How to Improve Learning From Mistakes | Diagnose, Repair & Prove the Fix

A wrong answer is not yet a lesson. A corrected answer is not yet a repaired learner. The useful question begins after the red cross: what changed inside the learner so that the same failure is less likely to return?

This article continues the eduKateSengkang How to Improve series after How to Improve Learning From Feedback. It is a practical companion to the canonical mechanism page How Learning From Mistakes Works | Error, Feedback, Repair and the Better Next Attempt. The mechanism page explains why errors can become useful information. This guide concentrates on the operational problem: how to diagnose one, repair the right cause, and prove that the repair survives beyond the original correction.

The examples below are instructional examples, not records of actual students and not official examination questions. Their purpose is to make the diagnostic decisions visible. A real learner may need different support, more evidence or specialist teaching before the same procedure applies.

The first rule: preserve the mistake before correcting it

When an error disappears under correction fluid, copied model working or an overwritten sentence, part of the evidence disappears with it. Keep the original visible long enough to ask what the learner was trying to do.

For a useful diagnostic, retain three things side by side:

the original task,
the learner’s original response or working,
the corrected or improved version.

The difference between the first and second versions is where learning can become visible. Without the first version, the adult may know the correct answer but not the route that produced the error.

This also protects against hindsight. Once the correct method is known, the wrong move can look absurd. Before correction, it may have been locally plausible to the learner. Understanding why it looked plausible is often the beginning of a durable repair.

Do not begin with “careless”

“Careless” can describe an outcome, but it is usually too broad to prescribe a repair. Two students can lose the same mark for completely different reasons.

One may not know the rule. Another may know it but fail to retrieve it. Another may select the wrong method. Another may understand everything but copy a negative sign incorrectly. Another may be rushing because of poor time allocation. Another may not notice that the answer violates a condition in the question.

Replace “careless” with a more useful description:

missed the final condition,
copied 0.06 as 0.6,
treated correlation as cause,
retrieved the wrong definition,
selected elimination when substitution was simpler,
made the correct inference but supplied no textual evidence,
failed to check an implausible result.

Precision is not about blaming more accurately. It is about making the next action possible.

Find the first useful cause, not merely the last wrong line

The visible error often appears late. The useful cause may appear earlier.

A Mathematics solution may end with the wrong number because the student made an arithmetic slip. But the more important problem may be that the learner chose a route with six unnecessary calculations when a direct representation would have required two. A Science answer may end with the wrong keyword because the student never identified which variable changed. An English paragraph may end with awkward expression because the claim itself was unclear before drafting began.

Trace backward until you reach an earlier decision that is both explanatory and realistically trainable. That point is the first useful cause.

Not every upstream cause deserves repair. A learner may have been hungry, annoyed and working under time pressure. Those conditions matter, but if the same conceptual mistake occurs under calm untimed conditions, the concept remains a stronger immediate target. Use evidence to decide which cause has the greatest educational leverage.

A practical error map

A useful working map separates several broad error families. These are hypotheses about a performance, not permanent labels for the learner.

Error family	What you may observe	First repair to test
Missing knowledge	The learner cannot explain the required concept even with time and prompts.	Teach or rebuild the missing relationship.
Retrieval failure	The answer returns after a small cue or when notes are visible.	Closed-book retrieval with spaced return.
Representation error	The problem is translated into the wrong diagram, equation, category or model.	Rebuild the representation before execution.
Selection error	The learner knows several methods but chooses the wrong one.	Contrast and interleave nearby cases.
Execution error	The method is sound but a local step is inaccurate.	Targeted practice plus a checking trigger.
Monitoring failure	An impossible or inconsistent answer is not noticed.	Teach plausibility and verification checks.
Transfer failure	The learner succeeds in familiar practice but fails after the surface changes.	Vary context and ask what deep structure remains.
State or timing failure	The error appears mainly under pressure, fatigue or particular task conditions.	Stabilise the underlying skill, then train under the relevant condition.

The classification does not need to be perfect on the first attempt. Its job is to generate a better repair hypothesis than “do more questions.”

Worked case 1: the same wrong algebra answer can require different repairs

Consider the equation 3(x + 4) = 24. Suppose two students both write 3x + 4 = 24.

The page looks identical. The diagnosis may not be.

Student A: the distributive relationship is missing

Ask Student A what the bracket means. The learner cannot explain why the three should affect both terms. Showing another answer immediately may produce copying without repair.

