Alto
dev

Spaced Repetition

Choose a scheduling algorithm for your profile under Settings → Practice → Question scheduling. ASRS and ARTS let you practice for as long as you like. SM-2 and FSRS give each card a review date. With either of those, the exercise pauses when you are caught up. Another review can come due later the same day.

Consider a single Note Identification question, F♯ in treble clef. A scheduler can choose whether to show F♯ on the next turn, or it can set a due date for its next review. Even schedulers that make the same kind of decision use different evidence.

Scheduler What changes after an answer? How does F♯ return?
ASRS Each attempt changes its strain. A weighted draw on each turn. Its weight recovers as seconds pass.
ARTS The first answer sets how fast its priority grows. It returns when its priority ranks highest. Each intervening question raises that priority.
SM-2 The first answer updates its run of successes and ease factor. Its review interval follows a sequence of day counts, with a short retry after a miss.
FSRS The first answer updates its estimated memory stability and difficulty. After the learning steps, it gets a review date near the model’s 90% recall point.

Self-grading a new card

In Settings → Practice, select Grade correct answers with Hard, Good, or Easy. After a correct answer, Hard and Good keep a new card in its short learning steps. Easy skips those steps. SM-2 starts its first Easy review at 4 days. With Alto’s default FSRS settings, the starting estimate is about 8.3 days.

A correct Note Identification answer with the review intervals above Hard, Good, and Easy A correct Note Identification answer with the review intervals above Hard, Good, and Easy
Choose a grade to schedule the card at the interval above that button.
First correct answer SM-2 FSRS
Hard 5 min 30 s 5 min 30 s
Good 10 min 10 min
Easy 3–5 days 6–10 days

Alto spreads reviews across nearby days so that cards learned together can come due on different days. This adjustment is called fuzzing. For example, two new FSRS cards can show 7 days and 9 days for Easy even though both start with the same estimate. If you reopen the same card before grading it, its preview stays the same.

ASRS

The Alto Spaced Repetition System (ASRS) is the default scheduler. For each turn, it generates a few questions and adds some you have answered before under the same exercise settings. It draws from that pool at random, giving more weight to questions you find difficult. A question you just answered gets less weight for a while. The better you know it, the longer that effect lasts.

Strain

Each question has a score called strain, which says how much practice it needs. It starts at 1.5, stays between 0 and 6, and changes with each answer.

Answer Change in strain
Wrong Add 1.5 for each miss.
Right on the first try Subtract 0.5, then adjust for speed.
Right after a miss Subtract 0.2.

Response time

For a correct first answer, ASRS compares the time you took with the exercise’s par time. At par there is no speed adjustment. At half the par time or faster, ASRS subtracts another 0.6. At four times par or slower, it adds 1.0. The adjustment changes gradually in between. A slow but correct answer can therefore make the question more likely to return.

Par times by exercise
Exercise Par time
Note Identification, Keyboard Note Identification 3 s
Key Signature Identification, Clef Basics, Notation Identification 5 s
Interval Identification, Keyboard Interval Identification, Chord Quality, Note Construction, Key Signature Construction 6 s
Circle of Fifths 7 s
Scale Identification 8 s
Arpeggio Identification 9 s
Order of Keys 12 s
Scale Construction 25 s

Probability of the next draw

For a question with strain \(s\), ASRS sets a recovery time \(c\) in seconds. If \(t\) seconds have passed since the last answer, its weight \(w\) recovers according to

\[c = 120 \cdot 2^{3-s}\] \[w = e^{0.6s}\max\left(0.02,\,1-e^{-t/c}\right).\]

Higher strain gives a question both a higher eventual weight and a shorter recovery time. A fresh question gets a weight of \(e^{0.6\cdot 1.5}\approx 2.46\). To get the probability of drawing a question, divide its weight by the sum of all eligible weights.

Suppose you miss F♯ once, moving its strain from 1.5 to 3, and leave it there. Later, the only two eligible candidates for a draw are F♯ and an unseen B. At strain 3 the recovery time is 120 seconds. After two minutes, F♯ has weight \(e^{1.8}(1-e^{-1})\approx 3.82\). Its chance of coming up next is

\[\frac{3.82}{3.82+2.46}\approx 61\%.\]

B can still win that draw.

ASRS probability of drawing F♯ With one unseen alternative, a missed F sharp has a 35% chance at 30 seconds, 49% at one minute and 61% at two minutes. 0 60 120 180 240 Seconds since the last answer to F♯ 0 20 40 60 80 Chance of drawing F♯ (%) 61%
With the same two candidates, elapsed time alone changes the odds.

ARTS

Adaptive Response-Time-based Sequencing (ARTS) measures the gap between questions in turns. Its clock advances each time you give a first answer in this exercise. Retries during the same showing and time away from practice leave the clock where it was.

Priority per turn

Let \(h\) be the number of intervening first answers since you last answered F♯, and \(T\) the time in milliseconds you took on that first attempt. Alto uses a 100 ms reference time. The natural logarithm, \(\ln\), makes a slower correct answer count more, with a diminishing increase for each extra millisecond. Mettler, Massey, and Kellman, equation 1

Last first answer Priority now
Unseen \(1\)
Incorrect \(2h\)
Correct \(0.11h\ln\!\left(\frac{\max(T,101)}{100}\right)\)

Alto compares eligible questions under the current exercise settings and selects the highest priority. It breaks ties at random.

