Feedback does not improve learning merely because it exists. A comment can be accurate and still produce no change. A mark can be informative and still become only a judgement. A correction can show the right answer and still leave the original error mechanism untouched. A student can receive pages of detailed advice, agree with every word, and make the same mistake again two weeks later.
This is the central problem of feedback failure: information reaches the learner, but the information does not become a changed decision, changed strategy, changed answer or changed capability. The failure is not necessarily that the teacher explained badly or that the student ignored advice. The break can occur anywhere between evidence, interpretation, action, practice, retrieval and transfer.
This guide is the failure-mechanism owner inside the Learning Problems Hub and the wider Recovery Atlas. It does not replace How to Translate Feedback Into Change, which teaches the action loop after useful feedback has been identified. This page asks the earlier diagnostic question: why did the feedback fail to change learning in the first place?
The evidence base supports treating feedback as a process rather than a format. The Education Endowment Foundation’s Teacher Feedback to Improve Pupil Learning guidance emphasises strong initial instruction, clear learning intentions, assessment of learning gaps and feedback that moves learning forward; it also notes that not all feedback has positive effects. The EEF’s updated Metacognition and Self-Regulated Learning guidance similarly stresses planning, monitoring and evaluating learning. These principles matter because a learner must do something cognitively useful with feedback before it can alter later performance.
Feedback is an input, not an outcome
Teachers, tutors, parents and students often talk as though feedback is a completed action: “I gave feedback.” “She got corrections.” “The teacher told him what was wrong.” But from the learner’s perspective, those are only inputs. The educational outcome is whether the next relevant performance changes.
A complete feedback loop is closer to:
performance → evidence → feedback → learner interpretation → chosen change → practice → retrieval → fresh task → changed performance.
If the chain breaks at any point, feedback can remain visible while learning remains unchanged. A marked script full of comments may create the appearance of a feedback-rich environment even when the learner cannot state what they will do differently on the next fresh task.
Failure mechanism 1: the feedback describes the product but not the cause
“Needs more detail.” “Careless.” “Weak vocabulary.” “Show working.” “Answer the question.” These comments may point toward a genuine weakness, but they often describe the appearance of the final answer rather than the mechanism that produced it.
Consider a Secondary English student whose comprehension answer is too general. The visible product is vague. But the cause could be one of several things: the student did not identify the exact demand; selected the wrong evidence; understood the passage but did not convert evidence into a precise answer; or wrote a broad explanation because previous practice rewarded length over scope. “Be more specific” does not distinguish those mechanisms.
Similarly, a Mathematics student may be told to “show working”. The real cause may be that the student jumps mentally between steps, does not know which intermediate state matters, or believes working is merely presentation rather than a tool for controlling error. The same surface comment can point to different repairs.
Diagnostic test: ask the learner to explain why the error happened, not only what the correct answer should have been. If the explanation remains “I was careless” or “I forgot”, the mechanism is probably still too vague to guide repair.
Failure mechanism 2: the learner cannot decode the feedback
Feedback can be technically correct but inaccessible. A comment such as “develop evaluation”, “justify inference”, “improve coherence”, “show causal reasoning” or “use a more efficient method” assumes the learner already understands what those actions look like.
An expert compresses knowledge. A teacher may use one phrase to represent a whole network of decisions. The learner may hear only the phrase. This creates a dangerous illusion of shared meaning: both people use the same words while imagining different actions.
Diagnostic test: ask the learner, “What would I see you doing differently in the next answer?” If the learner can only repeat the comment, the feedback has not yet been operationalised.
Failure mechanism 3: the feedback arrives after the learner has mentally closed the task
A marked test can feel finished. The learner wanted the score; the score arrived; attention moves to the next chapter. Feedback then competes with the learner’s psychological sense that the task is over.
This is one reason corrections can become rituals. The learner copies the correct answer, signs the paper, files it and proceeds. The school has a correction process, but the cognitive system has not reopened the decision that produced the error.
Diagnostic test: after corrections, give a fresh question that requires the same decision but changes the surface. If the original error returns immediately, the correction closed the old task without changing the underlying rule.
