Studying with Repeto

A Repeto session is a queue of eligible learning, review, and new cards from the selected study scope. Studying a parent deck includes its subdecks. Studying without a deck filter uses the collection-wide scope.

Before the first card

The Today and deck screens summarize due work. These counts describe the collection, while the final session also respects daily limits and the selected deck’s scheduling copy.

The web, iOS, and Android clients ask the API for up to 9,999 session cards, rather than stopping a large due queue at 200. A deck can still have a lower Maximum reviews per day or New cards per day setting.

The review loop

For every card:

1. Read the prompt and attempt active recall.
2. Reveal the answer.
3. Compare it with what you recalled.
4. Choose a rating.

The four ratings mean:

• Again — you could not recall the answer or recalled it incorrectly.
• Hard — you recalled it, but only with substantial effort or uncertainty.
• Good — you recalled the essential answer with reasonable effort.
• Easy — the answer was immediate and confident.

The interval shown with each button is a preview based on the current card state and deck configuration. After a rating, the card’s FSRS state and review history are updated and the queue advances.

New and learning cards

New cards have never been reviewed. Their first rating moves them into learning or directly into a scheduled state according to the configured learning steps. A later Again on a graduated review card can move it through relearning steps.

Short steps are time-based, so a card may reappear inside the same session. Repeto persists the learning-step position, due scheduler day, active queue, and new-card position so a reload or offline replay does not silently restart a multi-step card. Intraday learning stays ahead of interday learning, review, and new work.

Open Settings → Study to set the account-wide Learn ahead limit. The default is 20 minutes, the allowed range is 0 to 1,440 minutes, and 0 means wait for the full learning delay. Future intraday learning cards enter the queue only after ordinary due learning, review, and new work is exhausted. They do not consume the requested session-card limit.

Review cards and elapsed time

FSRS estimates memory stability and difficulty from the review history. Repeto also passes real elapsed time into the scheduler, so a card reviewed after a long break is not treated as though it was answered on its original due date.

Desired retention controls the tradeoff between workload and recall. Raising it generally creates more frequent reviews. Avoid changing FSRS weights unless you understand the full 21-value parameter set.

Ordering

Current deck options include:

• new cards in position order or random order;
• new cards mixed with, before, or after reviews;
• reviews by due order, due then random, random, shortest interval, or longest interval.

This is a useful subset, not yet the complete Anki ordering matrix. Interday learning has its own durable queue and priority. Manual burial and the three automatic sibling-bury rules are supported by the scheduler. Their switches are available in global presets and each deck’s private scheduling copy; an existing deck receives later preset changes only when you explicitly sync it. Hierarchical daily quotas and Limits start from top are enforced; separate gather/sort phases and the remaining ordinary-queue orders stay in the parity roadmap.

Increase today’s limits

Choose Custom study → Increase today’s limits to add extra new or review capacity to the ordinary deck until the next account rollover. This is additive: it raises the effective limit instead of replacing the deck’s stored limit, caps each category at 9,999, can be reset early, and does not create a filtered deck.

Custom study with a filtered deck

Choose Custom study on the deck list to build a temporary queue from the same search language used by the collection browser. A filtered deck can use one or two filters. Each filter has its own query, card limit from 1 to 9,999, live match count, and order: oldest due, random, smallest or largest interval, most lapses, oldest added, latest added, or Most likely forgotten first. The second filter runs after the first, so the same card is never gathered twice.

Choose Reschedule cards based on my answers for a real study session. Each answer uses the card’s home-deck settings and keeps the updated schedule. Review cards return home after the answer; learning and relearning cards remain in the filtered deck until their active home-deck steps finish. Choose Preview only when answers should not change scheduling or review statistics; Repeto restores the card’s exact scheduler state when it returns home.

You can create the definition without building, create and build it immediately, study the current queue, rebuild it from the saved filters, or empty it while keeping the definition. Deleting a filtered deck first returns every unstudied card to its home deck. Rebuild, empty, and delete show the restore consequence before they run.

Live match counts are estimates before build exclusions. Suspended, buried, already gathered, and duplicate cards are skipped, so the final queue can be smaller. Filtered definitions, memberships, home decks, and tombstones participate in normal cross-device sync and Repeto-native backup/restore. Most likely forgotten first uses current FSRS retrievability against the home deck’s retention target and falls back to overdue/interval data when memory state is incomplete.

Filtered decks deliberately use each card’s home-deck learning and relearning steps; there is no separate filtered-only step editor. Preview mode has configurable Again, Hard, and Good delays, with Anki-style defaults of 60 seconds, 600 seconds, and 0 seconds. A positive delay keeps the card in the preview queue for another attempt; Easy or any zero-delay answer returns it to its home deck. Review-ahead scheduling with an early-review algorithm, dynamic filter terms, and a guaranteed new random shuffle on every rebuild are not supported yet.

Evaluate FSRS settings

Open Statistics → FSRS tools to evaluate the current model with log loss and RMSE, or optimize the selected preset or all eligible presets from review history. Optimization requires enough complete history, returns a candidate only when neither metric becomes worse, and never applies it without explicit confirmation. Applying a global preset does not silently rewrite a deck’s private copy; sync that deck from the preset when you want the new weights there.

The same workspace provides a read-only due-date preview for a proposed retention target and a deterministic workload simulation with a retention sweep. These are decision aids, not promises. The preview never changes data by itself; after reviewing it, a separate confirmation can apply that exact due-date plan. Commit changes due timestamps and scheduler days while preserving review history, counters, intervals, and FSRS memory. Stale, invalid, and replayed previews are rejected. There is not yet an Undo action for this commit. Incomplete history blocks optimization instead of manufacturing a result.

Ending and resuming

Choose End session or press Esc to leave. Pending review writes are flushed before navigation. When you return, already recorded reviews stay complete and the API constructs a fresh queue from current due state.

If a network interruption happens during review, Repeto persists the rating before advancing: web uses an account-scoped IndexedDB outbox, while iOS and Android use atomic on-device files. All three replay accepted ratings in FIFO order with stable idempotency keys. The server serializes submissions and revalidates card state, queue and learning step, counters, interval and FSRS result, due class, deck configuration, staleness, and idempotency before accepting one. A permanently conflicting item is isolated as blocked instead of corrupting the schedule or stopping later valid ratings; retryable network, server, authentication, and rate-limit failures stay pending. Do not keep multiple reviewer tabs open for the same collection; a stale card may be rejected after another tab changes it.

For editing, suspension, undo, and shortcuts, continue to Reviewer actions.