Flashcard workload calculator

The model

A spaced repetition scheduler shows each card at growing intervals. In this model every card follows the same rounded due-day ladder. If you add n cards every day, one group of n cards comes due for each review step reached so far:

reviews on day d = n × (number of due days at or before d)

Each new due day adds another daily group of reviews. Once every step in the ladder is active, the load reaches its plateau:

steady-state reviews/day = n × (number of reviews a card ever gets)

Which is why the familiar rule of thumb — “reviews land at about ten times your new-card rate” — holds: a card gets about ten reviews in its life.

The ladder here starts at one day and multiplies by a growth rate at each step until it passes a 20-year horizon, after which a card is treated as learned and stops counting. Each retention setting selects an assumed growth rate; these settings are not calibrated predictions of your own retention. At 90% it is 2.2, a little below SM-2's default ease of 2.5, so a card added today falls due on days 1, 3, 8, 19, 42, 94, 207, 456 and gives about 11 lifetime reviews. At 95% it is 1.7 — smaller steps, about 16 lifetime reviews. At 85% it is 2.6, about 9. Higher retention costs reviews because the intervals grow more slowly, not because each review takes longer.

“Reviews per card, first year” is the count of steps whose cumulative due date falls inside 365 days. It is the honest way to read the cost of adding a card: not one review, but six or more in this model.

“Steady state” takes years, and that matters

The headline number is the long-run plateau: what the load becomes if you hold the same new-card rate until the earliest cards have intervals measured in years. Almost nobody experiences that number early. At 20 new cards a day and a 90% target, the model puts the load near 80/day after a month, 100 after three, and 140 after a year — the plateau of 220 only arrives if the rate holds for several years.

This is why the ramp table matters more than the headline for anyone starting out, and why the practical guidance in how many flashcards per day quotes a lower range for the first months. Both are the same curve read at different points.

How the finite deck plan works

The deck planner introduces cards from day zero until the deck is exhausted. Each review step counts only the cards introduced on day − due day, including a smaller final group when the deck size is not divisible by the daily rate. Completion takes ceil(deck size / new cards per day) learning days.

The table samples the last learning day and four later days. A zero means no cards fall due on that particular day; longer intervals can still bring reviews back afterward. The ongoing-additions estimate below the planner assumes you keep adding cards indefinitely.

What the estimate leaves out

• Lapses are not modelled. A forgotten card re-enters at a short interval and pays its early steps again. At a 90% target, roughly one review in ten lapses, and each lapse adds several reviews.
• Real schedulers adapt per card. FSRS and SM-2 both set intervals from your answer history, so no two cards follow the ladder above. The ladder is an average shape, not a schedule.
• Cards are not uniform. Chinese characters with several readings, or near-synonyms, lapse far more than concrete nouns and cost more than the average.

If the calculator says 200 reviews a day, allow room for lapses and changes to the intervals. Use the number to ask “will I do this in a bad week?” rather than as a forecast.

Download and cite the model outputs

The calculator's reference outputs are available as a reusable CSV dataset. It covers 5–100 new cards per day at 85%, 90%, and 95% target retention, with review-load estimates after 7, 30, 90, 180, and 365 days and at steady state. These are model outputs, not observed learner data, and should be described that way when cited.

Suggested attribution: Fulin Labs, “Flashcard review workload model outputs,” 2026. The dataset is licensed under CC BY 4.0; link to this page so readers can inspect the assumptions and limitations.

Embed this calculator

Writing about spaced repetition? Paste this into your page to give readers the working calculator. It runs in their browser; please keep the attribution link. The datasets behind it are listed on the open datasets page.

<iframe src="https://fulinlabs.com/embed/flashcard-workload/" title="Flashcard workload calculator" width="100%" height="560" style="border:1px solid #d0d7de;border-radius:8px;max-width:640px" loading="lazy"></iframe>
<p>Calculator by <a href="https://fulinlabs.com/tools/flashcard-workload-calculator/">Fulin Labs</a>.</p>

What to do with the number

The usual mistake is picking a new-card rate from ambition and discovering the review load six weeks later, when quitting feels like the only way out. Two levers help:

• Lower the new-card rate. It moves the plateau proportionally and immediately.
• Lower the retention target. 85% instead of 95% takes lifetime reviews per card from about 16 to about 9 in this model, at the cost of forgetting more between reviews.

For the reasoning behind the rates themselves, see how many flashcards per day, FSRS vs SM-2, and spaced repetition for Chinese. To run a deck on these numbers, Fulin Flashcard schedules cards in the browser. If your cards come from Chinese texts, the Chinese text analyzer shows which characters in a passage are above your level and worth adding.

Questions

How many new flashcards per day should I add?

Pick the number whose steady-state review load you will still do on a bad day. In a mature deck reviews accumulate at roughly ten times the new-card rate, so 20 new cards a day is a commitment to a bit over 200 reviews a day, not to 20.

Why do reviews keep growing when I add the same number of cards each day?

Every new card generates a series of future reviews at growing intervals. The daily load is the sum of all cards still in circulation, so it rises for weeks after you start and only levels off once the earliest cards reach long intervals. That plateau is what this calculator estimates.

Is this an estimate or a measurement?

An illustrative estimate from the interval ladder stated on the page. Real schedulers adapt each interval to your answers, and lapses can add reviews. The retention settings are assumptions, not measured outcomes for your deck.

What happens if I stop adding new cards?

The review load decays instead of levelling off, because no new short-interval cards enter and existing intervals keep growing. Pausing new cards for a week is the standard way out of a review backlog.

Does a higher retention target cost more time?

Yes, and by more than most people expect. Higher retention means intervals that grow more slowly, so a card is seen more times before it is effectively learned. In the model on this page, moving the target from 85% to 95% takes lifetime reviews per card from about 9 to about 16 — close to double the total work for the same deck.

Think the ladder or the retention multipliers are wrong? Corrections with a source go to hello@fulinlabs.com.