The LearnClash SRS Retention Curve: 72% at 7 Days [2026]

The 1/3/7/21 schedule on every memory blog is a calendar dressed up as spaced repetition.

LearnClash’s 3-stage SRS uses a performance-based memory retention curve. When we mapped the curve across the active question pool in April 2026, roughly 72% of cards pass the 7-day Known check on first attempt, 81% pass the 90-day Mastered check, and mastered cards exit the active pool at around day 97. That curve is the artifact this article describes.

Below: how the curve looks in our data, why we picked 3 stages instead of an interval ladder, what demote-by-one actually means, and three things that surprised us when we plotted the numbers. Duel me on memory psychology →

That matters because a retention curve is only useful if it changes product decisions: it should tell us which cards to show, which cards to retire, and which cards look mastered but are actually stuck in a loop. Otherwise it is just a chart.

What Is an SRS Retention Curve?

An SRS retention curve plots the percentage of cards a learner still answers correctly at each spaced interval. In LearnClash, the curve traces three checkpoints: how many cards pass the 7-day check, how many pass the 90-day check, and how many exit the pool as mastered. It’s our overlay on top of the classic forgetting curve Hermann Ebbinghaus measured in 1885.

Figure 1: The Ebbinghaus forgetting curve drops to ~33% at 24 hours without review. LearnClash’s 3-stage SRS resets the curve at the 7-day and 90-day checkpoints, then retires the card.

The unmodified forgetting curve is brutal. Ebbinghaus measured 58% retained after 20 minutes, 44% at one hour, 33% at 24 hours, 21% at one month. Murre and Dros replicated those numbers in 2015 using a different language and got nearly the same shape, 130 years later. The curve is one of the most durable findings in cognitive psychology.

Did you know? Ebbinghaus ran his entire 1885 study on himself, sitting alone memorizing nonsense syllables. The curve has now been replicated across dozens of studies in multiple languages without changing shape.

A retention curve is what you get when you draw the opposite of forgetting. Each spaced review resets the slope. The interval-by-interval pass rate is the curve’s signature, and it’s how learning apps benchmark whether an SRS algorithm is doing its job.

LearnClash’s curve has three points instead of an open-ended ladder. That’s the design choice the rest of this article unpacks.

What Does the LearnClash Retention Curve Look Like in April 2026?

In April 2026, the curve hits 72% at the 7-day Known check and 81% at the 90-day Mastered check, with hard-difficulty cards compressing to ~31%. Across the active LearnClash pool, that produces a retention shape with a finite endpoint at roughly 97 days per card. The numbers are observational, from our April 2026 internal retention mapping, and directionally validated against the full LearnClash data set.

Figure 2: LearnClash’s retention curve in April 2026. The 7-day check holds at 72%; the 90-day check climbs to 81% because cards that survived the first interval are pre-selected. Random review (no SRS) flatlines near 30-40% within the same window.

Here’s the data behind the line:

Transition	Interval	Pass rate	Demote rate
Wrong → Known	7 days	~72%	n/a
Known → Mastered	90 days	~81%	n/a
Known → Wrong (miss)	7 days	n/a	~28%
Mastered exits pool	~97 days	n/a	n/a

The results surprised us.

The 7-day pass rate (72%) sits well above what a baseline forgetting curve predicts at the same interval (~25-33% for unfamiliar nonsense material per Ebbinghaus, higher for meaningful material). The lift comes from active recall during the original encounter. Testing once at encoding and again at 7 days roughly doubles unaided retention, which matches the Roediger-Karpicke testing-effect literature.

Key takeaway: The 7-day pass rate measures memory plus the testing effect from the original encounter, which is why it sits well above raw forgetting predictions.

The 90-day pass rate (81%) is higher than the 7-day rate, which surprises people. The reason: a card that reaches the 90-day check has already passed the 7-day check, so the 90-day cohort is pre-selected for cards the player has actually encoded. Survivor bias works in the curve’s favor.

Did you know? A 2024 meta-analysis of 29 spaced-practice studies found 25-50% better long-term retention than cramming. LearnClash’s 7-day and 90-day intervals sit inside that band.

