Mistakes change learning when an incorrect answer becomes usable information about what should happen next. Searches for learning from mistakes, how mistakes help you learn, error-based learning, corrective feedback, common student mistakes and how to stop repeating mistakes all point to the same practical problem: a wrong answer does not teach by itself. Learning begins when the learner detects the mismatch, explains its cause, repairs the underlying rule or process and then succeeds on a fresh attempt.
For students, parents and teachers, the important distinction is between correcting an answer and correcting the system that produced the answer. A vocabulary error may come from meaning, retrieval, collocation or context. A mathematics error may come from representation, method selection, transformation or checking. A comprehension error may come from evidence, inference or answer scope. The same red cross can therefore require very different teaching.
This guide explains why mistakes can improve learning, why they sometimes do not, how feedback and error correction work together, how to diagnose repeated mistakes and how to turn errors into better performance in vocabulary, reading, writing, mathematics, science and examinations. The central proposition is simple: an error becomes educationally valuable only when it changes a later decision. A recent synthesis of research on errors and failure in educational contexts likewise emphasises context, emotion, feedback, learner processing and instructional support. Research overview.
Your 50-second route
Do not begin with “be more careful”. Find the first meaningful divergence. Ask whether the learner lacked knowledge, failed to retrieve it, chose the wrong method, misread the task, executed a valid method inaccurately or failed to check. Correct the governing principle. Ask the learner to explain it without copying. Give a changed example. Return later. If the same mechanism reappears, change the repair rather than simply increasing the number of questions.
Expandable contents — diagnosis, repair, subjects and long-term transfer
1. What a mistake tells you · 2. Mistakes do not teach automatically · 3. Error, failure and misconception · 4. The first meaningful divergence · 5. Knowledge gaps · 6. Retrieval gaps · 7. Selection errors · 8. Execution errors · 9. Interpretation errors · 10. Language errors · 11. Corrective feedback · 12. Confidence and familiarity · 13. Productive struggle · 14. Error climate · 15. Emotion after errors · 16. Motivation and identity · 17. Wrong worked examples · 18. Vocabulary repair · 19. Reading repair · 20. Writing repair · 21. Mathematics repair · 22. Science repair · 23. Examination error taxonomy · 24. Careless mistakes · 25. Repeated mistakes · 26. Primary learners · 27. Secondary learners · 28. Parents · 29. Teachers · 30. Self-correction · 31. Peer correction · 32. Spacing corrections · 33. Transfer after correction · 34. Useful error logs · 35. Ten-minute repair protocol · 36. When mistakes are not useful · 37. AI and error detection · 38. Seven-day experiment · 39. Thirty-day review · 40. World-return test
Continue through Why Feedback Changes Learning, Why Practice Changes Learning, Why Spaced Repetition Changes Memory, the Vocabulary Learning Hub and the How X Works library.
1. What a mistake tells you
The starting principle is that an error is evidence of a mismatch between a target and a response, not a complete diagnosis. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
2. Mistakes do not teach automatically
The starting principle is that an incorrect answer becomes useful only when it is noticed, interpreted, corrected and followed by another attempt. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
3. Error, failure and misconception
The starting principle is that local errors, unsuccessful outcomes and stable misconceptions require different levels of response. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
4. The first meaningful divergence
The starting principle is that the earliest point where valid reasoning became invalid often explains many later wrong steps. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
5. Knowledge gaps
The starting principle is that missing prerequisite knowledge requires teaching before repetition can help. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
6. Retrieval gaps
The starting principle is that understanding may be present even when a label, fact or relationship cannot be independently recovered. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
7. Selection errors
The starting principle is that knowing several methods differs from recognising which method fits the current task. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
8. Execution errors
The starting principle is that a valid plan can fail during calculation, transcription, grammar or multi-step execution. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
9. Interpretation errors
The starting principle is that content knowledge can be present while the learner answers a different question from the one asked. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
10. Language errors
The starting principle is that language can obscure otherwise sound subject understanding. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
11. Corrective feedback
The starting principle is that feedback should identify the consequential gap, explain the governing principle and create a next action. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
12. Confidence and familiarity
The starting principle is that feeling certain is not the same as producing an accurate independent answer. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
13. Productive struggle
The starting principle is that difficulty is useful only while the learner has enough knowledge and support to continue meaningful thinking. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
14. Error climate
The starting principle is that learners expose more useful reasoning when errors can be examined without humiliation and standards remain clear. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
15. Emotion after errors
The starting principle is that frustration and embarrassment can determine whether a learner studies an error or avoids it. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
16. Motivation and identity
The starting principle is that correction should create a controllable path forward rather than a fixed judgement about ability. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
17. Wrong worked examples
The starting principle is that erroneous examples can teach when learners detect, explain and repair the faulty reasoning. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
18. Vocabulary repair
The starting principle is that word errors can involve meaning, nuance, morphology, collocation, grammar, register or retrieval. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
19. Reading repair
The starting principle is that comprehension errors can arise from evidence selection, reference, vocabulary, inference or answer scope. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
20. Writing repair
The starting principle is that writing contains many dimensions, so revision must prioritise errors instead of correcting everything at once. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
21. Mathematics repair
The starting principle is that mathematics errors differ across representation, method selection, transformation, calculation and checking. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
22. Science repair
The starting principle is that science errors can occur in factual knowledge, mechanism, terminology, evidence use or conclusion scope. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
23. Examination error taxonomy
The starting principle is that lost marks become more informative when classified by mechanism instead of treated as one score. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
