Video Transcript Contract

Scenario

Use before selecting short clips from a long recording when a transcript has word or segment timestamps from an approved transcription tool.

Input and output

Input is redacted JSON with ordered segments containing start, end, and text. Output contains normalized segments, duration checks, and transcript_hash.

Execution and recovery

Run python scripts/normalize.py examples/input.json. Missing text, negative times, overlaps, or reversed boundaries produce visible errors. Correct the transcript and rerun; no media file is opened.

Privacy boundary and acceptance

Use a redacted transcript and placeholder speaker labels only. The sample must produce ordered segments, a positive duration, and a stable hash.