Failure mechanism 4: the learner sees the answer but does not reconstruct the decision
Seeing a correct answer can create a strong sense of understanding. Once the worked solution is visible, every step appears obvious. That feeling is not reliable evidence that the learner can generate the solution independently.
A useful correction therefore asks for reconstruction. What clue in the question should have triggered the method? Where did the learner first diverge? What rule applies? What alternative answer looked plausible and why is it wrong?
This converts correction from answer exposure into decision analysis.
Failure mechanism 5: the feedback asks for too many changes at once
A heavily annotated piece of work can be accurate and still be unusable. The learner sees ten things to fix: grammar, vocabulary, structure, evidence, handwriting, timing, checking, explanation, relevance and presentation. The result is often no clear priority.
Feedback competes for working attention. If everything is important at the same moment, the learner may make superficial changes or default to the easiest visible correction. The highest-leverage error remains.
Diagnostic test: ask, “What is the one change that would most improve the next attempt?” If neither teacher nor learner can choose, the feedback may need prioritisation before further practice.
Failure mechanism 6: the feedback targets something the learner cannot yet do independently
Feedback can correctly identify a target that sits above several missing prerequisites. Telling a student to “evaluate the writer’s effect” does not help if the student cannot first identify the relevant language choice. Telling a Mathematics student to “use a more efficient algebraic method” does not help if basic manipulation remains unstable.
The correct feedback target may therefore be too high in the dependency chain. Recovery requires descending to the first missing prerequisite and rebuilding upward.
This is the core logic of Academic Recovery: repair the first broken dependency rather than repeatedly commenting on the final symptom.
Failure mechanism 7: the learner does not get a second attempt soon enough
Feedback without a nearby opportunity to use it decays into advice. The learner may understand the comment in the moment but never convert it into a new action while the reasoning is still active.
The best next task is not always another full paper. Often it is a small fresh task that isolates the exact decision the feedback is meant to change. Once that succeeds, the skill can return to mixed and representative contexts.
Failure mechanism 8: the second attempt is too similar
A student can appear to respond to feedback because the next question closely resembles the corrected one. The learner remembers the surface rather than applying the principle. This produces local improvement without transfer.
After an initial corrected attempt, change something that should not matter: context, wording, numbers, order, passage topic or representation. If the learner can still apply the changed decision, the feedback is becoming portable.
Failure mechanism 9: the learner changes the answer, not the rule
A correction can fix one response while leaving the learner’s internal decision rule untouched. “For this question, use method B” is not the same as “When these conditions appear, method B is appropriate because…”.
Transfer depends on rule change. The learner needs a cue that can be recognised in a future task.
Useful prompt: “What will you notice next time that tells you to act differently?”
Failure mechanism 10: the feedback is experienced as judgement rather than information
Marks and comments can become identity signals: “I am bad at English.” “I am careless.” “I cannot do algebra.” Once feedback is interpreted globally, it stops pointing toward a bounded change.
The repair is not empty reassurance. It is to return the feedback to observable evidence. Which task? Which decision? Which condition? Which next attempt would show improvement?
Rebuilding Confidence is useful when repeated performance feedback has become disconnected from the learner’s actual capability.
Failure mechanism 11: the feedback is accurate but too late for the next decision
A teacher can return detailed comments after the class has already moved through several related tasks. The learner now has to reconstruct the original thinking before the feedback is meaningful. The distance between performance and feedback increases the effort required to connect the two.
When delay is unavoidable, reactivation helps. Ask the learner to revisit the question, explain what they remember doing, attempt one similar item or predict the feedback before reading it. The aim is to reconnect the comment to the decision state it is meant to modify.
Failure mechanism 12: the feedback is consumed passively
Reading comments is not the same as processing them. Nodding, highlighting, underlining or saying “okay” may create recognition without generating a plan.
Useful processing requires the learner to produce something: classify the error, explain the cause, rewrite a rule, attempt a correction from memory, compare two answers or state the next decision in their own words.
