During production, amber flags appear on the timeline as issues arise.
Before export, the Film Audit scores eleven categories: character appearance, character state, wardrobe, props, locations, time and weather, story contradictions, voice consistency, lip-sync, lighting and technical quality.
Each finding is actionable, and Fix All (or selective repair) regenerates exactly the affected nodes via the graph.
An honest boundary: the structural checks are real — they read the Bibles, the state machine and the shot metadata, and they find genuine contradictions like "Scene 27: John's arm is broken; Scene 29 shows him climbing two-handed". The perceptual checks (does this frame actually look like that character) need a vision model. Where one is not configured, those categories are reported as not run rather than passed. A green audit that never looked is worse than no audit.