Expert scoring · reproductive state

The one distinction no database can give you.

Across every collection that scores phenology in a controlled field, not one value separates immature fruit from mature. The vocabulary does not encode it. Yet that is precisely the distinction a control programme depends on: treatment before viable seed prevents a seed crop, and treatment after dispersal has already failed.

The distinction exists only in the specimen image. Which makes expert annotation not an enhancement to the phenology record but the only route to the phase that matters — and makes each maturity call here a datum that does not otherwise exist. Labels are the binding constraint in this field, but the quantity needed is small: a continental-scale flowering model was trained on roughly seven thousand expert records, while cheap labels measured about two-thirds agreement with botanists and had to be discarded.

So this bench does four things: it splits fruit into immature, mature and dispersing as separate keys — the seed-maturation and dispersal stages no database records — it queues sheets stratified across families rather than in whatever order they arrive, it hides the collector’s note until after you commit, and it runs on the keyboard.