Why 3 Stages Beat the 1/3/7/21 Heuristic Most Memory Blogs Cite

The popular 1/3/7/21 schedule is a calendar dressed up as spaced repetition. LearnClash’s 3-stage SRS advances cards on accuracy, not time. We chose 3 stages because the data says you need a few performance gates to predict memory retention, not a daily review cadence borrowed from a high-school study planner.

Figure 3: The 1/3/7/21 schedule schedules everyone the same way. The 3-stage SRS branches based on whether the player actually recalled the card.

Here’s the side-by-side:

	1/3/7/21 heuristic	LearnClash 3-stage SRS
Trigger for next review	Calendar day	Player accuracy
Stages	4 (or more)	3 (Wrong, Known, Mastered)
Wrong-answer handling	Reset to day 1	Demote one stage
Mastery	Open-ended	Exits pool at ~97 days
Backed by	Study-planner blogs	Performance data

The 1/3/7/21 idea has a real cognitive-science cousin: Cepeda et al. (2008) tested 1,354 people across review gaps from 0 to 105 days and final tests up to one year out. Their finding was scaling, not a fixed schedule: the optimal review gap is roughly 10 to 20 percent of how long you want to remember the material.

Figure 4: Cepeda’s 10-20% rule visualized. LearnClash’s 7-day and 90-day intervals sit inside the band for monthly and permanent retention.

For a one-week test, review after 1-2 days. For one-year retention, review after 1-2 months. No single ladder fits every retention goal.

Retention goal	Cepeda 10-20% optimal gap	LearnClash interval
1 week	1-2 days	n/a (covered by encounter + 7-day check)
1 month	3-5 days	7-day Wrong→Known check
1 year	1-2 months	90-day Known→Mastered check
Permanent	3-6 months	Mastered exits pool

The 7-day and 90-day intervals in LearnClash sit inside that band. Seven days is the 10-20% gap for monthly retention; 90 days is the 10-20% gap for permanent retention. The two checkpoints hit two real retention windows instead of pretending one ladder fits everything.

So why three stages and not five or twenty? Two reasons.

The statistical reason

With only two interval checkpoints, you need exactly three states: pre-check, between-checks, post-check. Adding more stages would mean adding more checkpoints, and the marginal information gain past day 90 is small (Cepeda’s data flattens by ~6 months for most retention goals). More stages means more reviews per card without much retention payoff.

The product reason

Players don’t want a card to keep cycling forever. The pure-SRS school keeps reviewing forever; LearnClash retires cards. Mastered means done, not “we’ll see you in 365 days.”

Key takeaway: The popular 1/3/7/21 schedule is a heuristic, not science. Real SRS branches on performance, not calendar days. LearnClash’s 3-stage SRS keeps the science (Cepeda’s 10-20% rule) and adds a finite endpoint (mastery retires the card).

How Does Demote-by-One Differ from Anki’s Reset?

A wrong answer in LearnClash demotes the card by exactly one SRS stage. A missed Known card drops to Wrong with a fresh 7-day cooldown, not a full reset to day zero. That’s a deliberate departure from how Anki SM-2 handles wrong answers, and it matters more than people expect.

Figure 5: Three SRS algorithms side by side. LearnClash’s demote-by-one is the cleanest middle ground between SM-2’s ease penalty and FSRS’s full-model rebuild.

Here’s how the three algorithms handle a wrong answer:

Algorithm	Wrong answer effect	Long-term consequence
Anki SM-2	Reduce the ease factor by 0.20	”Ease hell” if a card flips repeatedly
Anki FSRS	Rebuild difficulty/stability/retrievability	Marginal ease change per review
LearnClash 3-stage SRS	Demote one stage; cooldown restarts	No ease drift; clean state machine

SM-2 has a known failure mode the Anki community calls ease hell: cards that repeatedly drift below default ease pile up and become impossible to clear. The open-spaced-repetition team built FSRS partly to solve that. FSRS reaches the same memory retention as SM-2 with 15-30% fewer reviews by modeling difficulty, stability, and retrievability separately rather than collapsing them into a single ease number.

But what does that look like in practice?