A fast and a slow identification

Suppose F♯ and an unseen B are eligible. If you last identified F♯ in one second, after three other questions its score is \(0.11\cdot 3\ln(10)\approx 0.76\). B wins with 1. Had you taken four seconds, the same three questions would give F♯ a score of \(0.11\cdot 3\ln(40)\approx 1.22\), and F♯ would win. If you missed F♯, just one intervening answer would give it a score of 2.

ARTS priority by intervening answers An unseen question stays at priority one. A four-second correct answer overtakes it after three other questions. A one-second answer overtakes it after four. 0 1 2 3 4 Other questions answered since F♯ 0.0 0.5 1.0 1.5 2.0 Priority score Correct in 4 s Correct in 1 s Unseen question
After a correct answer, response time sets the increase in priority per turn.

SM-2

SuperMemo 2 (SM-2) keeps a count of successive successful reviews, the last interval in days, and an ease factor \(E\) for each card. A new card starts with \(E=2.5\). By default, Alto grades a correct first answer Good and a miss Again. Under Settings → Practice, select Grade correct answers with Hard, Good, or Easy to choose the grade yourself. Each button shows its interval above it. A wrong answer records Again and advances automatically. The exercise’s Next Question preference is disabled while self-grading is on. Original SM-2 specification

Review intervals and ease

First submission with automatic grading Change to the card Next review
Good (grade 4) Add one success. Keep the current ease. 1 day for the first success and 6 days for the second. After that, \(I_{n}=\lceil I_{n-1}E\rceil\), the previous interval times the ease, rounded up to a whole day.
Again (grade 2) Reset the success count and interval to zero. Set the ease to \(\max(1.3,E-0.32)\). 1 minute. The next showing starts a new review.

Self-graded SM-2

The buttons use Anki’s default learning steps. A new card shows 5 min 30 s for Hard and 10 min for Good. Easy skips both steps and schedules a review around 4 days later. At the second step, Hard repeats the ten-minute step and Good graduates the card with a one-day interval.

Once a card graduates, Hard starts at 1.2 times the previous interval. Good multiplies the previous interval by the card’s ease, and Easy applies a further 1.3 multiplier. Good credits half the days overdue and Easy credits all of them. Hard lowers ease by 0.15, Easy raises it by 0.15, and Good preserves it. A miss lowers ease by 0.20 and starts a ten-minute relearning step. At that step Hard waits 15 minutes, Good schedules one day, and Easy schedules two days. Anki deck options

Three reviews of F♯ with automatic grading

Identify a new F♯ correctly on day 0 and it is due on day 1. Get it right on day 1 and it is due on day 7. A third success on day 7 gives an interval of \(6\cdot 2.5=15\) days, so the next due date is day 22.

SM-2 interval after each successful review The first, second and third consecutive successes on a new card give next intervals of one, six and fifteen days. 0 3 6 9 12 15 Days until the next review First success Second success Third success 1 6 15
Each bar starts at a successful review and measures the gap to the next due date.

Suppose instead that you miss F♯ on day 7. Its ease falls to 2.18 and Alto schedules a retry one minute later. Three successive correct reviews from there give gaps of 1, 6, and 14 days. The shorter third gap comes from the ceiling calculation \(\lceil 6\cdot 2.18\rceil=\lceil 13.08\rceil=14\).

FSRS

The Free Spaced Repetition Scheduler (FSRS) models the chance you can recall a card as time passes. Alto uses FSRS-6 with its default parameters. A correct first submission earns Good by default. With self-grading on, you can choose Hard, Good, or Easy after a correct answer. A miss always earns Again. For a new card, Good puts its next learning step ten minutes away and Again puts it one minute away. A miss on a card already in review starts a ten-minute relearning step. Python FSRS scheduler

With self-grading, a new FSRS card starts with the same Hard and Good learning steps as SM-2. Easy graduates it directly using FSRS-6’s estimate of memory stability at 90% desired retention. On review cards, the three intervals use the same random adjustment and are ordered Hard, Good, Easy up to the maximum of 36,500 days. The interval printed above a button is the one Alto schedules. Anki review scheduling

Estimated recall and the review date

FSRS tracks difficulty (\(D\), from 1 to 10) and stability (\(S\), measured in days). A stability of 8 days means the model expects a 90% chance of recall eight days after a review. Retrievability (\(R\)) is the predicted chance of recall after \(t\) whole days. With Alto’s default parameters the curve is approximately

\[R(t)=\left(1+0.98035\,\frac{t}{S}\right)^{-0.1542}.\]

The factor 0.98035 is chosen so that \(R(S)=0.90\). At the 90% target the base review interval therefore equals the updated stability in days. FSRS rounds it to whole days and may vary the scheduled day slightly to spread reviews out.

