Open a flashcard app after a week away and the first thing it shows you is a number. 312 due. The number is accurate, and it is the wrong first thing to show. It answers a question you did not ask (how far behind am I?) and hides the one you did (how long will this take?).
immit's review screen asks a different question: how much time do you have today? Everything below follows from that question and from three rules we hold ourselves to. The promise is time, not cards. The scheduler is never lied to. Complexity can be hidden, but the deal cannot.
The review badge itself is the first sign of this. It shows the estimated minutes for what is due, capped at 15, and never a count.
When you open review, immit asks how much time you have today and offers up to four options: 5 min, 15 min, 30 min, and Everything. The timed options show how many cards fit at your pace ("15 min ≈ 90 cards"). Everything shows the exact number of cards due and roughly how long they would take, so even "all of it" is priced in minutes rather than handed to you as a count.
A timed option only appears when it is shorter than the estimate for the whole pile. With a short pile you only see Everything. As the pile grows, the shorter options appear one by one. At the starting pace of 10 seconds per card, before immit has learned yours, the ladder looks like this:
Cards due
Estimate
What immit offers
10
2 min
Everything
85
15 min
5 min, Everything
150
25 min
5 min, 15 min, Everything
210
35 min
5 min, 15 min, 30 min, Everything
The card counts move with your pace: someone who flips a card every six seconds sees the 30-minute option later than someone who types every answer.
The default is 15 minutes whenever it is offered. We picked it because it is the session that survives contact with a real week: long enough to move the pile, short enough to fit before work or on a train, and small enough that saying yes to it is easy. If you have more time, the longer options are one tap away. If you have less, 5 minutes is a real session, not a consolation prize. And if 15 is simply not your number, you can change which option is preselected in Settings, under Review (Default Session Length). The choice follows your account, so the web, the desktop app and the extension agree.
Why time and not a card count? Because time is the unit your day is already planned in. Fifteen minutes with the morning coffee, the train ride to work, a quarter of an hour before bed: those slots exist whether or not you have flashcards, and a session sized to fit them gets done. "Ninety cards" has no natural place in a day. You cannot tell whether it fits until you have already started.
The same design is what keeps review going after a lapse. Skip a few days and 300 cards are waiting. Pushing through all of them in one sitting is draining, and that drain is usually what turns a missed week into a quit. Fifteen focused minutes today, fifteen tomorrow, and the pile comes down on its own while you get back into rhythm. Catching up gradually is easier on you, and the sections below explain why the scheduler is fine with it too.
If you would rather not leave anything for tomorrow, that is what Everything is for: every due card, in the scheduler's order, nothing left out.
A timed session ends when the timer does. At 0:00 you finish the card on screen, that answer counts, and no new card is dealt. If you clear the pile before the clock runs out, the session ends there instead. The timer is a ceiling, never a floor.
The "≈ 90 cards" next to the 15-minute option is an estimate built from your pace in the review mode you are using: typing takes about twice as long as flipping, so each mode has its own number. As you review, immit measures your seconds per card and refines the estimate from your last few hundred answers. Until it has enough of your own history it uses a population default. It gets closer over time. The timer was never approximate.
Pausing stops the clock, and switching to another window or tab pauses it for you. Paused time is not counted anywhere: not against your budget, and not in the per-card timing at the end. The timer measures attention, not wall-clock time, because attention is what you budgeted.
When the session ends you get four numbers: cards answered, accuracy, seconds per card, and best streak, meaning consecutive correct answers inside this session. They are a receipt for the session you just did. The longer view, your day streak, days learnt, words saved, and a calendar of the days you reviewed, lives on your dashboard on immit.co if you have an account. We keep the two apart on purpose. The review screen is about today's promise; the dashboard is where you go when you want to look back.
If the timer stopped you with cards left over, the end screen says "That's your 15 minutes" and offers +5 more min. The extension continues the same queue, plus anything that came due while you were reviewing, and the stats keep counting, so 15 minutes plus one extension is one 20-minute session. Everything has no timer and no extension. It shows a plain counter ("1 / 32 cards") and ends when the pile does.
A 15-minute session on a 150-card pile means about 60 cards wait until tomorrow. Which cards you see is the decision that makes or breaks a short session, and it is made by the scheduler.
immit orders due cards by retrievability: the probability, right now, that you could still recall each one. Highest first. After the due cards come new cards, the words you saved but have never reviewed. A card you get wrong is not lost. It does not come around again in the same session, but the scheduler brings it back soon, on its own timing: a brand-new word you miss is due again in about five hours.
Highest-first sounds backwards, and it took us a while to come around to it. What convinced us was looking at what a single review actually does. A review right before you would have forgotten something is what pushes the memory further out. A card you can still recall at 85% is worth rescuing today. A card that has sunk to 30% is, for practical purposes, already forgotten, and one more day changes little. When time is short, the session goes where it does the most good. That is why the order is left to the scheduler: a short session should be spent on the cards that need it today, and the scheduler is the only part of immit that knows which ones those are.
The cards you did not reach are not rescheduled, softened, or dropped. They keep their original due dates, and the scheduler sees exactly what happened: nothing. This is the second rule at work. Time budgeting only decides which due cards you review today. It never touches intervals, due dates, or review history, because the moment an app starts quietly editing the schedule to make the pile look smaller, the schedule stops meaning anything. Tomorrow's picker asks the same question with a slightly different pile.
