How many reviews does FSRS need before its personalised parameters become accurate?

The Anki team's guidance is around 200 reviews per deck before running the FSRS optimiser, and at least 1,000 reviews before the personalised parameters meaningfully outperform the global defaults. For chess repertoires under 500 positions, you may use the default parameters indefinitely - they are already fit on hundreds of millions of reviews and produce strong scheduling. Custom optimisation is most useful for serious users with 1,500+ positions and 6+ months of consistent reps.

What desired-retention setting should I use for chess openings?

90% is a sane default for general study. For tournament weeks raise to 93-95% to over-prepare. For experimental sidelines or recently added positions you are still verifying, 85% is fine - it lets the algorithm space them out faster and you save drill time on lines you may delete next month. Avoid 97%+ unless you are a working coach or correspondence player; the review load grows non-linearly above that threshold and most of the extra cost buys very little practical improvement.

Does FSRS punish me for missing a few days of reviews?

Yes, but less harshly than SM-2. When you skip days, FSRS treats overdue cards as 'partially decayed' and re-fits the next interval based on your actual recall when you review them. SM-2 effectively resets to the previous failure interval if you fail an overdue card. Practically, missing a week of reviews under FSRS produces a backlog you can clear in roughly half the time it would take under SM-2, because correctly recalled overdue cards still earn longer intervals.

Can I export my FSRS-trained Anki deck and re-import it into another FSRS tool?

The card content (PGN, FEN, comments) exports cleanly through .apkg or .csv. Review history exports too, but the raw history is the input - the receiving tool re-runs FSRS on those reviews to derive the personalised parameters. Different tools may use slightly different FSRS versions (4.5, 5, 6) which produce somewhat different intervals from the same history. Plan on a 2-3 week recalibration period after migrating, even if both tools advertise FSRS.

Why does FSRS sometimes give shorter intervals than SM-2 for cards I find easy?

FSRS uses your full review history, not just the last rating. If you rated a card Easy yesterday but failed it twice last month, FSRS treats it as still slightly fragile and waits less time before testing again. SM-2 would have applied the Easy multiplier to your last interval and shipped you out for several months. FSRS's 'shorter than expected' interval is usually the correct call - the same card under SM-2 frequently re-fails three months later, costing more total reviews than FSRS's earlier check would have.

FSRS vs SM-2 for Chess: Why FSRS Wins by 20-30%

Disclosure: ChessAtlas is our product, and we use FSRS as our default scheduler. We have aimed for a fair, source-backed comparison, but readers should weigh our perspective accordingly.

By Antoine Tamano, building ChessAtlas. Last updated May 2026.

You memorise the Ruy Lopez Closed to move 12. You drill it for a month. On game day, your opponent plays a move-order you only saw twice in your prep, and the position you thought you knew slips away. This is not a motivation problem, it is a scheduling problem. The algorithm deciding when to show you a position is the difference between lines that stick and lines that vanish under time pressure.

Most chess opening trainers still run on SM-2, the spaced-repetition algorithm Piotr Wozniak published in 1987 for SuperMemo. A newer algorithm, FSRS (Free Spaced Repetition Scheduler), was released for Anki in late 2022 and built into Anki since version 23.10 (October 2023). FSRS is now the recommended scheduler for new decks. Per the open-spaced-repetition benchmark across 500+ million review logs and the official Anki team's testing, FSRS requires roughly 20 to 30 percent fewer reviews at the same retention level.

What Are FSRS and SM-2?

Both algorithms schedule reviews to maximise long-term memory while minimising wasted effort. In chess, they decide whether you see a critical Najdorf branch tomorrow, next week, or next month based on how accurately you played it last time.

SM-2: The 1987 Heuristic Still Running Most Chess Trainers

SM-2 uses fixed multipliers and an "ease factor" that moves up or down based on your Hard / Good / Easy rating. It treats every user identically, which was the only option given the compute and data available in 1987. The common failure mode, widely documented in the Anki community, is ease hell: rate a card Hard too many times and the ease factor collapses, forcing the card to return daily long after you have actually mastered it. In chess, this happens regularly with tight move orders (Winawer French, Mar del Plata King's Indian, Botvinnik Semi-Slav) where you fumble early, rate Hard, and then never escape the short interval.

