworthy systems preserve the evidence token, especially through failure.

## Accessibility and claim boundary

The poster is the complete static equivalent. Captions and transcript derive from the same reviewed Caption JSON source. Synthetic examples teach relationships; they do not certify a production detector.