FSRS predicted recall at two stabilities A card with two days of stability reaches the 90% target on day two. A card with eight days of stability reaches it on day eight. 0 2 4 6 8 10 Days since the last review 75 80 85 90 95 100 Predicted recall (%) Stability 8 days Stability 2 days 90% target
Both cards start at 100% estimated recall. The more stable card takes longer to reach the review target.

A successful review on day eight

Suppose your F♯ card has \(D=5\) and \(S=8\) days. On day eight its predicted recall is 0.90. You identify F♯ correctly. For a Good review at least a day after the previous one, the stability update has two parts. The factor \(k\) depends on the card’s current difficulty and stability.

\[k = e^{1.8722}(11-D)S^{-0.1666}.\] \[S_{\mathrm{new}}=S\left[1+k\left(e^{0.796(1-R)}-1\right)\right].\]

The quantity in square brackets is about 3.29, giving a new stability of 26.29 days and a base interval of 26 days. The term involving \(R\) explains why timing matters. If you recall the same card earlier, \(R\) is higher and the gain in stability is smaller. A miss uses a separate stability update before relearning.

Review intervals for F♯

Suppose you see a new F♯ card five times in Note Identification. Each row shows your answer and the gap until the next showing. For ASRS and ARTS, assume you answer one other question every 10 seconds. After each of those answers, F♯ competes with one unseen card. For SM-2 and FSRS, assume you review F♯ whenever it is due.

Showing and answer ASRS (median gap) ARTS SM-2 FSRS (base interval)
1. Correct in 4 s 9 questions (90 s) 3 questions (30 s) 1 day 10 min
2. Correct in 1 s 19 questions (190 s) 4 questions (40 s) 6 days 2 days
3. Correct in 1 s 19 questions (190 s) 4 questions (40 s) 15 days 11 days
4. Miss (corrected on the same showing in ASRS and ARTS) 8 questions (80 s) 1 question (10 s) 1 min 10 min
5. Correct in 4 s 11 questions (110 s) 3 questions (30 s) 1 day 2 days

After the first answer, ASRS draws F♯ again within nine intervening questions about half the time. ARTS picks it as soon as it outranks the unseen card, after three questions. On the fourth showing, ASRS counts both the miss and the correction. ARTS uses the first answer. In SM-2 and FSRS, the miss ends the showing. The fifth row is the scheduled retry. FSRS uses base intervals here. In practice, its small random adjustment to longer intervals can also change the dates of later reviews.

Daily limits

For SM-2 and FSRS, exercise tiles show new cards in blue, missed cards awaiting a correct answer in red, and due reviews in green. Below the search bar, New, Learn, and Due show the totals for checked exercise tiles that match the search text. A missed card stays in the red count while it waits and when it becomes due. Another miss on that card still counts as one. Due reviews come first. The default allowance for each exercise is 10 new cards and 200 reviews per local day. Change those numbers in Settings → Practice, or open an exercise’s Daily Limits submenu and clear Use global to override either number for that exercise. Check Unlimited to remove the daily cap on new cards, reviews, or both. A limit of zero pauses that kind of practice.

On the exercise screen, the counts appear at the top, immediately to the left of the score and streak, in blue / red / green order: new cards, outstanding incorrect cards, and due reviews. The underlined number identifies the current card’s queue. A retry that is due appears in both the red and green counts, with red underlined while you answer it.

Keyboard Note Identification with three colored counts next to the streak and Next interval beneath the keyboard Keyboard Note Identification with three colored counts next to the streak and Next interval beneath the keyboard
With automatic grading, Next interval shows when this card will return if you answer correctly.

After a miss, Alto sets the card aside and offers other eligible questions. If the missed card is the only review left within your daily limit, the exercise shows its return time and resumes automatically when it is due.

An exercise waiting for a missed card, with its return time and the red outstanding-card count An exercise waiting for a missed card, with its return time and the red outstanding-card count
You can leave the exercise open. Practice resumes when the missed card is due.

The two algorithms share the day’s allowances and keep separate review dates. Opening a new card uses one introduction slot. If you leave before answering, you can return to that card without using another slot. The allowances reset at local midnight, and overdue reviews remain due.

You can see the next review date for a specific card in the database browser.

Switching schedulers

When a card first moves between SM-2 and FSRS, it keeps its due date, full review interval, and learning or relearning step. A card on a 180-day interval with 20 days left will still be due in 20 days. After that first transfer, each scheduler resumes its own saved schedule for the card.

For mature cards, FSRS estimates stability and difficulty from the SM-2 interval and ease, assuming 90% historical recall. In the other direction, SM-2 uses the full FSRS interval and starts ease at 2.5. FSRS conversion from SM-2

ARTS and ASRS record your answers in your practice history. Calendar due dates, new cards, missed cards awaiting review, and daily allowances are independent of that practice.

Your profile

Alto tracks individual questions, such as identifying F♯4 in treble clef in D major. Changing the order of the answer buttons still counts as practice on the same question.

Every submission is saved under your profile. A question that takes three tries leaves three entries, each with your answer, whether it was correct, and how long it took. Switching algorithms keeps that history. You can switch profiles in the menu (☰) and see your accuracy and response times in Statistics.