24. Careless mistakes
The starting principle is that the label careless is too broad when it hides a repeated reading, attention, copying or checking problem. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
25. Repeated mistakes
The starting principle is that the return of the same error can show that the repair is targeting the answer rather than the generator. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
26. Primary learners
The starting principle is that younger learners often benefit from short concrete corrections, visible contrasts and rapidly fading support. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
27. Secondary learners
The starting principle is that older learners increasingly need diagnosis of multi-step reasoning, transfer and performance under constraints. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
28. Parents
The starting principle is that parents can ask what the error reveals and what will happen next without turning every correction into a lecture. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
29. Teachers
The starting principle is that patterns across a class can reveal weaknesses in examples, explanations, sequencing or task wording. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
30. Self-correction
The starting principle is that the long-term destination is a learner who detects discrepancies and initiates appropriate repairs independently. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
31. Peer correction
The starting principle is that peer review can develop evaluative judgement when criteria are bounded and teacher verification prevents misconception spread. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
32. Spacing corrections
The starting principle is that an immediate correction should return later so the learner can show whether the repaired principle remains available. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
33. Transfer after correction
The starting principle is that the strongest practical test is whether a repaired principle survives a changed surface form. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
34. Useful error logs
The starting principle is that an error log should preserve recurring mechanisms and next actions without becoming an unusable archive. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
35. Ten-minute repair protocol
The starting principle is that a compact routine can convert one consequential error into diagnosis, correction, transfer and a future checking rule. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
36. When mistakes are not useful
The starting principle is that errors are not inherently beneficial when they arise from inaccessible instruction, unsafe conditions or misinformation. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
37. AI and error detection
The starting principle is that AI makes verification and independent explanation more important because fluent output can still be inaccurate. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
38. Seven-day experiment
The starting principle is that a short experiment can test whether one repair routine reduces one recurring mechanism without pretending to prove a universal law. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
39. Thirty-day review
The starting principle is that a monthly review can show which error categories are shrinking, persisting or being replaced by more advanced difficulties. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
40. World-return test
The starting principle is that error repair has succeeded when a better distinction improves later decisions outside the original worksheet. A visible wrong answer is an outcome, not a complete explanation. Two learners can produce the same answer through different routes, and one learner can produce several different errors from one underlying misunderstanding. Useful analysis therefore asks what information was available, what decision was made, what rule or representation controlled that decision, and what support changes the response.
Imagine a learner produces an incorrect response in an ordinary lesson. Preserve enough of the original working to inspect it before replacing it with the model. Ask what the learner thought the question required, which clue seemed important, and why the chosen response appeared reasonable at the time. This short reconstruction often separates missing knowledge from a misleading cue, an overgeneralised rule, a retrieval failure, a language problem or an execution slip. The categories matter because they imply different next actions.
The immediate repair should target the smallest consequential mechanism. If knowledge is absent, teach it. If a cue is misleading, contrast it with the cue that should control the decision. If the learner understands but cannot retrieve a term, create a brief closed-source attempt. If execution fails, isolate the vulnerable step and add a check. Giving the final answer can support teaching, but it should not be the final evidence of learning. The learner should reconstruct the corrected reasoning without simply copying.
Then change the example. Alter the names, numbers, wording, representation or context while preserving the principle being repaired. The original question is now familiar, so success on it can reflect memory for the correction rather than understanding of the governing relationship. A changed example asks whether the learner can recognise the same principle when surface cues move. Record any prompt that was needed. Prompted success is valuable evidence, but it is different from independent recognition.
Timing creates another distinction. An immediate successful correction shows that the learner can act while the explanation is recent. A later attempt asks whether the repaired knowledge remains available when the teacher’s wording and the original page are no longer carrying the response. This is where error repair connects with retrieval practice and spaced review. The delay does not magically fix the error; it creates another opportunity to see whether the repair survived.
For teachers, a useful record is a mechanism plus a next action. Instead of writing only “wrong”, note knowledge, retrieval, interpretation, selection, execution, evidence, language, checking or transfer where that category genuinely fits. For parents, the equivalent conversation is “What did this mistake show, and what will you try differently?” For students, it is “Where did my reasoning first stop matching the task?” Each question turns a judgement into a decision.
Avoid two extremes. One is treating mistakes as proof of low ability, which gives the learner no controllable next step. The other is celebrating every error as automatically productive. Errors can be ignored, rehearsed or reinforced. Their educational value is conditional on detection, accurate feedback, explanation, correction and another opportunity to act. This is why the broader research literature treats learning from errors as a process shaped by learner, task, emotion, context and instructional support rather than as a simple slogan.
The world-return test for this section is straightforward: after support is reduced and the surface form changes, does the learner make a better decision? If yes, the correction has begun to change capability. If not, return to diagnosis. The repair may have been too narrow, the prerequisite may still be missing, or the new task may add a difficulty that was not present before. Keep the conclusion proportional to the evidence and let the next attempt refine the model of what the learner needs.
Research floor and further routes
Research floor: Learning from errors and failure in educational contexts. Practical eduKateSG routes: feedback, practice, spaced repetition, study quickly, vocabulary learning and How X Works. All classroom cases in this guide are teaching illustrations unless a research source is explicitly identified.