Use a concrete or verbal representation: three groups of (x + 4) means three x terms and three groups of four. Then translate:

3(x + 4) = 3x + 12.

Now compare with 3x + 4. The first repair target is conceptual: what multiplication over a bracket means.

Student B: the rule is known but not monitored

Student B explains distribution correctly and successfully expands five separate expressions. The error appeared only while solving the full equation quickly.

For this learner, another explanation of distribution may have low value. A better repair might be a checking trigger: when a multiplier sits directly outside a bracket, point to every term it acts on before simplifying. Then test whether that trigger survives inside longer equations.

Same wrong line. Different cause. Different repair.

Prove the algebra repair on a fresh task

Do not use the original equation as the only evidence of learning. It now contains memory of the correction.

Try 5(2x − 1) = 35. The correct expansion is 10x − 5 = 35, giving x = 4. Then use a contrast case such as 5(2x) − 1 = 35. The second expression does not place the minus one inside the bracket.

The contrast matters because a learner who merely memorised “multiply the second number too” can overgeneralise. The deeper rule concerns the structure of the expression.

Later, place bracket expansion among unrelated algebra questions. If the learner only succeeds when told that the question is about brackets, method recognition may still depend on external cueing.

Worked case 2: a comprehension mistake can be evidence, inference or scope

Use this invented extract:

When the results were announced, Nila folded the corner of her paper and placed it under her workbook. She congratulated her friend, then stayed behind after class to ask the teacher about two questions.

Suppose the question asks what Nila’s behaviour suggests about how she feels about her result.

A student writes: “Nila is angry because she hides her paper.” The answer is not absurd. It notices a relevant action. The problem lies in how confidently the action is interpreted.

Possible repairs differ:

If the learner cannot identify textual evidence, teach evidence selection.
If the learner treats one behaviour as proving a single emotion, teach inference under uncertainty.
If the learner ignores the later action of asking about two questions, teach whole-passage integration.
If the learner invents a reason not stated in the text, teach scope control.

A stronger answer could say that Nila appears disappointed or concerned about the result because she puts the paper away and remains after class to ask about mistakes. The wording should reflect what the text supports rather than claim access to Nila’s inner state with certainty.

The improvement target is not “use better vocabulary.” It is evidence-to-inference control.

Test the English repair by changing the surface

A fresh passage should not simply repeat “student gets disappointing result.” Use a different setting:

At the end of rehearsal, Marcus remained beside the piano while the other performers packed their bags. He played the opening four bars again, stopped, and wrote something above the first line of music.

Ask what Marcus’s actions suggest. A learner who has repaired the deeper operation should identify evidence, offer a restrained interpretation such as concern with improving or correcting the opening, and avoid inventing a detailed history not present in the extract.

That fresh performance is stronger evidence than a polished rewrite of the Nila answer.

Worked case 3: a Science mistake may contain correct facts in the wrong relationship

Imagine an investigation in which identical wet cloths are placed in two locations. One location has moving air from a fan; the other does not. The cloth near the fan loses more mass over the same period.

A student writes: “The fan gives the cloth more heat, so the water evaporates faster.” The student may know that evaporation can be affected by conditions, but the proposed mechanism does not follow from the stated comparison.

The repair should not be “memorise that fans increase evaporation.” Ask what variable differs between the set-ups. The intended changed condition is air movement. Then rebuild the causal explanation at the appropriate level: moving air carries away water vapour near the cloth, helping maintain a difference between the water vapour near the wet surface and the surrounding air, so evaporation can continue more rapidly under the stated conditions.

The important repair is not an extra keyword. It is the relationship between the changed condition and the measured outcome.

For the dedicated subject route, connect this style of correction to How to Turn PSLE Science Feedback Into a Testable Repair rather than duplicating that specialist page here.

Do not repair a misconception by memorising an exception

A learner says, “Heavier objects always fall faster.” One unusual example may convince the learner to memorise, “except these two objects.” The original model remains intact and receives an exception.

A stronger repair replaces the underlying rule. Ask what variables affect the motion in the situation, what is being held constant and what the evidence actually shows. The learner should leave with a more accurate model, not a longer list of exceptions.

This principle applies beyond Science. A grammar misconception should be replaced with a better grammatical model. A Mathematics misconception should be repaired at the structural relationship. A comprehension misconception should be repaired by returning to what evidence licences the inference.