This links feedback to metacognition: plan, monitor, evaluate, then alter the next action. The EEF’s current metacognition guidance emphasises explicitly teaching and modelling those processes rather than assuming pupils naturally infer them.
Marks are not automatically feedback
A mark tells the learner where a performance landed on a scoring system. It does not automatically explain why the result occurred or what to do next. Two students can receive the same mark for entirely different reasons.
A 60% Mathematics result might come from broad conceptual gaps, excellent knowledge with catastrophic time management, repeated sign errors, one missing prerequisite that contaminates many questions, or weak transfer from routine exercises. The mark identifies a state. Diagnosis identifies the next move.
The same applies in English and Science. The score is useful evidence, but it becomes instructional only when connected to mechanisms.
Comments are not automatically instructions
“Explain more clearly” describes a desired quality. It does not necessarily tell the learner what operation creates clarity. “Use evidence” names a requirement. It may not teach how to choose, embed or explain evidence. “Check your work” is a reasonable expectation. It may not specify which checks catch the learner’s recurring errors.
Turn comments into behaviours. A useful behaviour can be observed, practised and checked.
| Comment | Possible operational translation |
|---|---|
| “Be more specific.” | Underline the exact question demand; select one phrase of evidence; write one sentence that directly links the evidence to that demand. |
| “Show working.” | Record each state change in a new line and label the relationship before calculating. |
| “Use better vocabulary.” | Replace one general word only when the new word matches meaning, register and collocation. |
| “Check units.” | Circle the required unit before solving and compare it with the final line before moving on. |
| “Explain the cause.” | Name the changed factor, mechanism and resulting effect in one causal chain. |
Corrections are not automatically learning
Corrections matter when they change future behaviour. Copying a model answer may improve the page and leave the learner unchanged. The learner must reconstruct the reasoning, then test it on something fresh.
A minimal correction loop is:
- Hide the solution.
- State what went wrong.
- State the rule or decision that changes.
- Redo the item from the point of failure.
- Attempt one new item requiring the same decision.
- Return after delay.
If the fresh or delayed item fails, the correction has not yet become learning.
The feedback uptake ladder
| Level | Learner can… | What is still missing |
|---|---|---|
| 1. Receive | See or hear the feedback. | Meaning. |
| 2. Restate | Repeat what the feedback says. | Operational change. |
| 3. Explain | Describe why the previous attempt failed. | Reliable changed action. |
| 4. Correct | Fix the original item. | Transfer. |
| 5. Apply | Use the changed rule on a fresh similar task. | Variation and durability. |
| 6. Transfer | Use the rule when the task changes form. | Longer-term stability. |
| 7. Monitor | Notice the risk and self-correct before external feedback is needed. | Ongoing maintenance. |
Teachers and parents often stop at level 2 or 4 because the learner appears to understand. The strongest evidence sits higher: the learner applies the change in a fresh context and eventually detects the problem independently.
Feedback failure in English
English creates many feedback traps because comments often refer to qualities rather than discrete procedures. “Better expression”, “stronger analysis”, “more mature vocabulary” and “more developed ideas” can hide several distinct mechanisms.
- A student told to “elaborate” may add length without adding causal or evidential depth.
- A student told to “use evidence” may quote more but fail to explain relevance.
- A student told to “improve vocabulary” may substitute rare words that distort tone or meaning.
- A student told to “answer precisely” may still not know how command words change answer scope.
The fix is to operationalise the quality, model the decision, practise it in a small fresh task, then return it to authentic reading or writing.
Feedback failure in Mathematics
Mathematics feedback can fail when corrections emphasise final method without the selection cue. A student may understand the worked solution and still fail to recognise the method in a new question.
- “Use simultaneous equations” needs the learner to recognise what relationship makes simultaneous equations appropriate.
- “Check your sign” needs a checking procedure at the point signs commonly change.
- “Show clearer working” needs a representation rule, not prettier handwriting.
- “Be careful” needs a named error mechanism and an intercepting check.
Use fresh mixed problems to verify that the feedback changed mathematical decision-making rather than only the corrected script.
