What a Wrong Answer Is Worth

The finding reframes everything that follows. Feedback is not fertiliser — spread it and things grow. It is an intervention, with a dose, an aim, and a risk profile. Ruth Butler's experiments found that comments produced better subsequent performance than grades — and that adding a grade to comments destroyed the comments' advantage, as if the number absorbed all the attention 6. John Hattie and Helen Timperley's synthesis organises effective feedback around three questions the learner is implicitly asking: where am I going?, how am I going?, and where to next? 4. Valerie Shute's review compresses decades into guidance: feedback should be specific to the answer, timed to the learner, and — above all — should say something about the task rather than the self 5.

An error, in other words, is valuable only inside a system designed to receive it well. The system is the rest of this dispatch.

Failing first

Now the stranger half of the literature. Nate Kornell, Matt Hays and Robert Bjork asked students to take a pretest on material they had not been taught — pairs from distant countries, wild guesses required — and then taught the correct pairings. On a final test the students who had failed the pretest remembered more than students who had only studied 7. Lindsey Richland, Nate Kornell and Lisa Kao replicated the effect with realistic classroom materials: unsuccessful retrieval attempts enhanced learning even when the attempts were wrong 8. The wrong answers were not merely survived. They helped.

Why? The leading explanation is that a failed retrieval is a survey of your own ignorance: it activates related knowledge, exposes exactly what you do not know, and makes the eventual correction land on prepared ground. The effect has a cost, drawn honestly on the left of the figure — the pretested group starts behind, having studied confused material first — and a benefit paid out over time, as a heavier tail. Anyone who has ever guessed at an exam answer, been corrected, and found they could never again forget the correction has felt this mechanism personally.

Manu Kapur carried the logic from single items to entire lessons. In his productive failure studies, students explore novel problems in small groups before any canonical instruction — flail, invent methods that mostly fail, surface the real constraints — and are then taught the standard method. Compared against direct instruction followed by practice, with the same total time, the exploration classes produce worse solutions during the lesson and better learning on delayed tests, including deeper structure and transfer 910. The figure above draws both schedules against their exact sixty-minute budgets. The classic order looks better the whole way through and ends up behind — the assistance dilemma of the hint dispatch, seen from the syllabus level.

None of this licenses chaos. Productive failure works because the exploration is structured — around problems whose core idea the curriculum will consolidate — and because the instruction afterwards connects to what students invented. Failure earns its keep when someone collects it.

The timing of the answer

When the correction comes also matters, and here instinct needs correcting twice.

First: feedback reliably helps, but its effect size varies enormously across studies, and Bangert-Drowns and colleagues traced much of the variance to what kind of feedback — right/wrong alone barely moves learning; feedback that explains does far more 2. Second, and less intuitively: immediacy is not always king. Shute's review finds immediate feedback favours acquisition during initial learning, while delayed feedback tends to favour retention and transfer 5. Butler and Roediger found the delayed advantage directly in multiple-choice studies: students given the correct answer after a delay outperformed those given it instantly on later tests 13. The figure above draws the crossing. The mechanism is pleasingly economical — by the time the delayed answer arrives, the learner's original answer has itself become a retrieval attempt, with all the benefits retrieval carries 15. The wait is practice wearing a delay's clothing.

There is a practical tension here and no free lunch: the tutor who answers instantly teaches the step today and forfeits some of next month. Our resolution is to let the surface answer quickly and schedule the repair slowly — which is where the ledger comes in.

The ledger

Collect the strands and a machine assembles itself. When an answer is wrong:

1. Verdict — honest, immediate, unemotional: this is not yet correct.
2. Diagnosis — what kind of wrong? A slip under time pressure, a misread stem, a missing concept, a half-learned procedure applied past its edge? Ohlsson's analysis of performance errors separates these classes precisely because their remedies differ 12; Metcalfe's review of learning from errors makes the same point developmentally — errors predict learning exactly to the extent that they are diagnosed and corrected rather than repeated 11.
3. Schedule — the topic re-enters the queue, at an interval that widens as repairs succeed: days, then a week, then two. The arithmetic is the same expanding ladder the spaced-repetition dispatch derived from first principles — a repair attempted while the error is fresh is cheap and shallowly earned; a repair attempted after a real gap is stronger evidence and stronger practice.
4. Closure — the entry leaves the ledger when it can be answered cleanly after a delay, not when the sting fades. Janet Metcalfe's summary of the error literature lands on the same criterion: it is not making the error nor disliking it that predicts learning; it is successfully generating the correct response afterward 11.

The hero plate draws the loop. Notice what it refuses: a wrong answer never produces only a mark; a repaired answer never closes on the same day. The ledger is the bridge between assessment and memory — Paul Black and Dylan Wiliam's long argument that formative assessment is the engine room of learning, made mechanical 1. And it is why Scott Dunlosky and colleagues' landmark review rated practice testing and distributed practice as the two highest-utility techniques in their survey of ten 16: the ledger is both at once, aimed at exactly the items that need them.

Where the model thumbs the scale

Three honest limits.

Errors entrench as easily as they teach. A misconception practised is a misconception strengthened; the pretesting effect requires that a correction eventually arrives and is attended to. An uncorrected wrong answer in the wild — a friend's confident explanation, an answer key with a typo — teaches backwards. The ledger's diagnosis step is not bureaucracy; it is the difference between the two outcomes.

Motivation is not in the model. The same feedback event can read as information or indictment depending on the learner's history and the framing around it; Butler's grades-versus-comments result is the sharpest edge of this 6. A ledger presented as surveillance becomes another thing to game or fear. Presented as a workshop list — here is what is not yet yours, here is when we ask again — it is a tool. The words around the machinery carry weight the machinery cannot compute.

Transfer is not automatic. Repairing item after item builds skill on those items; whether the repaired understanding generalises to unfamiliar framings is a separate question that the transfer meta-analyses treat cautiously 14. The ledger reduces forgetting; it cannot by itself manufacture insight. That remains the tutor's work — and, where the tutor is a machine, the boundary of its honesty.

What we read from it

EuraStudy runs a correction ledger — der Fehlerheft — built exactly along these lines. Every incorrect attempt enters the queue with its diagnosis; the queue schedules repairs at widening intervals; entries close on clean delayed answers, never on elapsed time. The reading we take from this literature is threefold. Feedback is medicine — dose and aim matter, and over a third of it harms, so our corrections arrive attached to reasons rather than numbers. Failure is tuition — the wrong answer you generate yourself buys more than the right answer you read, provided the system collects on it. And the error is the syllabus — the most personal, most current map of what to teach next is not the chapter list anyone wrote in August; it is the ledger of what you got wrong this week, sorted by when you'll be asked again.