← All posts

Best spaced repetition settings for USMLE Step 1 (FSRS, explained)

If you're preparing for Step 1, your scheduling algorithm decides how many cards you see every day and how much you actually remember on test day. Most med students inherit their settings from a Reddit thread and never touch them again. This guide explains what the settings actually do, and how to pick them deliberately.

SM-2 vs FSRS: what changed

Anki's classic scheduler descends from SuperMemo's SM-2 algorithm (1987). It multiplies your interval by an "ease factor" that moves up and down as you answer. It works, but it has well-known failure modes — most famously "ease hell," where a few lapses push a card's ease so low that you see it constantly without remembering it any better.

FSRS (Free Spaced Repetition Scheduler) is a modern replacement built on a memory model with three components per card: difficulty, stability (how long the memory lasts), and retrievability (the probability you can recall it right now). Instead of blindly multiplying intervals, it predicts the exact day your recall probability drops to a target you choose, and schedules the review there. In public benchmarks against hundreds of millions of real reviews, FSRS predicts recall meaningfully better than SM-2-family schedulers — which translates into fewer reviews for the same retention.

Gyrall runs FSRS natively, so the settings below map directly; if you're on Anki, enable FSRS in the deck options (Anki 23.10+).

The one setting that matters: desired retention

FSRS has effectively one user-facing dial: desired retention — the recall probability at which a card comes back.

  • 0.90 (default) — the right choice for most of your Step 1 timeline. Solid retention at a sustainable review load.
  • 0.85 — use this if your backlog is crushing you. Review load drops substantially and you still retain most material; the marginal reviews between 85% and 90% are the most expensive ones.
  • 0.93–0.95 — dedicated period only. Review load grows steeply as you push retention up (the curve is non-linear: going from 0.90 to 0.95 costs far more than going from 0.85 to 0.90). Turn it up 6–8 weeks before your exam, not before.

Practical Step 1 configuration

  1. Preclinical years: desired retention 0.90, no daily review cap. A cap silently converts into a backlog — the reviews don't disappear, they queue.
  2. New cards per day: whatever keeps total daily time under your budget. Every new card costs several future reviews; if reviews are eating your day, the fix is fewer new cards, not a lower retention target you'll regret in dedicated.
  3. Dedicated period: raise desired retention to 0.93–0.95, stop adding new cards in the last 2–3 weeks, and let the scheduler concentrate on consolidation.
  4. Don't manually "reset" struggling cards. FSRS already accounts for lapses in its difficulty estimate. Deleting and re-adding cards destroys the history the model learns from.

Retention isn't the goal — passing is

A scheduler optimizes memory of your cards. It can't fix cards that test recognition instead of recall, or a deck that doesn't match your exam blueprint. Spend your settings-tweaking energy once (10 minutes, the numbers above), then put the rest into better cards: single-fact, cloze-style, source-linked so you can verify a card against the original material when it feels off.

That last part is the core of how Gyrall works — every AI-generated card links back to the exact passage in your lecture PDF it came from, so you never memorize a hallucination. Try it free, or read more about how we generate cards from PDFs.