MindOS · Learning Operation · Compare → Locate → Classify → Repair → Reattempt → Transfer → Return
Wait, What? Telling a Student the Answer Is Wrong May Teach Them Almost Nothing
A red cross is information.
It tells the learner that the produced answer does not meet the expected standard.
But it does not necessarily tell them where the reasoning departed, why it departed, what should change, or whether the learner can now perform the repaired operation independently.
MindOS calls this Feedback State: can the learner convert information about performance into a precise next action?
Quick Answer
Useful feedback closes a gap between current performance and a target by helping the learner identify the discrepancy, locate its cause, choose a repair and reattempt. The important unit is not the comment. It is the feedback loop.
ATTEMPT ↓ COMPARE WITH CRITERION ↓ LOCATE DISCREPANCY ↓ CLASSIFY THE ERROR ↓ CHOOSE SMALLEST USEFUL REPAIR ↓ LEARNER REATTEMPTS ↓ REMOVE MODEL / PROMPT ↓ NEW ITEM ↓ DELAYED RETURN
The Owned Learner Job
This page owns feedback uptake: turning performance information into repair and independent reattempt. It does not own marking systems, motivation as a whole, correction as a separate future branch, or every form of teacher communication. The learner job is narrower: after receiving useful information, can they act on it?
The reflective article A Corrected Answer Is Not Necessarily a Repaired Mistake raises the problem. This MindOS manual operationalises it.
Feedback Is Not the Same as a Grade
“7/10” measures performance. “Wrong” identifies a mismatch. “Use more evidence” points toward a criterion. “Your inference is plausible, but the quotation you chose does not support it; choose evidence that directly changes the probability of your claim” is more actionable.
But even excellent teacher feedback can fail if the learner does nothing with it. The loop is incomplete until the student produces a changed response.
Observable Signatures of a Weak Feedback State
- The learner reads comments but repeats the same error.
- They copy the model answer as the “correction.”
- They can state what the teacher wrote but not what operation should change.
- They repair the original item while the answer is visible but fail a parallel item.
- They interpret every error as “careless.”
- They make many simultaneous changes and cannot tell which one fixed the problem.
Discriminate Before Blaming Feedback
Does the learner understand the criterion? If “explain” or “justify” is itself unclear, feedback cannot close a gap the learner cannot represent.
Is prerequisite knowledge missing? “Check your formula” is useless if the learner never learned the formula or cannot retrieve it.
Is the feedback too large? Ten comments on one paragraph may create a load problem. Prioritise the earliest or highest-leverage repair.
Is the learner merely copying? Hide the model and ask for a fresh reconstruction. If the repair disappears, the answer was corrected but the operation may not have changed.
Intervention 1: Locate the First Divergence
Do not begin at the final wrong answer if the error started earlier. Compare the learner’s route with the criterion and find the earliest useful divergence.
- misread question;
- missing concept;
- retrieval failure;
- wrong representation;
- poor strategy choice;
- execution slip;
- evidence mismatch;
- expression that fails to show correct reasoning.
The category matters because each implies a different repair.
Intervention 2: Convert Comments Into Actions
Translate vague feedback into an observable learner move.
- “Be clearer” → underline the claim and identify the exact noun each pronoun refers to.
- “Show working” → write the intermediate state that makes the next operation checkable.
- “Use evidence” → identify the claim, then select the sentence that most directly supports it.
- “Careless” → locate the precise checkpoint that would have caught the error.
- “Wrong method” → state the cue that should have triggered a different strategy.
Feedback becomes useful when the learner knows what to do differently on the next attempt.
Intervention 3: Require a Reattempt
The most important moment may come after the explanation. Remove the answer and ask the learner to perform the repaired operation again.
If the tutor rewrites the sentence, solves the equation or supplies the Science explanation, the artifact is repaired. The learner still needs a turn.
Intervention 4: Use the Smallest Useful Feedback
Too little feedback leaves the learner lost. Too much can perform the task for them. Calibrate assistance to state.
