YouTube Edit List

Input: a timestamped transcript or rough-cut notes, plus the retention drops and cutting pattern if the video is a re-cut. Data source: if ~/.youtube-skills/metrics/<VIDEO_ID>.json exists (see youtube-connect), take the drops from its retention.cliffs. Output: an edit decision list an editor can execute without watching the whole file, chapters, and the end frame.

Steps

1. Mark the first cut. Speech or the payoff frame starts by 0:02.5. Everything before it is removed. Name the first on-screen change and its timestamp.
2. Walk the transcript in order. At every span of 8 seconds with no change of state (cut, zoom, graphic, b-roll, text, music), write one. At every span under 3 seconds with more than one change, remove one. Change of state means the viewer sees something new, not a jump cut on the same frame.
3. Apply the cutting pattern from youtube-retention if given. Otherwise pick it from the video type: tutorial is hybrid tempo, explainer is progressive rhythm, commentary is contrast.
4. If retention drops were given, treat each drop timestamp as a hard cut point: remove or move the span the retention skill named. The fix list from that skill becomes rows in this list.
5. Write the chapters. One per segment payoff, worded as a micro-promise with a number, a comparison, or a finding. Never "Intro" or "Outro".
6. Set the caption chunks: 1-5 words per screen, break on the spoken pause, keep the filler words that carry rhythm.
7. Mark the end frame. The video ends on the payoff or the pointer to the next video. No outro, no logo hold.

Rules