Video Digest

Video review is dual-modal: the transcript/audio gives reasoning and claims, and actual frame inspection (what's drawn, highlighted, or pointed at) covers the visuals. OCR is secondary, only for exact numbers and labels; OCR alone does not complete visual coverage.

Tag important facts SPOKEN, SHOWN, or SPOKEN+SHOWN. If speech-to-text and the screen disagree, keep the misheard version in the audit notes and use the on-screen value in the digest. Never invent unreadable values; mark them unresolved.

Report coverage as three separate fields: review coverage (complete/partial/blocked), visual evidence available (yes/partial/none), and claim verifiability (verified/partly verified/speaker's word only). Don't mark a talking-head video incomplete just because it has no visuals.

Use when summarizing a YouTube video or a new upload from a watched channel, so the digest covers what's shown on screen as well as what's said.

Use for the first conversation with a new owner of the Video Digest bot, to set up which videos to watch and how.