One study engine for the cards you already make — six ways to study a set, with spaced repetition that decides what to review next, built to embed anywhere in the platform.

What Homeroom Study Sets is

At the centre is a small, plain idea: a study set is a list of cards you want to learn. A card can be a term and its definition, and it can carry an optional picture or a sound. A card can also be a bank item — it references the same canonical assessment item the quiz engine already scores, so a study card and a quiz question can be the same underlying thing rather than two copies drifting apart.

Around that set, the engine does two jobs. First, it gives you six ways to study: read through and learn, flip flashcards, study by ear, match terms to definitions, take a test, or write the answer out. Second, it runs spaced repetition: for each card, for each student, it keeps a tiny schedule that decides when that card is worth seeing again, so review time lands on the cards that need it.

It is built to be embedded. The study surface is a small module another part of the platform can mount inside its own page, so a language class, a music theory drill, or a lyric-learning screen can all study with the same engine instead of each reinventing one. And it is private by default: a set starts private, a student’s practice schedule is treated as private minors data, and a shared set only shows its author’s name when a multi-party consent check passes.

Honestly, where the build is: the engine underneath this page is written and tested, but this product surface is early access and is being opened up gradually. Where something is a plan rather than an open door, this page says so. The honest next step is to ask about early access.

Spaced repetition, worked out plainly

Behind the study modes is a spaced repetition scheduler that decides, per card and per student, when a card is worth seeing again. It follows the published SM-2 method (Wozniak, 1990). Each card keeps three small numbers: how many times in a row you have gotten it right, the gap in days until it is next due, and an ease factor that starts at 2.5.

The schedule reads simply. The first time you get a card right, it comes back in 1 day; the next time, in 6 days; after that, each gap is the last gap multiplied by the card’s ease, so a card you keep getting right returns less and less often. Miss a card — any answer graded below the passing mark — and it drops back to a 1-day relearn step, but it keeps the ease it has earned; a lapse does not wipe your progress on that card. After every review the ease nudges up a little when the answer was easy and down a little when it was hard, and it never falls below a floor of 1.3.

It is deterministic. All of that interval and ease math is pure arithmetic: it reads no clock and no random number. The clock is used for one thing only — to stamp the date a card is next due. The same card state and the same answer always produce the same schedule, every time, which is what makes the behaviour testable rather than mysterious.

Two other views sit on top of the same state: a Leitner box number, for a simpler at-a-glance sense of how well a card is known, and a mastery read that marks when a card has been held long enough to count as learned. A due queue pulls the cards that are ready for review right now to the front.

The quality score that feeds the scheduler comes from the same shared grader the graded modes use — not a private study-engine score. And it stays honest at the edges: when a card cannot be objectively graded, the engine hands the scheduler no quality at all instead of a made-up one, so a fabricated grade can never drive a real schedule.

Private by default, and honest about the data

A study set does not have a “public to the whole internet” setting, by design. A set is private, shared to a school, or unlisted — and there is no fourth, public option to choose. New sets start private. If a visibility value ever arrives that the engine does not recognise, it falls back to private rather than opening a set up, and the database itself carries a check that refuses to store a set marked public. Private-by-default is enforced in the code and again at the row.

A student’s practice schedule — the per-card ease, interval, and due dates the spaced repetition keeps — is treated as private minors data. It lives in its own table behind the school’s row-level security, following the same hardened subject-privacy pattern the platform uses for sensitive student rows: a student sees only their own schedule, staff see it through the school’s student-data rules, and a support session scoped to a school reads none of it. A student never matches another student’s row.

That schedule is also walled off from sharing. When a set is shared, the shared view carries the cards and nothing else — no review state, no schedule, no keys. One student’s practice history cannot ride along inside a set someone else opens.

Names on a shared set are consent-gated. Before a shared set shows who made it, the author’s name is run through the platform’s multi-party consent check — the chain that asks the school, the district, and the parent. Only when that check passes does the name appear; otherwise the set shows an anonymous label. And to be plain about it: this is not a “we hold no data” claim. A practice schedule is real data. The honest claim is that it is private minors data, kept behind the school’s walls, and never carried into anything shared.