Spaced repetition for the drills — and why Leitner, not SM-2
The pattern and collocation drills reshuffled the whole deck every time, wasting reps on chunks you'd already nailed. Added per-card spaced repetition on the account: the daily drill now surfaces what's due plus a trickle of new cards. Picked a Leitner box system over SM-2 because the drill's two outcomes map to it exactly.
With the drill streak now living on the account, the next gap was obvious: "Start daily drill" reshuffled all 246 patterns / 202 collocation items every session. So you'd re-drill depend on (long since automatic) at the same rate as result from (which you keep fumbling). The owner's actual ask all along was efficient daily practice — and that's exactly what spaced repetition is for.
Why Leitner, not SM-2
The app already runs SM-2 on the clip Review side, and it would have been tempting to reuse it. But SM-2 wants a 0–5 quality grade per review — it fits the clip review's four buttons (Again / Hard / Good / Easy). The drill has only two outcomes: Got it or Again. Forcing those into SM-2's scale would throw away most of its precision and add complexity for no gain.
A Leitner box system fits the binary outcome exactly: a correct recall promotes the card one box (it comes back later), a miss drops it to box 1 (back tomorrow). Box → interval is a fixed ladder: 1 → 2 → 4 → 7 → 14 → 30 days. Right tool for the shape of the data, even though a "fancier" one was sitting right there.
The shape
Backend, extending the practice slice from yesterday: a practice_card row per (user, card), where a card is one drill item keyed by a stable string — pat:on-depend-on#0, col:for-responsible-for#1. The Leitner rules live as domain behavior on the entity. Two endpoints: read all card states, and grade one. Grading a card also records a streak rep, so spaced-repetition state and the daily streak come back in a single round-trip.
Frontend: the day's session is built client-side from the static deck + the user's card states — everything due today (shuffled) plus up to 12 new cards. The two drill buttons are the grades: "Got it" promotes, "Again" demotes; both count as practicing. Each page shows a {due} due · {new} new hint, and a caught-up toast when there's nothing due.
Verified three ways: ten Testcontainers integration tests (promote-on-correct, reset-on-miss, per-user isolation, validation), a live HTTP round-trip (a card walking box 1 → 2 → back to 1 on a miss, due dates moving 06-01 → 06-03), and a direct practice_card query confirming the rows. The drills are now a real daily practice loop, not a shuffle.