Feedback failure in Science
Science feedback often becomes keyword chasing. A student learns that an answer “needed evaporation” or “needed friction” and inserts the word next time without repairing the causal model.
Useful science feedback checks the relationship: what changes, why it changes and what observable outcome follows. A correct keyword inside an incorrect causal chain is not recovery.
Why repeated feedback can make the problem worse
When the same feedback is repeated without success, adults often increase intensity: more comments, stronger reminders, more frequent corrections. But repetition can make the comment feel like a stable label. “You always rush.” “You never check.” “Your writing is vague.” The learner hears a description of self rather than a changing mechanism.
The correct escalation is diagnostic. If the same comment has been given three times, stop giving the same comment. Test why it is not changing behaviour.
A feedback-failure diagnostic
- Name the repeated feedback. What comment, mark pattern or correction keeps appearing?
- Ask for learner interpretation. What does the learner think it means?
- Ask for the changed behaviour. What should happen differently next time?
- Find the prerequisite. Can the learner actually perform that behaviour?
- Test on the original type. Can the learner correct the old task?
- Test on a fresh task. Does the changed decision transfer?
- Delay and retest. Does the change survive time?
- Remove support. Does the learner notice the problem without being told?
The first point of collapse becomes the repair target.
When the feedback itself is poor
Not every failure belongs to the learner. Feedback can be vague, contradictory, excessive, disconnected from learning intentions, focused on low-value surface features or delivered without a realistic opportunity to act.
The EEF feedback guidance is useful here because it moves attention away from arguments about whether feedback should be written or verbal and toward the principles that make feedback effective. The format is secondary to whether the feedback is grounded in good instruction, addresses a learning gap and enables improvement.
When the feedback is good but the system around it is poor
A teacher may provide excellent feedback while the learner has no time, no second attempt, no retrieval routine and no follow-up. The educational system then treats feedback as an endpoint.
Repair the surrounding loop. Reserve time for response. Require one fresh application. Revisit the target after delay. Reduce the number of simultaneous targets. Track whether support can be faded.
The parent role
Parents can help by asking process questions rather than re-marking the work.
- “What did the teacher want you to change?”
- “What will that look like in the next answer?”
- “Can you show me one fresh example?”
- “Which part can you now do without help?”
- “What will you check for before submitting the next task?”
If the learner cannot answer these, the feedback may need translation before more practice is assigned.
The teacher or tutor role
The adult does not need to write more. Often the higher-value move is to reduce feedback to the smallest decision that can change the next performance, model that decision clearly, create an immediate attempt and observe whether the learner can use it.
When the learner succeeds, fade prompts. When the learner fails, use the failure as evidence. The loop becomes adaptive rather than decorative.
When the learner should generate the feedback
The long-term goal is not permanent dependence on comments. A capable learner increasingly compares work against criteria, notices mismatches, predicts likely errors and chooses a correction before an adult intervenes.
This is why feedback and metacognition meet. The learner gradually internalises the checking questions that were once external.
How to know the feedback loop is working
- The learner can explain the feedback in their own words.
- The learner can name the changed behaviour.
- The original task can be corrected without copying.
- A fresh task shows the same changed decision.
- The improvement survives delay.
- Prompts can be reduced.
- The same feedback appears less often.
- The learner begins to detect the risk independently.
The strongest sign is not that the correction page looks better. It is that the learner needs the same feedback less.
Where to go next
If the failure is that the learner understands the comment but does not convert it into action, continue to How to Translate Feedback Into Change.
If the same mistake keeps returning despite correction, continue to How Error Analysis Fails.
If marks remain flat despite continued effort, use Why Marks Stop Improving.
If you need the broader state-change map, return to the Recovery Atlas.
Feedback-failure control line: performance → evidence → comment/mark → learner meaning → operational change → fresh attempt → variation → delayed retest → fading support. If the loop breaks, repair the first break rather than adding more comments.
Feedback recovery laboratory: test whether the comment changed the learner
The fastest way to discover whether feedback worked is not to reread the correction. Test the next performance. A feedback system should be judged by what the learner can now notice, decide and do with less external support.