Did you know? FSRS officially landed in Anki 23.10 and reached mainstream adoption by 2025. The Anki community recommends it as the default for new decks.

LearnClash’s demote-by-one rule sidesteps both failure modes. There’s no ease factor to drift. There’s no per-card retrievability model to mistune. A card is in exactly one of three states, and a wrong answer moves it back exactly one state. So the algorithm is trivial to reason about and trivial to debug. A player can look at any card and know what state it’s in and why.

The trade-off is gentler punishment for a miss. Anki SM-2 can swing a stuck card into a daily review loop until it’s relearned. But LearnClash’s 7-day Wrong cooldown means a missed card waits a full week before you see it again. That’s too long if you’re cramming for a Friday exam, intentional for the lifelong-learning loop the 3-minute duel format is built around.

Key takeaway: Demote-by-one is a state-machine SRS, not a parameter-tuning SRS. It loses fine-grained ease control but gains transparency and avoids ease hell.

What Does the 90-Day Mastery Threshold Actually Mean?

In LearnClash, a card is mastered when it clears both the 7-day Known check and the 90-day Mastered check on first attempt. Mastered cards exit the active review pool at roughly day 97 and never come back. The active SRS pool stays clean instead of accumulating thousands of “we’ll see you in 5 years” cards the way Anki decks do.

Figure 6: A card’s path from first encounter to mastery. The 97-day exit point is what makes the LearnClash retention curve finite instead of open-ended.

The 90-day interval is the back half of the Cepeda 10 to 20 percent retention-scaling rule. For knowledge you want to keep for one year, the optimal final review sits between days 60 and 110. Ninety days is the midpoint of that window.

Figure 7: Short-term to long-term storage at 90 days. Surviving the gap is the consolidation signal LearnClash retires the card on.

A card that survives a 90-day gap has crossed from short-term cortical buffer into long-term cortical storage. That’s what the 2025 ventromedial prefrontal cortex study on spaced learning describes at the neural level.

Why retire mastered cards at all? Three reasons.

Reason	Why it matters in LearnClash
Cognitive load	A cleared card burns 1 of 18 duel-question slots; cost is much higher than Anki’s “few seconds”
Player motivation	Mastery has to mean done, not “see you in 5 years”
Pool freshness	Active topics use prime counts (37, 43, 47, 53, 89); retirement keeps pools near target

The cognitive-load argument is the easy one. A card a player has cleared at 90 days is no longer informative for the SRS scheduler. Re-asking the same question burns a duel slot that could go to a card the player still needs to learn. Anki keeps reviewing forever because Anki is a personal-knowledge tool; the cost of a 5-year-out review is a few seconds.

Key takeaway: The retirement choice is product-driven, not science-driven. Pure SRS would keep reviewing; LearnClash retires because each duel slot is expensive.

The player-motivation argument is subtler. Mastery has to mean something. If the card comes back five years later, “mastered” is just a label. By retiring the card with a sentinel date, LearnClash ties the word to a real terminal state.

The pool-freshness argument is the one most pure-SRS apps ignore. Active topics in LearnClash use a prime-number question count (37, 43, 47, 53, 89). When mastered cards exit, the active pool stays close to that target size. A topic with 200 review-due cards is just slower than one with 47 active cards plus a bank of mastered ones, not more rigorous.

How Does the Retention Curve Hold Up Across Difficulty Tiers?

The shape of the retention curve holds across difficulty bands, but the height drops sharply. Easy LearnClash questions pass the 7-day check at ~82%, medium at ~56%, hard at ~31%. The accuracy gap matches the macro accuracy data on the LearnClash by the numbers page and tracks the predicted desirable-difficulty zone.

Figure 8: Retention by difficulty band. The hard-question line compresses to ~31% across re-encounters; the easy line floats above 80% throughout.

The numbers split like this:

Difficulty	7-day pass rate	90-day pass rate	Cards reaching mastered
Easy	~82%	~88%	~70% of cohort
Medium	~56%	~71%	~38% of cohort
Hard	~31%	~52%	~14% of cohort

A first reaction is “fix the hard questions, the pass rate is too low.” But that’s the wrong instinct. The 31% hard-tier pass rate is intentional, and it’s what the desirable-difficulty zone described in Bjork’s research on training conditions actually predicts.

