The receipts.

Spaced repetition works

The spacing effect (the same hours of study buy more memory when spread out than when crammed) is one of the oldest and most replicated findings in experimental psychology. It was first documented in 1885 and has survived every methodological fashion since. The definitive review is Cepeda and colleagues (2006), which synthesized 839 assessments of distributed practice across 317 experiments: spaced practice reliably beats massed practice.

Dunlosky and colleagues (2013) surveyed ten popular learning techniques for a major review and rated only two as high utility: distributed practice and practice testing. Verbamor is, without exaggeration, those two techniques in a trench coat.

One catch matters for software: the best gap between reviews isn't fixed. It grows with how long you want to remember (Cepeda et al., 2008). Which is why a scheduler, not a calendar, should be running this.

• Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in verbal recall tasks: A review and quantitative synthesis.

  Psychological Bulletin, 132(3), 354–380 · meta-analysis, 317 experiments

• Cepeda, N. J., Vul, E., Rohrer, D., Wixted, J. T., & Pashler, H. (2008). Spacing effects in learning: A temporal ridgeline of optimal retention.

  Psychological Science, 19(11), 1095–1102

• Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Improving students’ learning with effective learning techniques.

  Psychological Science in the Public Interest, 14(1), 4–58 · review of 10 techniques

Forgetting is the default state

Hermann Ebbinghaus taught himself thousands of nonsense syllables, tested his own recall at intervals, and plotted the result: memory falls steeply within hours and days, then flattens. That was 1885. In 2015, Murre and Dros re-ran the experiment under modern conditions (one heroic subject, 70 days of relearning sessions) and got the same curve.

Two honest footnotes. Languages are not nonsense syllables: meaningful, connected material decays more slowly, but along the same shape, and the shape is what the scheduler models. Your own retention number, computed from your actual reviews, is on your home screen.

• Ebbinghaus, H. (1885/1913). Memory: A Contribution to Experimental Psychology.

  Full text, Classics in the History of Psychology · the original forgetting curve

• Murre, J. M. J., & Dros, J. (2015). Replication and analysis of Ebbinghaus’ forgetting curve.

  PLOS ONE, 10(7): e0120644 · open access replication

Recalling beats re-reading

The testing effect: actively retrieving something from memory strengthens it far more than looking at it again. Roediger and Karpicke (2006) showed students who practiced recall held onto material a week later much better than students who spent the same time re-studying. Karpicke and Roediger (2008), published in Science, made the point with foreign-language vocabulary. Swahili word pairs, about as close to your lesson deck as lab work gets: once a word could be recalled, continued retrieval practice determined whether it survived the week; extra studying added almost nothing.

A 2017 meta-analysis across 272 independent effects confirmed the advantage of practice testing over restudy is robust and sizable. Fill-in-the-blank prompts collect a second dividend: information you generate yourself is remembered better than information you're handed. That's the generation effect (Slamecka & Graf, 1978).

• Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention.

  Psychological Science, 17(3), 249–255

• Karpicke, J. D., & Roediger, H. L. (2008). The critical importance of retrieval for learning.

  Science, 319(5865), 966–968 · foreign-language vocabulary

• Adesope, O. O., Trevisan, D. A., & Sundararajan, N. (2017). Rethinking the use of tests: A meta-analysis of practice testing.

  Review of Educational Research, 87(3), 659–701 · meta-analysis, 272 effects

• Slamecka, N. J., & Graf, P. (1978). The generation effect: Delineation of a phenomenon.

  Journal of Experimental Psychology: Human Learning and Memory, 4(6), 592–604

The scheduler, and its 21 tweaks

FSRS (the Free Spaced Repetition Scheduler) is an open-source algorithm that models every card in your deck with three quantities: difficulty, stability, and retrievability. It descends from a model published at KDD 2022, trained on enormous review logs from real language learners, and it is developed and stress-tested entirely in the open.

The public benchmark pits it against alternatives on roughly 1.7 billion real flashcard reviews from about 20,000 learners. FSRS predicts recall more accurately than SM-2 (the 1987 algorithm most flashcard apps still ship), and accurate prediction is the whole game: it’s what lets a schedule hold your retention target without over-reviewing to be safe.

The current generation, FSRS-6, exposes 21 optimizable parameters, the “21 tweaks.” They start at defaults learned from the crowd; Verbamor then re-fits them to your own review history, which the benchmark shows improves accuracy further. It’s the same open-source FSRS either way. The difference is who does the tuning.

• Ye, J., Su, J., & Cao, Y. (2022). A stochastic shortest path algorithm for optimizing spaced repetition scheduling.

  Proceedings of KDD ’22 · the paper behind FSRS

• Open Spaced Repetition (ongoing). SRS benchmark.

  Public benchmark · ~1.7B reviews, ~20k learners · FSRS vs. SM-2 and others

• Anki Manual: FSRS.

  Reference implementation docs · parameters & per-user optimization

Why start with 650 words

Word frequency is brutally top-heavy. Corpus studies of everyday conversation find the most frequent ~1,000 word families account for roughly 85% of everything said, and 2,000 families get you to just under 95% (Adolphs & Schmitt, 2003; Nation, 2006). Studies of listening comprehension put comfortable understanding at knowing 2,000–3,000 families (van Zeeland & Schmitt, 2013). Every word you learn from the top of that curve does wildly more work than one from the middle.

• Adolphs, S., & Schmitt, N. (2003). Lexical coverage of spoken discourse.

  Applied Linguistics, 24(4), 425–438 · CANCODE spoken corpus

• Nation, I. S. P. (2006). How large a vocabulary is needed for reading and listening?

  Canadian Modern Language Review, 63(1), 59–82

• van Zeeland, H., & Schmitt, N. (2013). Lexical coverage in L1 and L2 listening comprehension.

  Applied Linguistics, 34(4), 457–479

Why every card gets an image and audio

Dual-coding theory (Paivio, 1991): information stored both verbally and visually has two retrieval routes instead of one, and pictures are remembered exceptionally well. The same conclusion holds across decades of multimedia-learning experiments: people learn better from words plus pictures than from words alone (Mayer, 2009).

Tested directly on foreign-language vocabulary, pictures beat plain translations (Carpenter & Olson, 2012), with one wry catch right in the paper’s title: pictures make learners overconfident. The antidote to overconfidence is a scheduler that tests actual recall and believes your results over your feelings. Conveniently, that is the entire rest of the product. Audio with a native accent ties each written form to the sound you’ll actually hear in your lessons, so the card you drill is the word as spoken, not just as spelled.

• Paivio, A. (1991). Dual coding theory: Retrospect and current status.

  Canadian Journal of Psychology, 45(3), 255–287

• Mayer, R. E. (2009). Multimedia Learning (2nd ed.).

  Cambridge University Press · the multimedia principle

• Carpenter, S. K., & Olson, K. M. (2012). Are pictures good for learning new vocabulary in a foreign language? Only if you think they are not.

  Journal of Experimental Psychology: Learning, Memory, and Cognition, 38(1), 92–101