The five-minute feedback uptake test
- Hide the marked work. Remove the visual cue of the teacher’s wording.
- Ask the learner to state the main feedback. If it cannot be recalled, uptake is already weak.
- Ask what will change next time. Require an observable action, not a repeated comment.
- Give one fresh micro-task. It should require the same decision in a new context.
- Observe support. Record whether the learner notices the issue independently or waits to be told again.
This small test distinguishes feedback exposure from feedback uptake. It is especially useful when adults feel they are “saying the same thing every week”.
Worked case: “be more specific” in English comprehension
A student repeatedly receives “be more specific”. The comment is accurate, but the next answers remain vague. The teacher increases the number of comments. Nothing changes.
The diagnostic conversation reveals that the student thinks “specific” means “write more details”. The learner therefore produces longer answers. The teacher means “answer the exact inference with evidence that supports it”. Both are using the same word for different actions.
The feedback is translated into a three-step behaviour: underline the exact demand, choose one phrase that indirectly supports it, write one sentence explaining the connection. The learner then practises on a fresh short passage. A delayed passage several days later checks whether the behaviour persists.
The improvement does not come from a better phrase than “be specific”. It comes from establishing shared meaning and an observable action.
Worked case: “check your work” in Mathematics
A Mathematics student loses marks to signs and units. “Check your work” appears on multiple papers. The student insists that checking happens, yet the same errors survive.
The problem is that “checking” has no target. The learner rereads the calculation from top to bottom, often seeing what they expect to see. The feedback is converted into two local rules: pause after any sign-changing algebraic step, and circle the required unit before calculation then compare it with the final line.
The next timed section measures whether those checks happen without prompting and whether the corresponding error frequency falls. Feedback becomes a checking system rather than a reminder.
Worked case: “explain fully” in Science
A Science student is told to “explain fully” and starts writing longer answers. Marks remain flat. The issue is not length. The student names facts but omits the causal mechanism linking changed condition to observed outcome.
The comment is translated into factor → mechanism → effect. The learner first practises orally, then in short written answers, then in changed contexts. The new behaviour is observable: every explanation must contain the relationship, not merely more sentences.
Feedback density: more comments can reduce usable information
A page can contain twenty accurate comments and still produce less learning than one well-chosen target. The learner has limited attention and cannot always coordinate many changes simultaneously.
Choose feedback by leverage. Which change would improve several later decisions? Which error repeats? Which prerequisite is blocking multiple tasks? Which one behaviour can the learner practise now?
Other issues can remain visible without becoming active targets. Feedback is not a requirement to comment on everything that could be improved.
Feedback timing: immediate is not always better, delayed is not always worse
Timing should serve the learning task. Immediate feedback can prevent a misconception from being rehearsed and is useful when the learner has little basis for self-correction. Delayed feedback can require more retrieval and self-monitoring when the learner already has enough knowledge to attempt independently.
The key diagnostic question is whether the timing helps the learner connect evidence to the decision that needs to change. If feedback arrives so late that the learner no longer remembers the reasoning, reactivate the task first. If feedback arrives so quickly that the learner never commits to an answer, delay slightly and allow an independent attempt.
Feedback source: teacher, peer and self-feedback have different jobs
Teacher feedback is valuable when expert diagnosis or modelling is required. Peer feedback can expose the learner to alternative interpretations and audience response when criteria are clear. Self-feedback develops independence when the learner has enough knowledge to compare performance against a standard.
Do not ask a novice to self-assess something they cannot yet recognise. Do not keep expert feedback permanently external once the learner can detect the issue. The long-term direction is from external diagnosis toward increasingly accurate self-monitoring.
The feedback comparison task
One of the strongest ways to make feedback meaningful is to compare two responses. Ask which better meets the criterion and why. The learner must then articulate the difference rather than merely receive it.
- Which answer is more precise?
- Which piece of evidence is relevant?
- Which Mathematics solution makes the state changes easier to inspect?
- Which Science explanation contains a complete causal mechanism?
- Which paragraph develops one idea rather than adding unrelated detail?
