Bird/Media Automation: inbox report + the record-to-Instagram roadmap

This already does most of the hard part of "find the good bird moment automatically": reads your Sony a7S III sessions, runs a largest-blob motion detector tuned against your real North Wing footage, and grades every clip 0–5 (V4/V5 = the trustworthy "social banger" tier, up to 60s). It writes graded Premiere sequences and public contact sheets at slowmotionbirds.com. Detection quality, not guesswork: it already caught and fixed a background-model deadlock and a false-positive band that generic "motion detection" would have missed.

Not yet built inside it: species ID (Phase 4, opt-in, gated on your own eBird life list), and any publishing step. Its own spec already says the publishing phase is draft-only with a required human approval click — that's relevant below.

There's no reliable way for anything automated to know a clip is "the weekend at Granny's." No metadata says that, and a vision model guessing from pixels will get names wrong in ways that are more annoying to fix than typing a real one. That part stays yours.

What automation can do, and what the table above already proves works: land everything in dated, device-grouped batches with a temp label and a thumbnail contact sheet, then ping you when a batch is ready. You open a link, look at 3–5 thumbnails, type the real name once. That's "the gig" — same pattern the bird pipeline already uses for its own contact sheets, just pointed at the Osmo folders too.