Figure 9: Bjork’s desirable-difficulty zone. Recall at 31% sits at the boundary where retrieval barely succeeds, which is where retention lift is largest.

Recall at 31% means the player is just barely retrieving the answer, which is the strongest version of the testing effect. And Roediger and Karpicke (2006) showed that retrieval just barely succeeding is what drives the largest long-term memory retention lift.

The flatness of the hard-tier line is also expected. Easy cards float at 80%+ across all checkpoints because the recall is automatic. Hard cards compress because each successful recall barely strengthens the trace; the cognitive effort is roughly the same on encounter two as on encounter one. That compression is the curve’s signature for a difficulty band sitting at the edge of recall.

Did you know? A LearnClash hard question is engineered around a specific wrong-answer trap: an answer that feels right because it’s adjacent to the correct one. “What year did the Berlin Wall fall?” pulls 1990 (reunification year) instead of 1989 (the actual fall). The 31% pass rate is what happens when distractors are plausible by design, not filler text.

The cross-difficulty data has a practical implication for how to study effectively: players climbing through Silver and Gold tiers shouldn’t grind harder questions inside one topic. They should add more topics. Hard accuracy stays near 31% across ELO tiers; what climbs with rank is breadth, not depth at the recall edge.

Train your brain with a duel →

What Surprised Us When We Mapped the Curve?

When we plotted the LearnClash curve in April 2026, three findings broke our prior assumptions hard enough to log here. The 7-day pass rate is bimodal, hard-tier retention compresses across re-encounters, and demote-by-one creates a “stuck card” failure mode we hadn’t predicted. Each one changed how we think about the next iteration of the algorithm.

Figure 10: The 7-day pass rate splits cleanly by response speed. Fast confident recalls clear at 88%; slow uncertain recalls fall to 51%. The middle is where the fluency illusion lives.

Surprise 1: the 7-day pass rate is bimodal by response speed

When we split the 7-day check by how long the player took to answer, the distribution stopped being a single smooth curve. Answers under 8 seconds passed at ~88%. Answers over 15 seconds passed at ~51%. The middle band (8-15 seconds) sits where the fluency illusion lives, where the player feels recall is happening but slow recognition reconstructed from partial cues is what’s actually going on. Bjork has called this the most counterintuitive finding in SRS research, and our 7-day data is a direct hit on it. So future versions of the algorithm will probably weight response speed alongside correctness, not just final-answer accuracy.

Surprise 2: hard-tier retention compresses, doesn’t expand

We expected the hard-tier line to climb across re-encounters because each successful recall should strengthen the trace. The 31% / 52% pass rates at 7 and 90 days look like a climb at first glance. But most of that lift is survivor bias.

What we expected	What the data showed
Hard-tier line climbs across re-encounters	Hard-tier line is functionally flat
Per-encounter consolidation delta	Per-encounter delta stays small
Pass rate gap closes by encounter 3	Pass rate gap stays open

Cards that pass the 7-day check are pre-selected for cards the player partially knew. The unselected cohort (cards that failed and demoted back to Wrong) re-enters the funnel at ~31% again. So the functional retention curve for hard cards is flat, which is consistent with desirable-difficulty research: cards at the recall boundary need many encounters to consolidate.

Surprise 3: demote-by-one creates a stuck-card failure mode

A small subset of cards (~3-5% of hard cards in a topic) bounce between Wrong and Known indefinitely. The cycle:

1. Pass the 7-day check (Wrong → Known)
2. Miss the 90-day check (Known → Wrong, 7-day cooldown)
3. Pass the 7-day check again (Wrong → Known)
4. Miss the 90-day check again (Known → Wrong, repeat)

We call these stuck cards. They never reach mastery and never exit the pool, so the topic’s effective active size grows slowly over time as stuck cards accumulate. The next iteration of the algorithm will likely add an explicit escape valve: after N round-trips between Wrong and Known, force a Mastered transition or a topic-level reset.

