Open Anki, add twenty cards, and it feels solved: the app promises to show each one exactly when you are about to forget it. That promise is real, but it is a promise made by an algorithm, and algorithms have assumptions. SuperMemo made the same promise thirty years earlier with a different model and a very different philosophy. Understanding what sits inside both tools — not the buttons, the math — is the difference between a system that compounds your knowledge and one that quietly feeds you the same cards forever.
This matters because the spaced repetition algorithm is the only part of the process you cannot eyeball. You can feel whether a card is badly written. You cannot feel whether your interval is being computed from a stability estimate that no longer fits your memory. So let's open the box.
What the algorithm is actually computing
Every serious spaced repetition system tracks two numbers per card. Stability is how long your memory survives before recall probability drops to a reference level — roughly how many days until you are at 90% chance of remembering. Difficulty (called the ease factor in older systems) says how fast that stability grows each time you review successfully. A hard card gains stability slowly; an easy card gains it fast.
The classic SM-2 algorithm, described by Piotr Woźniak in the late 1980s and still the backbone of Anki's default scheduler, is a set of rules. Each card starts with an ease factor of 2.5. Answer well and the next interval is the previous interval multiplied by that ease. Answer badly and the interval collapses, and the ease drops by a fixed penalty. It is deterministic arithmetic with a handful of constants — which is why it is easy to implement and why it has survived.
The newer FSRS algorithm (Free Spaced Repetition Scheduler) replaces the fixed constants with a model fitted to your own review history. Instead of "ease drops by 0.2", it estimates a forgetting curve per card from your actual answers and optimizes its parameters against your log. The practical consequence is what "desired retention" means: in FSRS you set a target — say 0.9 — and the scheduler computes intervals that should hit it. In SM-2 there is no such dial; retention is an outcome, not an input.
A walkthrough: setting up a deck that survives month three
Imagine you are learning medical terminology — 800 terms, six months, about 20 minutes a day. Here is the order a practitioner works in, and what breaks if you reorder it.
- Decide the card unit before writing anything. One fact per card. "Term → definition" and "definition → term" are two cards, not one. Mixing them doubles your review load and hides which direction is failing.
- Write 30 cards, not 800. Review them for a week. Cards you keep failing are badly written, not hard — they are ambiguous, or they test recognition when you need recall.
- Pick a scheduler and leave it alone. If you use Anki with FSRS, let it optimize only after you have a few thousand reviews; before that the parameters are guesswork and the default SM-2 behavior is safer.
- Set a daily new-card limit you can sustain on your worst day, not your best. Twenty new cards a day becomes roughly 150–200 reviews a day within a few weeks — the arithmetic of compounding, not a warning.
Do it in the wrong order — import a shared deck of 800 cards first — and you inherit someone else's card design, someone else's context, and a review queue that grows faster than your willingness to sit through it. Shared decks are the single most common way beginners sabotage themselves.
Where the scheme itself breaks
These are not beginner mistakes. They are conditions under which spaced repetition software stops delivering what it promises.
- Your material is not atomic. If a card tests a paragraph, the scheduler cannot tell which part you forgot, so stability estimates become noise. Sign: you answer "again" on cards you mostly know. Fix: split, but only the part you actually missed.
- Your review history is contaminated. Switching schedulers, resetting cards, or reviewing on autopilot corrupts the data FSRS optimizes against. Sign: intervals that feel absurdly short or long after a migration. Fix: keep a clean break — new cards under the new scheduler, old cards left on the old one.
- The tool's platform stops fitting your life. SuperMemo is Windows software with a long legacy; Anki's sync and add-on ecosystem have their own costs and plugin sprawl. Sign: you skip reviews because opening the app is friction. Fix: mobile-first review, fewer add-ons.
- You are reviewing passively. Recognizing an answer is not recall. The scheduler records a success either way. Sign: strong in-app performance, weak real-world retrieval. Fix: production-style cards and free recall before flipping.
Anki vs SuperMemo: the honest trade-off
| Anki | SuperMemo | |
|---|---|---|
| Scheduling | SM-2 by default, FSRS optional | Its own incremental algorithms |
| Philosophy | Deck of cards you author | Incremental reading — build knowledge from sources |
| Cost | Free on desktop; sync and mobile have costs | Paid, tiered |
| Platform | Cross-platform | Windows-centric |
| Setup burden | Low to start, high to tune | Steep learning curve |
The pricing gap and the platform gap are real, but the deeper difference is philosophical. Anki assumes you already know what to ask. SuperMemo's incremental reading assumes you are processing articles and books and want the tool to extract and schedule fragments — powerful, and demanding.
Which path fits you
Self-study with documentation and forums is free and works if you are disciplined about card quality. A structured course costs time and money but gives you feedback on decisions you cannot check alone — whether your cards are atomic, whether your intervals make sense, whether your "study session" is actually recall practice.
If you want a single decision rule: choose Anki if your material is discrete facts and you want cross-platform review; choose SuperMemo if your material is long-form text and you will invest in the learning curve. If neither fits, the problem may be your cards, not your tool.
This is exactly where Learning Skills & Memory Techniques fits: it walks through the SM-2 and FSRS calculations with examples, covers active recall and note-taking systems like Cornell and Zettelkasten, and gives you feedback on the decisions — card design, scheduler choice, review habits — that no forum thread can check for you.
You do not need to master every algorithm to learn well. You need to know what your tool is measuring, write cards worth reviewing, and notice when the system is feeding you noise instead of knowledge. Start with thirty good cards and an honest review session — the rest is arithmetic.