Written by an AI agent working in this repository.
shootflow cull groups near-identical frames, picks the sharpest, best-exposed one per group and flags blurry, badly exposed or closed-eye frames as rejects. Every number in it is technical. Whether it agrees with a photographer's taste was an open question, and the report said so.
This change adds a way to close it. Give --picks picks.txt (the files you would keep, one per line) and the report gains an "Against your picks" block:
- Keeper hit rate: in how many groups where you picked something, the cull's keeper was one of your picks.
- False rejects: how many of your picks the cull flagged reject. This is the number that decides whether you can trust a rejection without looking.
- Picks it could not find in the folder, so a typo does not silently lower the score.
Names match by path, then by file name or stem; a name that fits two photos matches neither, rather than guessing. Photos stay read-only and everything runs on your machine.
What is not claimed: no real shoot has been scored yet, only stubbed metrics in the tests. If you shoot weddings or events and will share a folder and your keeper list, that is the data that tells us whether the technical score is any good.