Separate a knowledge gap from a retrieval gap

Ask the learner to explain the relevant idea without the notes. If nothing useful can be reconstructed even with time, missing knowledge is plausible. If a small cue immediately restores a correct explanation, retrieval may be the stronger candidate.

The repairs differ.

Knowledge gap: teach or rebuild the concept.
Retrieval gap: require recall under progressively reduced cues and return after delays.

Do not infer too much from one cue. A hint can sometimes help the learner reason toward an answer they did not previously possess. Use several observations where the distinction matters.

Separate a method-selection error from a method-execution error

Students often practise a method until they can execute it, then fail when no one tells them which method to use. That is a selection problem.

Imagine a learner who can solve ratio questions accurately in a ratio worksheet and percentage questions accurately in a percentage worksheet. In a mixed set, the learner repeatedly applies percentage change to questions asking for part-to-whole ratios.

More blocked ratio practice may not solve the problem. The learner needs discrimination practice: compare nearby question types, identify the cue that matters, predict the method before calculating, and explain why the rejected method does not fit.

Execution skill answers, “Can I perform this method?” Selection skill answers, “Can I recognise when this method is the right one?”

Separate a one-off slip from a recurring process weakness

Not every error deserves a programme of remediation. Human performance contains noise.

If a learner makes one arithmetic slip across twenty accurate problems, correct it and observe. If the same type of slip appears repeatedly in similar conditions, treat it as a pattern worth investigating.

Ask:

Has this error appeared before?
Does it appear under a particular condition?
Does the learner detect it when prompted to check?
Does it disappear when the task is untimed?
Does a simple checking routine reduce recurrence?

The purpose is proportional response. Do not build a major intervention around noise, and do not dismiss a persistent pattern as a random slip.

When the learner is right for the wrong reason

A correct endpoint can conceal a faulty route. That makes correct answers part of error analysis too.

A learner may guess correctly, cancel two errors, follow a memorised surface pattern or select an option for an invalid reason. If the result is simply ticked and forgotten, the misconception may remain invisible until a harder question removes the accidental support.

For selected answers, ask:

How did you know?
Which evidence mattered?
Why does this method fit?
What would change your answer?
Which alternative did you reject?

The goal is not to interrogate every correct answer. Use the check when the answer is important, surprising, low-confidence or produced by a learner with a known misconception.

When the learner is wrong despite a good process

Outcome and process should also be separated in the other direction.

In uncertain reasoning, a learner can use strong evidence and still reach an answer that turns out to be wrong. In Mathematics, a long solution can be structurally excellent until one local arithmetic error. In writing, a thoughtful interpretation may be plausible even if another interpretation is better supported.

Preserve what worked. Repair the failing component without discarding the entire route.

This matters for motivation. If every wrong endpoint is treated as evidence that the entire thinking process was bad, learners receive distorted information about what to keep and what to change.

Use the smallest repair that matches the cause

Once a cause is plausible, choose the smallest intervention capable of changing it.

A missing concept may need explanation and examples.
A retrieval failure may need closed-book recall.
A selection error may need contrast and mixed practice.
An execution slip may need one checking rule.
A representation error may need diagramming or translation practice.
A monitoring error may need estimation or verification.
A transfer failure may need varied surface forms.

Do not prescribe more total work when one narrow operation is responsible. Precision improves both efficiency and the learner’s sense that improvement is controllable.

The correction must be produced by the learner

A tutor can produce an excellent correction while the learner remains unchanged. A model answer can make the page perfect without proving ownership.

After explanation or feedback, return the central operation:

rewrite the sentence,
redo the algebraic step,
redraw the diagram,
choose the method again,
state the evidence-to-claim link,
explain the mechanism in new words.

Support can remain around the learner where needed, but the skill being learned must eventually be expressed by the learner.

The MindOS Correction State makes this boundary explicit: can the learner repair the error without copying the fix?

A fresh reattempt is the first proof

The original item is contaminated by the correction. The learner has seen the route, the answer and perhaps the teacher’s annotations.

Use a fresh item that requires the same underlying decision.

new numbers for a Mathematics structure,
a new passage for an inference skill,
a new investigation for a Science reasoning error,
a new sentence for a grammar repair,
a new context for a vocabulary distinction.

