sort-enrich-photos-interactive-clusters

Sorting a camera roll by day clusters

A classification problem, not a file-moving one. Moving files is trivial; deciding where each photo belongs is the hard part, and the person you're working with is the only one who knows what a given Tuesday was.

Do all the mechanical work and bring them in only where their knowledge is required. Their attention is the scarce resource — spend it on the biggest groups first.

Files move and metadata is rewritten in place, so: show the plan, get approval, execute.

For photos an activity log can match automatically, use sort-enrich-photos-strava.

Conventions come from the library

No settings file. Tunable knobs and per-run decisions live in scripts/sort_photos.py — read it before filling it in.

Everything else comes from looking at the library:

• Copy the folder layout already in use. Don't assume a shape — it might be <year>/<date>_<place>/, Events/<name>/, or a flat pile. The script takes a literal relative path of any depth, so any layout works. Sample a few places; conventions drift over years.
• Read the tag facets before inventing a tag, to avoid Holidays next to an established Holiday.

Nothing to copy — empty library, first sort — means proposing two or three concrete shapes with examples and confirming once. It's the hardest decision to undo: a thousand files under a disliked scheme is a thousand files to move again.

Near-duplicate tags accumulate: Oliv beside Olivier, an accented and unaccented spelling. The tell is a pair differing by case, accent or prefix with lopsided counts — the rare one is usually the mistake. Ask which is canonical, then retag the smaller set rather than remembering an alias. A correction in a config file must be re-read forever; one applied to the photos is simply true.

1. Index and size up

Index the staging partition, get a count and date range, and also list the folder directly — the index and filesystem disagree in informative ways.