Comparison makes the quality criterion visible and can reduce dependence on abstract comments.
The feedback prediction task
Before showing the teacher’s comments, ask the learner to predict the main feedback. This reveals whether the learner can monitor their own work.
If learner and teacher identify the same issue, self-monitoring is developing. If they differ, compare the evidence. The purpose is not to make the learner guess what the teacher wants; it is to build a more accurate internal model of quality.
The response window: protect time to use feedback
Schools and tutoring systems can produce large amounts of feedback without protecting time for response. The learner moves immediately to the next task, so comments accumulate faster than behaviour changes.
Reserve a response window. The learner restates the target, corrects from memory, attempts a fresh item and records the changed rule. A ten-minute response window can be more valuable than another page of comments.
Feedback and emotional load
When a result matters deeply, feedback can be hard to process. A learner may read every comment as confirmation of failure. Another may become defensive and argue about each mark. Another may avoid the paper entirely.
Separate the judgement from the mechanism. “This answer lost two marks” is assessment evidence. “The evidence chosen does not support the inference” is a mechanism. The second statement creates a task that can be changed.
Where confidence has become a separate problem, use Rebuilding Confidence so feedback does not have to carry both technical correction and emotional recovery alone.
When praise becomes unhelpful feedback
Positive feedback can also be too vague. “Good job” may support morale but does not tell the learner what behaviour to repeat. When useful, make praise informational: “Your answer stayed inside the question scope and every sentence supported the inference.”
This helps the learner preserve what works while repairing what does not.
Feedback under examination time pressure
Near an examination, feedback must become more selective. There may not be time to repair every weakness. Prioritise recurring mechanisms that affect accessible marks and can still change quickly.
A tutor should be able to say, “For the next three days we are ignoring these lower-return issues and focusing on this repeated answer-scope problem because it appears across several components.”
Use Examination Triage when the feedback queue is larger than the remaining runway.
Feedback in a retake year
A retake year should produce a different feedback loop from the first attempt. If every paper receives the same comments, the plan is not changing the cause.
- Classify recurring feedback by mechanism.
- Choose one active target at a time.
- Require a fresh application after correction.
- Return after delay.
- Track whether the same comment frequency falls.
- Move stable targets into maintenance rather than keeping them permanently active.
Feedback should become less necessary as capability grows.
The teacher feedback audit
- Is the learning intention clear enough to judge the work?
- Does the feedback identify a meaningful learning gap rather than only surface presentation?
- Can the learner understand the language?
- Can the learner perform the requested change?
- Is one target more important than the others?
- Will the learner get an opportunity to use it?
- Will a fresh task test transfer?
- Will support be reduced when the learner succeeds?
This reflects the core principle in the EEF feedback guidance: effective feedback rests on good instruction, clear intentions and evidence of the gap it is meant to close, not on a preferred delivery format alone.
The learner feedback audit
- What did I think I was trying to do?
- What does the feedback say was missing?
- Do I understand the difference?
- What will I notice next time?
- What will I do differently?
- Can I show it now on a fresh task?
- Can I still do it next week without rereading this page?
Parent feedback audit
Parents do not need to add another layer of marking. Ask whether the child knows the active feedback target and whether a fresh attempt has been made. If the child cannot explain the change, the problem is not solved by adding more reminders.
- “What is the one feedback target you are working on?”
- “What does it look like in a new answer?”
- “What did you do differently this time?”
- “Do you still need someone to remind you?”
Feedback failure patterns in group tuition
Group teaching creates a particular risk: general feedback can be useful to the class but not specific enough for one learner’s mechanism. The tutor may explain a common error while an individual student’s failure has a different cause.
Use group feedback for shared patterns, then verify individual uptake with a short fresh task. The learner’s work decides whether the group explanation solved the problem.
Feedback failure patterns in one-to-one teaching
One-to-one teaching can create the opposite risk: so much immediate feedback is available that the learner rarely has to detect or recover from errors independently.