If the learner succeeds independently, the repair has passed its first useful test.

If the error returns, do not simply repeat the same correction. Revisit the diagnosis. The original cause may have been wrong, or the repair may not have been sufficient.

Change the surface to test transfer

A learner may succeed on a fresh question that looks almost identical to the correction. That proves less than success after a meaningful surface change.

Change one dimension at a time:

numbers,
context,
representation,
order of information,
response format,
which quantity is unknown,
which concept is placed nearby.

Then ask what deep feature still makes the repaired rule relevant.

This is where correction begins to become learning transfer rather than answer repair.

Return after a delay to test whether the repair held

Immediate success can depend on short-lived memory of the explanation. Return later, after the correction is no longer sitting in working memory.

There is no universal delay that fits every learner and task. Use a delay meaningful to the learning schedule: later in the week, in the next lesson, or inside a cumulative review. State the conditions if you are recording progress.

If the learner succeeds again, confidence in the repair rises. If the error returns, you have stronger evidence that the route is still fragile.

See How Spacing Works in Learning for the wider mechanism.

When the same mistake returns after correction

A recurring error is not simply “the same mistake again.” It is evidence that some part of the previous repair did not hold.

Check for several possibilities:

The correction was copied, not generated.
The explanation was understood but not retrieved later.
The learner practised only one surface form.
The learner knows the rule but cannot detect when it applies.
The checking routine disappears under time pressure.
The original cause was misdiagnosed.
A new task adds working-memory demands that were absent during correction.

Use the dedicated route Why Do the Same Mistakes Return After Correction? when recurrence itself becomes the main problem.

Build an error return, not an error museum

Correction books can become warehouses of old mistakes that are never revisited. Recording is useful only if it changes later learning.

For recurring or high-value errors, record:

the task or error family,
the first useful cause,
the repair,
the fresh test,
whether help was required,
the next return date or condition.

Once an error is stable across fresh and delayed cases, reduce its review frequency. Do not keep every past weakness permanently active. The purpose of the record is control, not identity.

Do not turn error history into a label

“She always makes sign mistakes” can become self-fulfilling if it replaces current evidence. Perhaps the sign error was repaired three months ago. Perhaps it only appears in long simultaneous-equation working. Perhaps the present issue is now time allocation.

Use error history as a prior, not a verdict. Check the current task. Old patterns deserve attention because they may recur, but they should not control the interpretation of new evidence automatically.

This is especially important for high-performing students whose error profile changes as the work becomes more advanced.

High-performing learners need error analysis too

A score of 90% can hide a decisive weakness. The learner may know nearly all the content yet repeatedly:

overclaim from incomplete evidence,
skip verification,
choose a long route under time pressure,
lose marks on proof structure,
become overconfident after familiar practice,
fail to transfer to unfamiliar representations.

At higher levels, the number of errors may fall while the cost of each error rises. Error analysis therefore becomes more selective, not less important.

Use confidence to find hidden mistakes

Ask for confidence before checking selected answers. A high-confidence wrong answer is especially informative because the learner’s internal model and external performance disagree strongly.

A low-confidence correct answer can also deserve attention. The learner may have the right knowledge but not trust or recognise it reliably.

The goal is not to make students uncertain. It is to make confidence more proportional to evidence.

Connect this to How Learning Calibration Works and How to Improve Metacognition.

Use prediction to reveal the learner’s model before correction

Before showing the answer, ask the learner to predict. Prediction commits the current model to an observable form.

Examples:

Which object will warm faster?
Which algebraic method will be shorter?
What emotion does this passage most strongly support?
Which word best fits this context?
Which variable should be controlled?

When the prediction and result differ, the learner has a clearer discrepancy to explain.

Do not engineer failure merely to create mistakes. Prediction is useful when it activates relevant prior knowledge and makes the learner’s current model inspectable.

Mistakes and productive struggle

A difficult task can generate useful mistakes because it forces search. But repeated failure becomes unproductive when attempts stop producing new information.

During struggle, ask:

What did this attempt reveal?
Which possibility can now be ruled out?
What representation should change?
Is the learner still generating useful information?
Has the point arrived for a hint or explanation?

See How to Improve Productive Struggle.

Mistakes and feedback

Feedback should identify enough of the discrepancy for the learner to make a better next attempt. It should not automatically perform the repair for them.