Key takeaway: These are first-hand observations from the April 2026 retention mapping. The percentages are directionally correct rather than audited; the pattern is clear enough that we’re already designing around it.

How LearnClash Uses the Retention Curve in Practice and Duel Modes

LearnClash hides the retention curve inside competitive gameplay. Practice mode is a 9-question session prioritized by retention-curve risk; Duel mode mixes SRS-due cards into the 18-question game alongside fresh ones. The curve runs while you play, which is the differentiator vs solo flashcard apps where SRS feels like homework.

Figure 11: Practice mode runs the SRS in the open; Duel mode runs it under the surface. The curve drives both.

Mode	Cards per session	SRS visibility	Primary purpose
Practice	9	Foreground	Pure SRS exposure, no opponent
Duel	18 (6 rounds × 3)	Invisible	Competitive recall + SRS
Quest	Variable	Tracked across chain	Topic-level mastery progression

In Practice mode, the algorithm picks 9 cards weighted by where they sit on the retention curve. Cards approaching the 7-day Known check get priority, cards approaching the 90-day Mastered check get the next priority, and new cards fill the remaining slots. The session takes roughly 3 minutes, the median LearnClash session length, and it’s pure SRS exposure with no opponent.

Practice card-selection priority order:

1. Cards within 24 hours of their 7-day check
2. Cards within 24 hours of their 90-day check
3. Cards last seen but not yet at a checkpoint
4. New cards (never encountered)

In Duel mode, the curve runs invisibly. Of 18 questions across 6 rounds, the matchmaker pulls a mix of SRS-due cards from the player’s pool and fresh cards weighted by topic and difficulty. The player’s opponent has no visibility into which cards are scheduled reviews and which are new. That competitive pressure adds emotional encoding solo apps can’t reproduce: the answer you got wrong in front of an opponent is one you remember.

Did you know? Players in competitive quiz games remember 25% more information in just 5 days compared to non-competitive review, per published competition-and-memory research.

In Quest mode, the curve runs across topic chains. Work through a topic’s question pool, and the SRS tracks mastery across the entire chain. By the time a player finishes a Quest, the topic’s mastered-cards count tells you which questions cleared both checks, which are still cycling between Wrong and Known, and which never made it past the first 7-day cooldown.

Key takeaway: The retention curve is never the foreground. It’s a “review your 47 due cards” prompt in solo apps; in LearnClash it sits underneath the game and shapes which cards appear, mediated entirely through duels and Quests. That’s why the 3-stage SRS spaced repetition foundations still feel like a game.

Practice your study technique →

The Bottom Line

LearnClash’s 3-stage SRS retention curve is a finite, performance-based version of the classic forgetting curve. Wrong cards return at 7 days, Known cards return at 90 days, Mastered cards exit the pool at around day 97. The curve hits 72% at the 7-day check and 81% at the 90-day check, sits inside Cepeda’s 10 to 20 percent retention-scaling band, and avoids the ease-hell failure modes pure SM-2 produces.

Figure 12: The four design choices that make the curve work. Pull any three without the fourth and the system stalls.

The trade-off is gentler wrong-answer punishment, which fits the lifelong-learning loop the 3-minute duel format is built around. The curve is also the foundation for two follow-on artifacts: a Substack essay version of this post for the founder-narrative read, and a YouTube video walkthrough where you can watch the curve plot live.

Key takeaway: The retention curve is the moat. The 7-day and 90-day checkpoints are inside Cepeda’s evidence band; the demote-by-one rule avoids ease hell; the 97-day exit makes mastery mean something. Remove any of the three and the system stops working.

If you want to read	Where to go
The broader LearnClash data set (ELO, accuracy, sessions)	LearnClash statistics page
The spacing-effect history and Leitner-to-Wozniak lineage	Spaced repetition foundations
The full nine-method study routine using the curve	How to study effectively
How Quizlet’s “SRS” actually works (it isn’t real SRS)	Does Quizlet have spaced repetition?

Explore more learning science: Spaced Repetition: Never Forget What You Learn | Does Quizlet Have Spaced Repetition? | The Testing Effect: Why Quizzes Beat Rereading | How the ELO Rating System Works | All Learning Science Articles