Because nothing is rescheduled, a backlog is just cards waiting, and the scheduler handles late reviews correctly. Get a card right three weeks after it was due and that is strong evidence the memory is solid, so the next interval grows more than it would have on time. Get it wrong and the card comes back soon, but not from zero. A word you once held for four months is not the same as a word you saw yesterday, and the scheduler knows the difference.
This is why the pile is not a debt. A backlog is measured in cards, but what you actually owe is minutes, and the minutes are usually fewer than the card count suggests, because the high-retrievability cards at the top of the queue are the fast ones. Pick 15 minutes today. Pick it again tomorrow. Everything is always there if you want it.
Saving a word in immit is one click with no dialog, and we intend to keep it that way. That choice has a cost: if every saved word went straight into review, the pile would fill with words you looked up once and will never need again. So a saved word has three places it can be.
Every new save starts in Learning and enters the review pool. From your Wordbook, or from the Saved button on the dictionary entry itself, you can move a word to Save for later, which keeps it in the book but out of review until you bring it back, or to Known, which retires it. During a session, Mark as Known does the same thing in one tap without leaving the card. A status move never touches the card's memory state, so a word you bring back to Learning resumes from its real history rather than starting over.
The time budget decides how much you review. These three statuses decide what is eligible. Together they keep the pile honest: the cards waiting for you are the words you actually meant to learn, which is what makes a 15-minute session worth taking in the first place.
immit runs FSRS, the open-source scheduler that Anki added in late 2023, in its current version, FSRS-6, through the same library Anki uses. FSRS keeps two numbers for every card: how hard the card is, and how stable the memory is. From those and the time since your last review it computes a third at any moment, how retrievable the card is right now. It schedules each card to come back when retrievability is about to cross a target, and that target is called desired retention.
In immit, desired retention is 90%. Every card comes back around the point where you have a 90% chance of still recalling it. There is no setting for it, and nothing in the app changes it behind your back.
We considered exposing the slider and decided against it, and this is the third rule at work. A slider asks you to make a decision without the information to make it well, to revisit it whenever reviews feel heavy, and to absorb a reshuffled pile every time you move it. The time budget already gives you a lighter day without changing the schedule underneath. So the retention target is a constant and the time budget is your lever.
Anki is the reference point for anyone who has done serious spaced repetition, and its settings panel is long for good reasons: desired retention, new cards per day, maximum reviews per day, learning steps, four answer buttons, an Optimize button. immit uses the same scheduler and removes most of the panel. Three choices matter most.
Two answers, not four. In Flip mode you turn the card and answer ✓ or ✕. In Type mode you type the answer and immit checks it, with some give: a near-miss in spelling counts, with a nudge to check it, and typing the word itself instead of its meaning gets a "try again" rather than a grade. The scheduler only needs to know whether you recalled the card or not. Extra buttons add a little information in theory. In practice, deciding whether a slow recall was Hard or Good is a judgment made hundreds of times a week, and a consistent lean in either direction skews the schedule. Yes or no removes the judgment, and with it a whole category of mis-graded cards.
One question instead of a settings page. Daily limits on new cards and reviews exist to keep a session bounded. The clock does that job directly, so the limits have nothing left to do.
No Optimize button, because it runs itself. FSRS has a set of weights that describe how your memory in particular behaves, and Anki asks you to press a button to refit them to your history. immit does that for you: once your account has enough reviews, the server refits your weights from your review log and keeps refitting as the log grows, and every device you use grades with the same weights. Without an account you review on the population defaults, which is what a new account starts on anyway.
Fewer numbers on screen. No retention rate, no forecast, no parameters to tune. Every number you see either helps you decide how long to review or tells you what you just did. Everything else is the scheduler's business, and keeping it there is what lets the review screen stay a single question.
Reviews only work if you come back, and in the weeks when reading falls off, immit will nudge you by email. Sparingly: only when you have an account, you have not reviewed on any device for three days, and at least one card is waiting. The note arrives at 9 in the morning in your time zone, never more than once in three days, and at most twice a week.
If you use immit without an account, you will not get them; signing in is what makes them possible. Turn them off any time in Settings, under Reminder, or with the unsubscribe link in any of the emails.
Nothing. They keep their due dates and are back near the top of tomorrow's pile if they are still the most recoverable ones. No penalty, no rescheduling.
Can I change the retention target?
No. It is fixed at 90%, and the app never changes it on its own. Use the time budget to have a lighter day instead.
I picked 15 minutes and there were only 12 cards. Now what?
That cannot happen. A timed option only appears when it is shorter than the estimate for the whole pile. With 12 cards due you would see Everything only. What can happen is that you review faster than the estimate and clear the pile before the clock runs out. Then the session simply ends there.
Does the estimated card count learn my pace?
Yes. As you review, immit measures your seconds per card in each mode and refines the estimate from your recent history, so the "≈ N cards" preview gets closer to your real pace over time. The timer is the guarantee either way.
Is Best streak a daily streak?
No. Best streak is the longest run of correct answers inside one session. Your day streak, the number of days in a row you have reviewed, is on your dashboard if you have an account.
Do Save for later and Mark as Known affect the scheduler?
They decide whether a card is eligible for review, nothing more. A status change is not a review: nothing is written to your review history, so it never counts as evidence about your memory, and the card's memory state is kept for when you bring it back.