If feedback says only “wrong,” the learner may not know what to change. If it rewrites the entire answer, the learner may not know what they changed. Useful feedback sits between those extremes and is followed by learner action.

Continue through How to Improve Learning From Feedback.

Mistakes and reflection

After correction, reflection should produce a new decision rather than a statement of regret.

Weak reflection:

I should be more careful.

Stronger reflection:

I answered before using the final condition. On the next similar question, I will mark every condition before calculating, then check whether the error disappears.

See How to Improve Reflection.

Mistakes and motivation

Mistakes can damage motivation when they are interpreted as evidence that effort does not work. They can support motivation when they reveal a tractable change and the learner later sees that the change improves performance.

That makes the repair loop psychologically important:

Error → Specific Cause → Controllable Repair → Better Attempt

The learner sees not simply “I was wrong,” but “this particular action changed the result.”

See How to Improve Motivation.

Mistakes and emotion

A learner may know how to correct an error and still avoid looking at it because the mistake carries embarrassment, disappointment or fear.

Separate the person from the performance. The answer can be wrong without the learner being “bad at” the subject. The standard does not need to be lowered. What changes is the interpretation of the error: it becomes information about a current state.

If emotional distress is intense or persistent, learning tactics alone may not address the whole situation. Use appropriate pastoral or professional support where needed.

Mistakes under time pressure

A mistake that appears only in timed conditions should not automatically be taught as a knowledge gap.

Compare conditions:

Can the learner solve the same type accurately untimed?
Does the error appear late in the paper?
Does it occur after unusually long questions?
Does a skip-and-return rule reduce it?
Does the learner’s checking routine disappear when time feels scarce?

Stabilise the core skill first if needed. Then train it under the condition in which it fails.

Mistakes caused by working-memory overload

A learner may know every component and still lose the route when too many elements must be held simultaneously.

Look for a pattern: performance is accurate on short versions but deteriorates as the number of steps, conditions or representations increases.

Possible repairs include:

externalising intermediate values,
marking conditions,
using diagrams or tables,
breaking the task into subgoals,
building automaticity in basic operations.

See MindOS Working Memory Load.

Mistakes caused by interference

Sometimes the learner knows two similar rules and retrieves the wrong one because they compete.

Examples include:

area and perimeter formulas,
mass and weight,
affect and effect,
mean and median,
similar algebraic transformations.

Repair through contrast. Put the competing ideas beside each other and identify the cue that decides between them. Then mix them so the learner must discriminate rather than rely on chapter order.

See How Interference Works in Learning.

Mistakes caused by overgeneralisation

A learner discovers a useful rule and begins applying it beyond its valid boundary.

Use non-examples and edge cases. Ask not only “When does this work?” but “What condition must be present?” and “What nearby case looks similar but follows another rule?”

Boundary knowledge is part of mastery. A rule that cannot be switched off when inappropriate remains dangerous.

Mistakes caused by undergeneralisation

The opposite failure also occurs. A student learns a method in one narrow form and does not recognise it elsewhere.

A learner can solve percentage questions about prices but not identical percentage structure in population or mass. A learner can identify cause and effect in a Science chapter but not in a reading passage. A writer can use evidence correctly in one essay topic but not another.

Ask what deep relationship is shared across the cases. Vary the surface while keeping the underlying rule stable.

Mistakes in AI-assisted learning

AI creates a special risk: it can repair the visible output while hiding whether the learner repaired the underlying skill.

If an AI system rewrites a paragraph, solves an equation or supplies a complete Science explanation, the final product may be correct while the learner’s original weak link remains untouched.

Use AI as a diagnostic or feedback tool more carefully:

show the original attempt,
ask it to identify the first questionable step rather than rewrite everything,
ask for one hint where possible,
verify factual claims,
hide the response,
reconstruct independently,
use a fresh transfer task.

Record AI assistance honestly where academic or school rules require it. A polished AI-assisted submission is not, by itself, a measurement of independent mastery.

See How AI-Assisted Study Works.

A five-minute error repair

Minute 1: Circle or name the first point where the route becomes wrong or uncertain.
Minute 2: State the most plausible error family.
Minute 3: Explain the missing rule, relationship or checking decision.
Minute 4: Repair the original without copying a model.
Minute 5: Attempt one fresh case.

If the learner cannot complete minute two or three, the repair may need more teaching. If the fresh case fails, return to diagnosis rather than automatically adding repetition.

