Why this works

Behind every decision in lern.app there is a finding from learning research. Point by point, here is what the research says. And what the app makes of it.

The dots after each title say how certain the research is: ⬤⬤⬤ checked and confirmed many times over, ⬤⬤◯ well evidenced, but with limits.

Recalling does more than reading again

⬤⬤⬤

The app rarely puts texts in front of you to read. It asks questions. Answering them is not the check at the end — it is the learning itself. No other finding in learning research is as well evidenced.

Evidence: Roediger & Karpicke (2006), Psychological Science 17(3), 249–255, Retrieval-Based Learning: A Decade of Progress (ERIC ED599273)

Comparing makes knowledge usable elsewhere

⬤⬤⬤

Hold two facts side by side and the pattern behind them shows — and it can then be applied to a new case. A question about a single fact does not do that. Which is why there are tasks that connect two or three facts.

Evidence: Gentner, Loewenstein & Thompson (2003), Journal of Educational Psychology 95(2), 393–408, Gentner, Loewenstein & Thompson, Analogical Encoding: Facilitating Knowledge Transfer and Integration

Practising several times with gaps beats one long session

⬤⬤⬤

The same effort achieves more when spread across several days. How long a question rests depends on how well it already sticks.

Evidence: Cepeda, Pashler, Vul, Wixted & Rohrer (2006), Psychological Bulletin 132(3), 354–380

Mixing topics beats one topic at a time

⬤⬤⬤

The app mixes the topics instead of working through them one after another. It feels harder. More of it stays.

Evidence: Rohrer & Taylor (2007), Instructional Science 35(6), 481–498

Writing it yourself does more than clicking

⬤⬤⬤

Wherever it works, your participant writes the answer instead of picking it. Of six questions on one fact, four are of that kind.

Evidence: Slamecka & Graf (1978), Journal of Experimental Psychology: Human Learning and Memory 4(6), 592–604

The wrong answers teach something too

⬤⬤⬤

Reading why the other answers are wrong teaches you along the way. That is why every option carries a short reason. It appears as soon as the question is answered.

Evidence: Little, Bjork, Bjork & Angello (2012), Psychological Science 23(11), 1337–1344, Little & Bjork (2015), Memory & Cognition, Retrieval-Based Learning: A Decade of Progress (ERIC ED599273)

An explanation follows every answer

⬤⬤⬤

With multiple choice, a wrong answer sometimes sticks in the mind. An explanation right afterwards makes that damage smaller and the benefit bigger. It does not remove it entirely.

Evidence: Butler & Roediger (2008), Memory & Cognition 36(3), 604–616, Beyond right or wrong: More effective feedback for formative multiple-choice tests (PMC7550480)

Feedback talks about the material, not about the person

⬤⬤◯

The app says “percentages: 48 % correct”. It does not say “you are weak at maths”. In the largest study on this, more than a third of all feedback made performance worse — and the more it spoke about the person, the more often that happened.

Evidence: Kluger & DeNisi (1996), Psychological Bulletin 119(2), 254–284, Weiner (1985), Psychological Review 92(4), 548–573

We reorder — we never leave anything out

⬤⬤◯

The app changes when and how often something comes up. It leaves nothing out. Easy questions appear in the exam too: whoever never sees them again forgets them.

Evidence: Bloom (1968), Learning for Mastery, Evaluation Comment 1(2), Kulik, Kulik & Bangert-Drowns (1990), Review of Educational Research 60(2), 265–299, Slavin (1987), Mastery learning reconsidered, Review of Educational Research 57(2), 175–213

We recommend a path, we do not prescribe it

⬤⬤⬤

Help that serves beginners holds advanced learners back. So anyone who already knows something may move straight on.

Evidence: Sweller & Cooper (1985), Cognition and Instruction 2(1), 59–89, van Merriënboer (1990), Journal of Educational Computing Research 6(3), 265 ff., Renkl & Atkinson (2003), Educational Psychologist 38(1), 15–22, Atkinson, Derry, Renkl & Wortham (2000), Review of Educational Research 70(2), 181–214, Kalyuga, Ayres, Chandler & Sweller (2003), Educational Psychologist 38(1), 23–31

Why Anki and similar flashcard programs do not fit here

Programs like Anki work out, for every person and every card, when they will forget it. The best gap is not a fixed number. It depends on how long you want to keep something. With Anki, everyone has to find that setting themselves — on cards they wrote themselves. ⬤⬤⬤

Anyone who wants to optimise themselves over years is well served by that. For exam preparation in a seminar it does not fit, for three reasons: the material comes from you, not from the learner. It is about a fixed exam date, not about keeping something forever. And you as the provider would see none of it.

lern.app therefore does three things differently. Instead of rigid gaps, a pool decides that weighs several signals against each other. What is learnt is the fact, not the card. And the right gaps emerge from the answers of all of a provider's learners, not from each person separately. The work every Anki user does alone is shared here by all participants — and you see the result.

Sources

These are all the works we rely on. How certain a finding is, and where it sits in the code, we maintain internally.

Recalling does more than reading again

Roediger & Karpicke (2006), Psychological Science 17(3), 249–255
Retrieval-Based Learning: A Decade of Progress (ERIC ED599273)

Comparing makes knowledge usable elsewhere