Why recall beats reading

There is a way of studying that feels wonderful and barely works: reading the thing again. You open the Gītā to a verse you’re learning, read it through, and it flows — every word is right there, familiar, easy. You close the book convinced you know it. Three days later you try to say it from nothing and discover you knew how to recognise it, which is a completely different skill, and not the one you needed.

Shlokas is built around the opposite, less pleasant motion — and around a fifty-year-old pile of evidence that the unpleasant motion is the one that lasts.

The idea: retrieval is the learning event

Most people picture memory as a bucket — study pours information in, testing checks the level. The research says that picture is backwards. The act of pulling a memory out — effortfully, from a near-empty mind — is not a measurement of learning; it is a learning event, and a stronger one than any amount of pouring. This is the testing effect, also called retrieval practice, and it is one of the sturdiest findings in cognitive psychology.

The classic demonstration is Roediger and Karpicke’s 2006 study, Test-Enhanced Learning. Students learned prose passages two ways: one group re-read repeatedly, the other read once and then practised recalling. On a test given five minutes later, the re-readers won — cramming flatters the short term. But on a test a week later, the order flipped and stayed flipped: the retrieval group remembered markedly more, roughly half the material against the re-readers’ third. Read again and you feel ready and fade fast; retrieve and you feel shaky and stay. Two years later Karpicke and Roediger drove the point home in Science under a title that is itself the thesis — The Critical Importance of Retrieval for Learning.

The problem it solves: the fluency trap

So why doesn’t everyone study by self-testing? Because re-reading feels better, and the feeling is a lie. Fluency — the ease of a familiar text sliding past your eyes — gets misread by the brain as mastery. Psychologists call retrieval one of the “desirable difficulties”: the methods that work feel like struggling, and the methods that feel smooth mostly don’t stick. Left to our own judgement we reliably choose the comfortable one and mistake comfort for progress.

For a Sanskrit verse the trap is sharper than usual, because the gap between recognising and producing is enormous. A śloka has no native meaning to fall back on, the word order is exact, the sandhi runs the words together, and the goal is to say the whole thing aloud, in sequence, with nothing in front of you. Re-reading trains recognition. The task is production. You can re-read Bhagavad-gītā 2.13 a hundred times, recognise it instantly every time, and still stand up in front of your class and produce silence. The only practice that touches production is production.

How we use it: an app with no re-read button

Knowing all this, the design follows almost on its own — Shlokas simply refuses to offer the comfortable path. There is no “read it again” mode. The learning flow is three stages, and two of the three are pure retrieval.

graph LR
    M["Memorize — exposure"] --> R["Recall — retrieval practice"] --> V["Review — spaced retrieval"]
    V -.->|forgot| R

Memorize is the only stage that hands you the text, and only briefly — the minimum exposure to make a first attempt possible. Then the door closes.

Recall is retrieval practice made literal. The verse is broken into cards that test it from every direction — verse-number to text, text to translation, translation back to number, and the reverse of each — so you are never recognising a fixed shape, you are generating the answer from a prompt that keeps changing. You commit before you check. The shaky, effortful pull the research is about is the entire interaction.

Review is that same retrieval, now spaced out over days by the scheduler we described in Cheating the forgetting curve — each successful recall stretches the gap to the next one. Crucially, the review is always a recall, never a re-read. And the grading is a confession, not a quiz score: swipe up for got it, down for gone. A forgot doesn’t just lower a number — it drops the verse back toward the start, because the system trusts the retrieval attempt as the real measure of what you know, exactly as the testing effect says it should.

Compared to the app’s purpose: a memory app is not a library

This is the cleanest way to see what Shlokas is for, and it comes out by contrast with our other work. Our flagship, Lectorium, is a library — thousands of lectures, searchable transcripts, an assistant that finds you the exact passage. Its entire job is to make recognition and lookup effortless: you should never have to remember where Prabhupāda said a thing, you should just find it. A library optimises for the verse being out there, instantly reachable.

Shlokas optimises for the opposite — the verse being in here, with no device required. Its job is not to help you find the śloka; it is to make you the place the śloka lives. And the two goals demand opposite interactions. A library that made you struggle to retrieve would be a bad library. A memory app that let you look things up would not be a memory app at all — it would just be a slower library training you to depend on it. The missing re-read button isn’t an oversight; it is the line between the two.

There’s an older version of this same distinction. The tradition names a progression — śravaṇam, kīrtanam, smaraṇam: hearing, then repeating aloud, then remembering — and it is, almost exactly, memorize → recall → review. You hear the verse, you say it back, and only through the saying-back does it become something you carry. The practice always understood that you don’t absorb a verse by reading it more times; you absorb it by producing it until it is yours. Shlokas is that intuition with a forgetting curve attached and a scheduler doing the arithmetic — but the core of it is old, and it was right.