- Blocked: model or explain the missing prerequisite.
- Fragile: point to the location and ask a guiding question.
- Stable: identify the discrepancy but let the learner generate the repair.
- Transfer-ready: give criterion-level feedback and ask the learner to diagnose the exact failure.
- Examination-ready: delay feedback until after a realistic independent attempt, then analyse the return.
This mirrors the logic of The AI Assistance Gradient: assistance should reveal and build capability, not permanently substitute for it.
Evidence: Feedback Works Through Design and Uptake
The Education Endowment Foundation’s Teacher Feedback to Improve Pupil Learning guidance emphasises foundations such as high-quality initial teaching, appropriately timed feedback, and planning for how pupils will receive and use it. That final point is central to MindOS: feedback that never becomes learner action has not completed the loop.
The evidence base does not imply that one feedback format is universally best. Timing, task, learner knowledge, specificity and opportunity to act all matter. MindOS therefore treats feedback as a controlled educational intervention whose result must be observed.
Feedback and Retrieval
When a learner cannot retrieve an answer, immediate exposure to the correct version may be useful—but the cycle should end with another retrieval attempt. Otherwise feedback can become repeated recognition. See Retrieval State.
Feedback and Strategy Selection
“Wrong method” is weak feedback unless the learner learns the cue that should change the selection next time. Link the error to the discriminating feature. See Strategy Selection.
Scaffold Fade
TEACHER IDENTIFIES ERROR + REPAIR ↓ TEACHER IDENTIFIES ERROR, LEARNER REPAIRS ↓ TEACHER IDENTIFIES REGION, LEARNER DIAGNOSES ↓ CRITERION ONLY ↓ LEARNER SELF-CHECKS ↓ NEW ITEM ↓ DELAYED ITEM ↓ EXAMINATION CONDITIONS
Transfer Test
- Give a parallel item without the original answer visible.
- Change the wording or representation.
- Ask the learner to identify the relevant criterion before attempting.
- Include a near-miss error and ask them to diagnose it.
- Return after a delay.
If the learner can repair only the marked original, the correction has not yet transferred.
Examination Test
During an examination, external feedback disappears. The learner needs internal feedback loops: does this answer match the question, is the magnitude plausible, did I use evidence, does the method fit, have I answered every part? Examination Craft is partly the ability to generate useful feedback before the marker does.
Return Test
- Can the learner name the exact discrepancy?
- Can they classify the error without a vague label?
- Can they state the repair action?
- Can they reattempt without copying?
- Can they catch the same error independently next time?
- Does the repair survive a changed problem and a delay?
Common Misconceptions
- “More comments mean better feedback.” Volume is not the same as actionability.
- “The model answer fixes the mistake.” It fixes the artifact unless the learner reconstructs the operation.
- “Careless” is a diagnosis. It often hides a missing checkpoint or earlier weak link.
- “Immediate feedback is always best.” Timing should match the task and learning purpose.
- “Students should discover every error alone.” Blocked learners may need explicit teaching; the support should then fade.
Parent and Tutor Teaching Guide
When work is wrong, resist replacing it immediately. Ask: Where did the route first diverge? What kind of error is this? What is the smallest change that repairs the underlying operation? Then hide the model and make the learner try again. Finally, test a new item. The goal is not a clean correction page. It is a learner who increasingly generates useful feedback for themselves.
MindOS Direction Graph
FEEDBACK STATE ├── Criterion unclear? → CLARIFY TARGET ├── Knowledge absent? → TEACH / RETRIEVAL ├── Error starts earlier? → FIRST WEAK LINK ├── Feedback too large? → PRIORITISE ├── Learner copies repair? → REMOVE MODEL + REATTEMPT ├── Repair works only here? → TRANSFER STATE ├── Learner self-detects? → FADE EXTERNAL FEEDBACK └── Self-check survives pressure? → EXAMINATION CRAFT
MindOS boundary: Feedback State concerns educational performance and repair. It does not diagnose motivation, attention, developmental or mental-health conditions from a student’s response to feedback.
