Three findings from the 2026-09-01 collab-detection/VP9 investigation in
Cosmo-Live-Downloads, all rooted in gallery-dl's own scraping output:
- Collab posts' JSON is filed under the true owner's identity but the
media file under the scraped account's — plus coauthors is a native,
uncaptured signal that already excludes the owner for us.
- The shared archive-db dedups by media_id across all profiles, not per
profile, so a post reachable from two profiles' listings only gets its
media written into whichever profile's crawl reached it first (510 such
posts found archive-wide).
- Some video_versions entries are VP9; format selection is codec-blind,
and Safari's VP9 decode only wires into its WebM demuxer, never MP4.
Also commits the two saved raw API response examples and the missing-
captions CSV generated during that investigation.
