videos.map.json — date → YouTube id, recorded once per upload. Drives the video thumbnail embed and the deep-linked chapters.
LLM extraction summaries are not committed here — they're regenerable staging in ../.tmp/summaries/. The index is built from the wiki pages, not the summaries.
indices/videos.json — video / meet-up chapters + per-operator backlinks.
indices/skillquest.json — quest entries + references (TBD).
indices/mentions.json — typed reference (op:…, ui:…) → the video moments it was discussed; loaded by the editor's help/tooltip window.

Generated by ../scripts/update_help_index.py plus the extraction skill.

Related, kept separate by lifecycle:

../.tmp/ — regenerable intermediates (transcripts, audio, draft review output). Git-ignored.
../.src/ — hand-authored source (video scripts, release-note drafts). Committed.