SM-2 powers Chessable's MoveTrainer, Anki's legacy scheduler, and Chess Position Trainer's custom variant. It works - it is simply not tuned to what 40 years of spaced-repetition data have since revealed.

FSRS: Personalised Scheduling Trained on Real Anki Reviews

FSRS was built by Jarrett Ye and the open-spaced-repetition organisation, released as an Anki add-on in December 2022, and integrated into core Anki since version 23.10 (October 2023). Ongoing releases through 2024 and 2025 (FSRS 4.5, FSRS 5, FSRS 6) refined the parameter set further. Default parameters are fit on hundreds of millions of reviews from tens of thousands of Anki users, then adapt to your personal review history once you have enough reps. FSRS models memory with three variables:

Retrievability (R): probability you recall a position right now
Stability (S): days required for retrievability to drop from 100% to 90%
Difficulty (D): how hard this specific position is for you

You also set an explicit target retention (often 85 to 95 percent). FSRS schedules intervals to meet that target on a per-card basis. The Expertium benchmark, last updated through 2025 against 500M+ Anki review logs, consistently puts FSRS ahead of SM-2 in prediction accuracy by log-loss for 99.5% of users tested. Community migrations - the official Anki team's testing, third-party tool migrations - report this translates to roughly 20-30% fewer reviews at equal retention.

Why FSRS Matters Specifically for Chess Opening Retention

Chess openings form deeply interdependent trees. A single choice on move three shapes the playable options five moves later, and one memory lapse collapses the line. A serious club-level repertoire typically runs to a few thousand position cards once transpositions and opponent sidelines are included. The following four properties of FSRS map directly onto these demands.

1. Personalisation Between Forced and Strategic Positions

Some positions stabilise fast in your memory once understood, because the only correct continuation is forced. Others remain fragile for weeks because several plausible-looking moves exist and only one is best. FSRS learns which is which for you; SM-2 applies the same multipliers to both.

Najdorf Sicilian after 7.Nb3, a fragile move-order tabiya — Najdorf Sicilian after **1.e4 c5 2.Nf3 d6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3 a6 6.Be2 e5 7.Nb3**. The kind of subtle move-order tabiya FSRS keeps on a tighter interval for most players, because small differences (Nb3 vs Nf3) flip the evaluation.

By contrast, a pure tactical refutation - punishing an early ...Qh4 with Nc3 in some lines - tends to stabilise within a handful of correct recalls. FSRS widens the interval on that card quickly, freeing review time you would otherwise waste.

2. Escape from Ease Hell on Sharp Lines

With SM-2, repeatedly rating a complex Grünfeld or Mar del Plata line as Hard drops the ease factor, forcing daily reviews long after you have internalised the position. FSRS recalibrates stability every time you recall correctly, so once the line clicks, the interval stretches out. Reviews follow your actual learning curve rather than punishing you for the weeks you struggled. For the broader retention strategy this enables, see our complete FSRS guide for chess openings.

3. Configurable Retention for Tournament Prep

Before a weekend open, you may want 95% retention on your main lines and accept 85% on sidelines. FSRS lets you set an explicit retention target, per deck or globally, and adjusts intervals to hit that goal. SM-2 has no equivalent dial.

4. Transpositions Handled as Shared Positions

Two different move orders can reach the same tabiya. FSRS-driven trainers that store by position rather than by sequence recognise this and schedule one card rather than two. This is an implementation detail more than an algorithm property, but it pairs naturally with FSRS-based tools and keeps your deck size grounded in real positions, not in cosmetic move-order duplicates.

How FSRS Works Under the Hood

After each review, FSRS updates stability (S), difficulty (D), and retrievability (R) for that specific card. The next interval is computed so that R drops to your target retention at exactly that time. The key difference from SM-2 is that the entire review history informs the next interval, not just the last rating. SM-2 uses only the latest rating to pick the next interval; if you blew the card three weeks ago and have nailed it five times since, SM-2 cannot weigh those signals against each other.

Personalised spaced-repetition scheduling based on per-card memory model

Default FSRS parameters come from the large-scale Anki dataset, so the algorithm performs reasonably from day one. With more of your own reviews, the optimiser can be rerun to fit parameters to your specific forgetting curve. On Anki this happens in the deck settings; on ChessAtlas it is automatic.