A deeper fifteen-minute error clinic

Preserve the original work.
Ask the learner to reconstruct what they were trying to do.
Locate the first useful cause.
Classify the failure tentatively.
Teach or prompt only the missing piece.
Have the learner produce the repair.
Give a fresh similar task.
Give a contrast or changed case.
Record whether help was required.
Plan a delayed return if the error is important or recurring.

The exact duration should fit the task. A sentence-level punctuation error and a conceptual modelling failure do not deserve identical time budgets.

A weekly error review

Do not review every mistake equally. Select the ones with the greatest expected value.

Which error repeated?
Which error cost many marks or blocked several topics?
Which correction failed to transfer?
Which high-confidence answer was wrong?
Which error disappeared after one narrow repair?
Which old error no longer deserves active attention?

The review should change next week’s practice. Otherwise it becomes administration rather than learning.

For parents: ask about the mechanism, not the embarrassment

A useful conversation can begin with: “Show me where the answer first stopped matching the question.”

Then ask:

What did you think at that point?
What makes the corrected version different?
Could you do a new one without looking?
What will help you notice this earlier next time?

If the child cannot explain the error, help them formulate a question for the teacher or tutor. Parents do not need to invent a subject explanation when the evidence is unclear.

Avoid turning every mistake into a character judgement. “You are careless” gives the child little to act on. “You used the correct method but copied the sign incorrectly on line three; what check could catch that?” returns the conversation to a controllable operation.

For teachers: diagnose before assigning more practice

Ten more questions are useful only if repetition targets the actual failure.

Before assigning volume, ask:

Does the learner know the concept?
Can they retrieve it?
Can they identify when it applies?
Can they execute it?
Can they monitor it?
Can they transfer it?

Then choose practice that exercises the weak operation. Blocked repetition may build execution. Mixed practice may build selection. Changed cases may build transfer. Delayed return may build durability.

For tutors: do not become the permanent error detector

A tutor can see mistakes faster than a learner. If the tutor always identifies every problem immediately, supported performance can improve while self-monitoring remains weak.

Where appropriate, ask the learner to check first:

Does the answer satisfy the question?
Is the result plausible?
Which step carries the highest error risk?
What would you inspect before I tell you?

Then add expert feedback where the learner’s internal model is insufficient. The long-term aim is earlier self-detection, not permanent tutor surveillance.

When not to use an error-first approach

Some tasks should not require learners to discover everything through failure.

Direct explanation may be preferable when:

the prerequisite knowledge is absent,
the task carries safety consequences,
the learner has no plausible entry point,
repeated failure is producing no useful information,
the goal is efficient introduction of a completely new procedure.

The educational value of an error depends on what can be learned from it. Failure is not a mandatory entrance fee to every lesson.

How to know the mistake has actually been learned from

Look for a sequence of stronger evidence rather than one neat correction.

The learner can locate the first useful cause.
The learner can explain why the original route failed.
The repair is produced rather than copied.
A fresh similar task is correct.
A changed case does not trigger the same misconception.
The learner detects or prevents the error earlier.
The repair survives a meaningful delay.
Less external help is required for the same operation.

Not every error needs all eight checks. Use stronger evidence when the error is high-value, recurring or foundational.

What not to conclude from one successful correction

One corrected problem does not prove broad mastery. One fresh question does not prove permanent retention. One week without recurrence does not prove the error can never return.

Use language that matches the evidence:

“The learner corrected the original independently.”
“The learner succeeded on two fresh examples without prompts.”
“The error did not recur in this mixed set.”
“The repair has not yet been tested under time pressure.”

Precise claims protect both optimism and caution from becoming exaggeration.

The strongest correction question

After any important mistake, ask:

What would I need to see on a different task, later, to believe this error has actually been repaired?

That question forces the correction to look beyond the original page. It connects diagnosis, repair, transfer, spacing and independence in one move.

Continue the How to Improve route

Final principle

A mistake becomes educationally valuable only when it changes a later decision or performance.

Preserve the evidence. Find the first useful cause. Repair the smallest relevant weak link. Return the work to the learner. Test the repair on a fresh case, then again after the correction is no longer fresh.

The corrected page is not the finish line. The better next attempt is.

Small Group Tutorials

How to Improve Learning From Mistakes | Diagnose the Error, Repair the Cause and Prove the Fix