Every answer you give — inside a lesson, on a card, in a mock paper — goes through one function. There is one record of what you know, not one per feature, which is why the app cannot congratulate you on one screen and warn you on another.

The date something comes back is computed by FSRS — the same open scheduler whose benchmark is linked below — and it runs on the server, so your phone and your laptop can never disagree about what is due.

Every card carries a “this is wrong” report. It posts to a real route, lands in a ledger, and takes the card out of circulation for the rest of the session.