What Changed When Chess Trainers Switched to FSRS

Anki: FSRS is Built-In and Recommended

FSRS has been integrated into Anki since version 23.10 (October 2023) as a built-in scheduler option. Anki recommends it for new decks, and the FSRS optimiser runs automatically once you have around 200 reviews of history. Older versions need an upgrade - the legacy FSRS add-on is no longer needed.

Third-party migrations

Several chess and language-learning trainers migrated from SM-2 (or in-house variants) to FSRS during 2024-2025 and reported review-load reductions in the 20-30% range at equal retention - the same number the open-spaced-repetition benchmark predicts. Users with deep, sharp repertoires (Sicilian, King's Indian) tend to see the upper end of that range because their cards include more "ease hell" candidates that FSRS handles cleanly.

ChessAtlas: FSRS from Day One

ChessAtlas ships with FSRS as the default scheduler, and pairs it with automatic Lichess and Chess.com game import plus Deviation Finder. When Deviation Finder flags a position you left your prep in a real game, that card is pushed to the review queue with FSRS scheduling. The feedback loop between study and real games stays tight.

For how these tools stack up in detail, see our 2026 trainer comparison.

Quick Comparison: FSRS vs SM-2 at a Glance

Feature	SM-2 (1987)	FSRS (2022)
Trained on real user data	No, fixed heuristic	Yes, hundreds of millions of reviews
Personalisation per user	No	Yes, after sufficient reviews
Explicit retention target	No	Yes (often 85 to 95%)
Escapes "ease hell"	Manual workarounds only	Yes, stability recalibrates
Memory model variables	Ease factor only	Retrievability, Stability, Difficulty
Review reduction at same retention	Baseline	~20 to 30% fewer
Used by	Chessable, Anki (legacy), CPT	Anki 23.10+ (built-in), ChessAtlas, recent Anki migrations

Common Misconceptions

"FSRS only helps if I have thousands of cards"

Default FSRS parameters already improve scheduling on small decks thanks to the global training data. Your personal fit sharpens over time, but configurable retention and dynamic intervals reduce waste from the first review. If your repertoire has 500 to 1,000 positions, the configurable retention dial alone is worth the switch.

"SM-2 is obsolete"

SM-2 still works. Chessable has shipped world-class courses on SM-2 for years and players legitimately memorise openings with it. The argument is not that SM-2 is broken, it is that FSRS is measurably more efficient and the switching cost is low. If a tool you love uses SM-2, stay; if you are picking a new trainer, pick FSRS.

"Switching scheduler resets my progress"

Migrations in Anki preserve review history. FSRS seeds initial parameters from your past ratings, then calibrates over a few weeks of reviews. You keep intervals and accuracy signals rather than starting from scratch.

Which Tools Use Which?

FSRS by default: ChessAtlas, Anki 23.10+ (built-in option, recommended for new decks)
FSRS opt-in: Older Anki decks (enable in deck options), community tools like Chessdriller
SM-2 still: Chessable MoveTrainer, Chess Position Trainer (legacy custom scheduler close to SM-2)
No spaced repetition at all: ChessBase (database tool), Lichess Opening Explorer, most engine analysis tools

For the broader repertoire-tools landscape, see our 7 tools roundup and the Chessable alternatives guide.

Action Plan

If you use Anki, upgrade to 23.10 or later and toggle FSRS on in Deck Options. Set desired retention to 90% to start.
If you use Chessable and your repertoire is large, export the PGN and drill it in parallel in a FSRS-based tool like ChessAtlas.
Give FSRS at least two weeks before judging - the optimiser needs roughly 200 reviews of personal history to fit your forgetting curve.
Track two metrics for the next month: daily review count and subjective recall under time pressure. The 20 to 30% review reduction shows up clearly within four weeks.
Adjust desired retention up before tournaments (95%) and back down after (85-90%) to manage drill load between events.

For the broader study framework this algorithm plugs into, read how to build your first repertoire, why spaced repetition works for chess, and how to memorize chess openings and actually remember them. Or create a free ChessAtlas account and start drilling with FSRS in two minutes.