When SRS Becomes Counterproductive: Anki Burnout, Leeches, and the Exit Signs

When SRS becomes counterproductive, the reviews keep coming, but the learning has quietly stopped. The spaced-repetition system that once felt like progress starts to feel like a debt. Anki burnout is rarely a sign that you lack discipline. More often, it means the deck is configured to defend a target you never chose, so the fix is diagnostic rather than motivational.¹

Overview

SRS is a tool, not the goal

A spaced-repetition system (SRS) schedules reviews to interrupt the forgetting curve: a card is shown again near the point of likely forgetting, and each successful recall lengthens the next interval. Anki's FSRS scheduler frames this as a probability target.

The "Desired Retention" setting is the proportion of cards you recall successfully when they come due. Its default is 90%, so cards are scheduled to give roughly a 90% chance of recall when they appear.¹²

The scheduler exists to serve retention at minimum cost, and that cost is not fixed. When desired retention is raised, intervals shorten and review volume climbs; the manual states plainly that "Higher retention leads to shorter intervals and more reviews per day."¹²

Review time that no longer buys retention, or that reading and listening could replace more cheaply, signals that the tool is being misused. It is not evidence that you have failed.

What "counterproductive" actually means here

The dividing line is structural, not emotional. A hard week is normal; a structural problem is when the maintenance cost of the deck outgrows its learning return.

FSRS makes this measurable. A higher desired retention raises the review count very quickly above 90% and can make the workload overwhelming above 97%. As a result, a deck can be configured so that most of a session is spent defending a retention target the learner never deliberately set.¹²

Falling behind is governed by deck settings, not willpower

The manual notes that when you fall behind, "if you introduce new cards, their reviews won't appear until you've gotten through your backlog," so adding new cards while behind only deepens the pile.³ Pausing new-card intake until the backlog clears follows directly, and that lever is a setting, not a matter of discipline.

The three symptoms of SRS gone wrong

The failure shows up as some combination of three symptoms. Most stalled learners recognize at least two of them, and each maps to a concrete fix later in this article.

Symptom 1: the mass review backlog

A backlog is the pile of overdue review cards that accumulates during a break (vacation, illness) or whenever the day's reviews are not finished. Anki's default daily limits are 20 new cards per day and 200 maximum reviews per day per deck. These defaults bound the load, and the new-card limit is the lever that keeps the pile from compounding.¹

The backlog compounds because, by default, the review limit also governs new cards: no new cards are shown once the review limit is reached, and when a learner falls behind Anki "prioritizes cards that have been waiting the longest."³ A large overdue pile therefore blocks new learning and puts the oldest, often least-relevant cards first.

That ordering creates the dread loop: opening Anki shows a wall of cards you no longer care about. Because new-card reviews queue behind the backlog by the manual's own account, the natural response is to pause new cards until you are caught up.³

A break does not destroy your deck

Returning after a long absence does not require starting over. Anki factors in the delay since a card was last seen when it schedules the next interval, so you resume from where you left off rather than resetting your progress.³

Symptom 2: joyless, autopilot reviews

The second symptom is harder to measure but easy to feel: reviews become a number to grind down, not a session that teaches anything. The sense of progress disappears even though the streak continues.

Underneath that experience is a concrete failure mode: recognition substitutes for recall. An SRS only buys retention when you genuinely recall the card before the answer is shown.

FSRS schedules around the probability of successful recall when a card is due. Pressing "Good" on a card you merely recognize, rather than one you could have produced, feeds the scheduler a false success.² The interval then lengthens past the real forgetting point. When the card returns as a failure, you get effort without return.

Symptom 3: leeches everywhere

A leech is Anki's term for a card you keep forgetting. Anki counts the number of times a card lapses, meaning it is answered "Again" after it has graduated. "When this counter reaches 8, Anki tags the note as a leech and suspends the card."⁴ The default leech threshold is therefore 8 lapses.

After the first leech notification, repeat warnings fire at half the threshold: "if you set the warning at 8 lapses, future warnings will happen every 4 lapses (at 12, 16, and so on)."⁴ The threshold and the action are both adjustable in deck options.

The two leech actions are "Tag Only," which adds the leech tag and shows a pop-up, and "Suspend Card," which tags the note and hides the card until it is manually unsuspended. The default action is Suspend Card, matching the manual's statement that Anki tags the note and suspends it at threshold.⁴

A swarm of leeches poisons a session because each one is, by definition, a card that has already failed up to 8 times. Leeches consume disproportionate review time relative to the retention they deliver. That is the workload-without-return pattern that defines counterproductive SRS.⁴

Triaging the backlog

Cap before you clear

The first lever is a hard daily limit, set so the backlog stops growing while you clear it. Anki's "Maximum Reviews/Day" caps the review cards shown each day. The manual notes this setting "can help to smooth out occasional peaks in due card counts, and can save you from a heart attack when returning to Anki after taking a week off."³¹

