Cards

Flashcards written from a board's own sections, scheduled by FSRS, drilled from three queues.

Every board has a deck. The cards are written from the sections on that board — the notes you wrote, pasted or imported — and not from the board's title, so they ask about your material rather than about the subject in general. Add a section and the next run asks about it; rename the board and the deck is unchanged.

A drill looks like this:

Making cards#

The maker panel sits under the deck on a board's Cards tab. It asks what to cover — the whole board, or sections you pick — and then writes questions from those notes.

What comes back is a proposal, not a deck: a tick-list you approve card by card. Nothing is saved until you add it, and an unapproved proposal survives navigating away and coming back. The questions already in the deck go out with the request, so a second run after folding in new notes writes cards for the new material instead of repeating what you have.

Image sections become cards with no model call at all: the picture is the question and the section's heading is the answer, which is the spot-diagnosis card you wanted anyway. Each picture gets one.

Making cards is board-scoped. The two pooled queues below have no single board to read, so the panel is not offered there.

Scheduling#

Cards are scheduled by FSRS, with the default weights. Those weights are trained on a very large body of real reviews, so the schedule is sensible from your first card and needs no history of your own.

You grade on four buttons — Again, Hard, Good, Easy, or keys 1 to 4 — and each one prints the interval it books. The printed interval is computed with the schedule's fuzz turned off, so a button that says 7d cannot then book 6.

Grading honestly is the whole mechanism, and it is the only thing the schedule has to go on. The interval is an estimate of when you would forget the card; pressing a grade tells it whether that estimate was right. Mark a card easier than it was and you buy nothing — the card goes further away and comes back after you have lost it. Mark one harder than it was and you pay for a few repetitions you did not need. Nothing is scored, ranked or reported, so there is no reason to flatter it.

The details worth knowing:

• Again is the only grade that records a lapse, and the only one that puts the card back into the current session — at the back of the queue, not immediately.
• Hard means recalled with effort. It is not a failure and does not count one.
• Intervals are capped at a year. A card next due in three years is not revision.
• New cards are capped at 20 per session, so making a hundred at once does not turn the next sitting into an avalanche. The rest stay due and come up next time.
• Cards made together are due together, so same-day ties are shuffled. Otherwise you learn the order rather than the content.
• A card failed five times is flagged as a leech. That points back at the board rather than suspending the card: a card you keep failing is usually a card written badly, or a note that needs rewriting.

There is no per-card review log. Per-user weight tuning would need one; default weights do not.

Queues#

The same drill runs at three scopes:

• This board — the board's own Cards tab, the deck above.
• This subject — every board the subject's index claims, pooled.
• Everything due — every deck you have.

The decks themselves are per board. The two wider queues pool them rather than being decks of their own: they take what is due from each, oldest first, and a grade lands back in that card's own board's deck whichever queue you graded it from. Nothing leaks between scopes — a subject's queue never shows a card from a board outside it.

Each queue keeps its own session. Leaving a subject drill part-way and opening a board's own deck does not continue the first one; going back picks it up where you left it, including on another device.

Cards from what you got wrong#

A marked paper and a viva attempt both produce misses: questions that lost marks, criteria an examiner marked you down on. Those are the most useful cards there are, because they are gaps you have demonstrated rather than gaps predicted from your notes — so they are written up as cards automatically and wait in a tray on the Cards tab, one tap each. A card you keep failing is treated the same way, except that the request is to break it into the simpler questions it was hiding rather than to restate it.

Nothing is added behind your back, and nothing is written twice. Two exact filters run on every proposed card, wherever it came from:

• Origin. A card remembers the miss it came from. Re-marking the same paper, or failing the same station on the same criterion next month, is the same miss and produces no second card.
• Wording. A proposed question whose text matches one already in the deck — ignoring case, punctuation and spacing — is dropped, including a repeat inside the same batch. Image cards are matched on the picture instead of the stem, because every one of them asks the same question.

Both filters are exact, so two different phrasings of the same fact can still get through. That is deliberate. Catching those would mean judging meaning, and a filter that judges meaning also throws away cards that were genuinely different — a quiet loss you would never see. The generator is told not to ask the same thing twice; the filter stays literal, so it can only ever drop a card that really is a duplicate.