What is FSRS? A practical guide to the algorithm

What FSRS means in practice

FSRS is not just "review later." It is a machine-learning spaced repetition algorithm developed by Jarrett Ye (L-M-Sherlock) that models each card as a memory state with three variables:

After every review, FSRS updates Difficulty and Stability, then schedules the next review so Retrievability is near your desired retention target when the card appears again. This usually produces fewer unnecessary reviews and fewer surprise lapses than older fixed-heuristic schedulers like SM-2.

FSRS was integrated into Anki 23.10 and is now the default scheduler for new decks. Deckbase, Mochi, and RemNote also use FSRS natively.

FSRS vs SM-2 and older schedulers

Independent benchmarks on over 700 million Anki reviews show FSRS-5 reduces retention prediction error to roughly 5.3% RMSE versus 16.2% for SM-2. In practical terms, that means FSRS can maintain the same retention level with approximately 20–30% less daily review time. Source: FSRS algorithm documentation.

Simple daily FSRS workflow

1. 1
   Keep cards atomic: one fact or concept per prompt. Smaller cards improve both recall and scheduler accuracy.
2. 2
   Study daily in short sessions (10-20 minutes). FSRS benefits from consistent review signals more than occasional long sessions.
3. 3
   Rate honestly after recall. Overrating hard cards leads to long intervals and avoidable forgetting.
4. 4
   Prioritize quality sources. For Deckbase users, scanning books/PDFs then editing AI drafts keeps card quality high.

In Deckbase, this is usually "capture source → generate cards → edit → review." If you create cards through MCP, follow your template schema with get_template_schema before bulk writes.

How to pick a retention target

FSRS works best when your target retention matches your context. A single universal number is rarely optimal: aggressive targets can inflate workload, while low targets can produce avoidable forgetting before high-stakes exams.

Start near 90-92% for most learners, then adjust after 2-3 weeks by looking at real outcomes: overdue count, daily review time, and lapse frequency. Small changes are usually better than large jumps.

Card quality checklist for better FSRS results

1. 1
   Use one clear question per card. Ambiguous prompts create noisy ratings and weaker interval predictions.
2. 2
   Prefer concrete answers over broad summaries. Specific recall events improve scheduling accuracy.
3. 3
   Add context only when needed. Extra text should disambiguate, not overwhelm the recall step.
4. 4
   Review recent lapses weekly and rewrite the worst 10 cards. FSRS performs best when input quality keeps improving.

Workload planning: retention is a budget decision

FSRS is powerful because it makes trade-offs explicit. Higher retention targets can reduce forgetting, but they usually increase daily review cost. Lower targets reduce workload, but can raise lapse risk before exams. The right setting depends on your available time, card volume, and consequence of forgetting.

Instead of chasing a single "best" number, treat retention as a budget: how many minutes per day can you actually sustain over months, not just one motivated week? Most learners improve faster when they pick a target that they can maintain consistently.

A practical default is 90-92% while you stabilize habits. Move upward only when your completion rate stays high and your daily review time is predictable.

A practical 30-day FSRS adoption plan

The biggest mistake with spaced repetition is optimizing settings before the workflow is stable. This 30-day plan prioritizes consistency first, then tuning. It works for medical learners, certification prep, language study, and technical domains where recall quality matters beyond short-term exams.

1. 1
   Days 1-7: establish baseline. Keep sessions short and daily. Cap new cards to a level you can review without backlog growth.
2. 2
   Days 8-14: fix card quality. Review your worst cards and rewrite unclear prompts. Prioritize one concept per card.
3. 3
   Days 15-21: tune retention gently. If lapses are high, increase quality before adjusting retention. If workload is too heavy, reduce new cards before lowering retention.
4. 4
   Days 22-30: lock your operating mode. Keep the settings that produce stable completion and acceptable lapse rates, then avoid frequent parameter changes.

By day 30, you should know your sustainable card volume, your realistic retention target, and where your card authoring process still creates avoidable errors.

FSRS troubleshooting guide

Most FSRS failures are not algorithm failures. They come from poor card inputs, inflated self-ratings, or unsustainable new-card volume. Use the table below to diagnose quickly before making large configuration changes.

Keep troubleshooting iterative. Change one variable at a time and monitor for 7-10 days. This prevents confusing signal overlap and helps you identify what actually improved recall.

Exam prep vs long-term mastery: two valid FSRS modes

Exam prep mode emphasizes recall reliability over a fixed horizon (for example 6-12 weeks), often with moderate-to-high retention targets and strict daily completion. In this mode, trimming low-yield cards is usually more effective than endlessly increasing review time.

Long-term mastery mode optimizes sustainability across months or years. You may run a slightly lower target with higher card quality and cleaner tagging. The objective shifts from short-term score maximization to durable knowledge with manageable workload.

Both modes are valid. Choose the one that matches your timeline, then keep rules stable long enough to evaluate outcomes from real review data.

Weekly FSRS operations checklist

FSRS performs best when you run it like an operating system, not a one-time setup. A short weekly review loop keeps your card pool healthy and prevents small quality defects from compounding into daily review fatigue.

1. 1
   Tag your misses. When cards fail, tag the cause (ambiguous prompt, overloaded card, weak context, or pure memory gap).
2. 2
   Fix highest-impact tags first. Work from the biggest lapse clusters rather than random card edits.
3. 3
   Keep card templates consistent. Similar card types should follow the same answer format so rating behavior stays reliable.
4. 4
   Record one weekly note. Write what changed, what improved, and what to test next week.

Worked scenarios: how FSRS settings change by goal

Learners often ask for one universal FSRS setup, but the optimal profile depends on constraints: timeline, error tolerance, and available daily minutes. The same settings that work for long-term language maintenance can feel overwhelming for short exam windows, and the opposite is also true.

Use the table below as a decision shortcut. Start with the closest scenario, keep settings stable for at least 10 days, then tune one variable at a time. This avoids false conclusions caused by changing multiple factors simultaneously.

The main idea is operational fit: a slightly lower retention target that you can sustain daily usually outperforms an aggressive target you abandon after two weeks.

Common failure patterns and how to recover fast

When learners say FSRS "is not working," the root cause is usually process quality, not the scheduler itself. Three patterns appear repeatedly: uncontrolled new-card intake, unclear card prompts, and inconsistent rating behavior after recall. All three produce noisy signals that make interval planning less reliable.

1. 1
   Backlog spiral: new cards keep growing while daily completion drops. Recovery: freeze new cards for 3-5 days and clear overdue reviews before resuming intake.
2. 2
   Ambiguity spiral: prompts ask multiple things, causing frequent "almost knew it" outcomes. Recovery: rewrite top lapse cards into one question, one expected answer, one clear context.
3. 3
   Rating drift: ratings become generous during busy weeks, then lapses spike later. Recovery: run 7 days of stricter self-rating and re-check interval behavior.

Fast recovery comes from reducing system noise first. Once your data quality improves, small retention adjustments become meaningful and easier to interpret.

Implementation checklist you can reuse monthly

Treat this as a recurring maintenance loop. The checklist helps keep review quality high as your card library grows and study goals change over time.

1. 1
   Set one explicit goal metric for the month (for example, keep completion above 85%).
2. 2
   Audit your top failing tags and rewrite the weakest 20 cards first.
3. 3
   Check whether average session time is rising faster than card volume.
4. 4
   Remove or suspend low-yield cards that repeatedly consume time without retention gain.
5. 5
   Document one process change and re-evaluate after 2 weeks of stable usage.

This monthly loop keeps FSRS aligned with your real workload. The best learners do not only review cards; they also maintain the quality of the review system itself.

FAQ

Related resources