Because the review limit also governs new cards by default, capping reviews simultaneously throttles new-card introduction, which is precisely the behaviour the manual advises when you are behind.³

When the open-ended configuration is itself the problem, J-Compass recommends Amenokori: a mobile FSRS app built on opinionated defaults. It avoids the freely tunable retention and limit knobs that invite the backlog spirals described above.⁵ Its per-level word counts are vendor leveling, not official JLPT vocabulary lists.⁵

Triage, do not grind

With the cap in place, the goal is to study what is due and cut stale, low-value cards, not to clear the whole pile. The deletion test is concrete: ask whether you would be "kicking myself if I can't remember this information later and know I deleted the card." If the answer is no, the card is a candidate for removal.⁶

The efficiency case for cutting is strong. Bjornstad's leech guidance reports that "Removing the most difficult 10% of material can improve learning efficiency by up to 300%," which means a small amount of deletion can buy back a large amount of session time.⁶

Filtered (custom study) decks let you isolate a subset for focused study without disturbing normal scheduling. That lets you pull only the cards you want to work through into a separate session.³ Restarting a deck (deleting and rebuilding) is a legitimate last resort, justified by the same efficiency argument: a deck made mostly of low-value leeches costs more to maintain than to rebuild from current, better-formulated cards.⁶

The leech protocol

Detect: lower the threshold so leeches surface early

The default leech threshold is 8 lapses. Lowering it makes Anki flag a struggling card after fewer failures, before it has rotted through many sessions.⁴ J-Compass recommends a lower threshold in the range of 4 to 6 lapses for intermediate decks, which surfaces problem cards earlier. This is a pedagogy recommendation, not an Anki default. The manual documents only the default of 8 and the fact that the setting is adjustable.⁴

Because repeat warnings fire at half the threshold, a lowered threshold also tightens the re-warning cadence. For example, a threshold of 8 re-warns every 4 lapses.⁴

Reformulate, suspend, or cut

Once a card is flagged, you have three options: reformulate it, suspend it, or cut it. The right choice depends on why the card is failing. Bjornstad's framework keys the choice to the root cause rather than to frustration.

The root causes are "imprecision and unclear wording," "memory interference" between similar cards, "missing foundational knowledge," and "poor question design, such as asking for enumerations rather than testable facts."⁶ Each cause points to a different action. Interference wants a disambiguating sibling card, overload wants splitting, missing context wants a supporting card, and genuinely low-value material wants deleting.⁶

The diagram below maps a flagged leech to its disposition by cause.

To reformulate is to fix imprecise wording, add supporting cards for missing context or interference, or reframe the question so it tests one retrievable fact rather than an enumeration.⁶ To suspend is to set a lower-priority card aside, focus on higher-priority material first, and return to it later. To cut is to delete non-critical material on the strength of the same efficiency argument.⁶ Suspending hides a card "from review until they are manually unsuspended," so a parked card waits safely until you can rewrite it.³

When to suspend, and when to stop adding

Suspending vs. deleting

Suspending is reversible parking. A suspended card is hidden from review "until they are manually unsuspended (by clicking the suspend button in the browser)," whereas deleting removes the card permanently.³

Suspend is the right move when a card may matter later or needs reformulating before it returns. Delete is right when the card is genuinely low-value.³⁶ Bulk-suspending means selecting many cards in the browser and suspending them at once. It buys breathing room without losing anything.

Burying is a one-day deferral, not a park

Burying "hides a card or all of the note's cards from review until the next day," so it only pushes work to tomorrow rather than parking it. The manual adds that "a card cannot be buried and suspended at the same time. Suspending a buried card will unbury it." When the goal is lasting breathing room, use suspend, not bury.³

Anki's own leech default uses suspension rather than deletion as the safe disposition, tagging and suspending at threshold instead of discarding.⁴ That is a useful instinct to copy: park first, delete only on reflection.

Turn off the tap

Pausing new cards follows from the manual's own account of the backlog: new-card reviews queue behind the overdue pile, so adding cards while behind only deepens it.³ The lever is the New Cards/Day limit (default 20), which can be set to 0 to halt new cards entirely.¹

A too-high retention target inflates reviews independently of new cards. The manual states that "Higher retention leads to shorter intervals and more reviews per day," and that "Above 90% the workload increases very quickly, and above 97% the workload can be overwhelming."¹

Check retention before concluding the deck is too big

A learner drowning in reviews should first check whether desired retention was pushed above the 90% default. Above that line, the review count climbs sharply on its own. The deck may be the right size but tuned to an unsustainable target.¹

When to switch to native-input-only

The intermediate handoff

The sourced case for shifting acquisition into context rests on Krashen's Input Hypothesis: learners acquire language by understanding input "a little beyond" their current level. This is the i+1 formulation. In this view, learners acquire vocabulary and structures through comprehensible context rather than deliberate study.⁷⁸

By an intermediate level, much vocabulary recurs frequently enough in reading and listening that context can carry retention an SRS would otherwise have to schedule. As your "i," or current level, rises, a larger share of new vocabulary becomes acquirable from i+1 input. That lowers the marginal value of mass review.⁷⁸

This is a methodology judgment, not a hard rule. SRS and comprehensible input are complementary, and Krashen's broader position is that comprehensible input is the central driver of acquisition; the practical claim here is only that its share of the work should grow as you advance.⁸

How to taper, not quit cold

The point of tapering is to keep the SRS gains you already have while letting immersion carry more of the load. No single source prescribes an exact ratio, so treat the ratio itself as your own judgment. The levers below are each sourced.

Lower the New Cards/Day limit toward a minimal high-value intake rather than zeroing it and abandoning the deck, since the new-card lever lives in deck options.¹ Keep desired retention at or below the 90% default so the maintained deck stays cheap to service.¹

Suspend rather than delete the cards you are de-emphasizing, so existing SRS gains are parked and recoverable rather than lost.³ The acquisition rationale is again Krashen: input supplies the recurring exposure that consolidates already-seeded vocabulary. A thin, high-value deck, the kind sentence mining builds from what you read, retains the items immersion does not surface often enough.⁷⁸

Good to know

The streak is not the metric

Anki schedules toward a retention probability (default 90%), not toward an unbroken-days count. The streak is not a quantity FSRS optimizes.¹² Defending a streak by adding new cards while already behind is exactly the behaviour the manual warns will worsen the backlog.³

Retention and a sustainable daily load are the real targets. A streak is a vanity metric. Protecting it is what drives the over-adding the manual tells you to stop.³

Recognition is not recall

FSRS optimizes the probability of successful recall when a card is due. Grading a recognized-but-not-recalled card "Good" feeds the scheduler a false success and lengthens the interval past the real forgetting point.² The SRS only buys retention for cards genuinely recalled before the answer is revealed.

Grade by whether you could have produced the answer, not by whether it looks familiar once shown. Recognition-grading is the autopilot failure mode that quietly undermines the scheduler.

One fact per card prevents most leeches

Bjornstad lists "poor question design, such as asking for enumerations rather than testable facts" and "imprecision and unclear wording" among the root causes of leeches. He recommends reframing a card to ask "what you actually want to know rather than difficult formulations" and adding supporting cards rather than cramming.⁶

The single biggest upstream fix is formulation. A card that tests exactly one retrievable fact rarely accumulates the 8 lapses that trigger leech status. Good formulation is leech prevention rather than leech cleanup.⁴⁶

Burnout is a config problem as often as a willpower problem

The overwhelm is often produced by configuration, not effort. An over-raised desired retention drives review count up sharply ("Above 90% the workload increases very quickly, and above 97% the workload can be overwhelming").¹ An uncapped or high review limit lets the pile grow,³¹ and the default leech threshold of 8 lets failing cards linger through many sessions before they are flagged.⁴

The manual's own remedies are setting changes: stop new cards, lower retention, cap reviews, and adjust the leech threshold.⁴³¹ That is the sourced basis for changing the settings rather than pushing harder against them.

See also

• Why Anki Has Become Painful (and How to Fix It): A Diagnostic Triage
• Beyond Anki: SRS Tools and Approaches Compared
• The Comprehension Threshold: How Easy Should Japanese Input Be?
• How to Get Back into Japanese After a Break: Recovering from Study Burnout
• Pure Input vs. Structured Study: How to Split Your Japanese Time at Each Level
• Second-Language Acquisition: A Primer for Japanese Learners

References

Footnotes

1. Anki Manual. "Deck Options" (Daily Limits and FSRS / Desired Retention sections). docs.ankiweb.net. https://docs.ankiweb.net/deck-options.html ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹ ↩¹² ↩¹³ ↩¹⁴ ↩¹⁵

2. Anki Manual. "FSRS." docs.ankiweb.net. https://docs.ankiweb.net/deck-options.html#fsrs ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

3. Anki Manual. "Studying" (sections on Suspending, Burying, and Falling Behind). docs.ankiweb.net. https://docs.ankiweb.net/studying.html ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹ ↩¹² ↩¹³ ↩¹⁴ ↩¹⁵ ↩¹⁶ ↩¹⁷

4. Anki Manual. "Leeches." docs.ankiweb.net. https://docs.ankiweb.net/leeches.html ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹

5. Amenokori. Product landing page. https://amenokori.com ↩ ↩²

6. Bjornstad, Soren. "Leeches" (Dealing with Leeches). Control-Alt-Backspace. https://controlaltbackspace.org/leech/ ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰

7. Krashen, Stephen D. The Input Hypothesis: Issues and Implications. Longman, 1985. (Comprehensible-input / i+1 hypothesis.) ↩ ↩² ↩³

8. Krashen, Stephen D. Principles and Practice in Second Language Acquisition. Pergamon Press, 1982. ↩ ↩² ↩³ ↩⁴