Allow productive struggle. Delay some feedback. Ask the learner to predict the issue. Use unseen tasks. Reduce prompts. The goal is not a flawless tutoring session; it is independent performance later.
Feedback failure FAQ
Why does my child understand corrections but repeat the mistake?
The learner may recognise the correct answer without changing the decision rule, may fail to retrieve the rule later, or may not recognise the cue in a new context. Use a fresh and delayed task to find where the chain breaks.
Should feedback always be detailed?
No. Detail is useful only when it helps the learner close the relevant gap. Too many simultaneous comments can dilute priority and overload response.
Is verbal feedback better than written feedback?
The useful question is whether the feedback is understood, actionable, used and tested. Delivery method matters in context, but the underlying principles are more important than treating one format as universally superior.
How do I know when to stop giving the same feedback?
If the same feedback has appeared repeatedly without changing performance, stop repeating it in the same form and diagnose why uptake is failing. Once the learner can detect and correct the issue independently across fresh tasks, the target can move into maintenance.
What if the learner disagrees with the feedback?
Return to the criterion and evidence. Ask what the task required, what the learner attempted and what the response demonstrates. Disagreement can be useful when it clarifies the standard rather than becoming a contest of authority.
Feedback should eventually disappear into self-regulation
The mature end state is not a learner who receives perfect feedback forever. It is a learner who increasingly anticipates criteria, monitors work, notices mismatches, chooses a correction and evaluates whether it worked.
That is why feedback belongs inside a larger recovery loop. It begins as external information and, when successful, becomes part of the learner’s own operating system.
Continue to How to Translate Feedback Into Change when the next task is implementation. Use How Error Analysis Fails when the same mistakes survive correction. Return to the Recovery Atlas when feedback failure is one part of a larger recovery problem.
The feedback implementation contract
For a high-value feedback target, write a six-part contract. Criterion: what quality or requirement matters? Gap: what does the current work show? Action: what observable behaviour changes next time? Opportunity: which fresh task will require that behaviour? Verification: what evidence will show the change worked? Fading: what support will be removed when the learner succeeds?
This keeps feedback aligned with the principles in the EEF guidance: feedback follows strong instruction and evidence of a learning gap, then must create a route that moves learning forward. It also connects naturally to metacognitive planning, monitoring and evaluation rather than treating the comment itself as the intervention.
Example: Criterion—an inference answer must connect indirect evidence to the requested attitude. Gap—the learner quotes relevant evidence but restates it literally. Action—write “This suggests…” and explain the attitude implied by the evidence. Opportunity—one fresh short passage. Verification—the new answer makes the inference without prompting. Fading—remove the sentence starter and ask the learner to identify the inference structure independently.
If the contract cannot be written because the criterion or gap is unclear, collect better evidence before adding more comments. If the contract is clear but the behaviour still does not change, test whether the learner lacks a prerequisite, cannot retrieve the rule, or fails to recognise the cue in a new task.
Feedback retirement rule
A feedback target should leave the active list when the learner can recognise the risk, perform the changed behaviour on fresh work, retain it after delay and do so with reduced prompting. At that point, move it into ordinary checking rather than continuing to write the same comment.
The long-term goal is a shrinking external feedback burden because the learner has absorbed more of the monitoring system. Return to Recovery Atlas if repeated feedback failure reveals a wider problem in confidence, habits, subject foundations or examination execution.
Feedback success criterion: the same comment should become less necessary. When the learner can anticipate the criterion, detect the mismatch, choose the correction and verify the new performance independently, external feedback has done its developmental job. Keep feedback specific enough to change the next attempt, and temporary enough that the learner eventually owns the monitoring process.
If the comment remains necessary indefinitely, diagnose the break rather than intensifying the wording. The issue may be meaning, prerequisite knowledge, retrieval, cue recognition, transfer or the absence of a genuine second attempt.
Diagnostic handoff: feedback that does not change later performance may have failed at interpretation, acceptance, memory, action selection or transfer. Return to Student Diagnostics to locate the first break before adding more comments or corrections.
Next action: turn useful feedback into changed performance
Move from diagnosis to Learning Intervention when the feedback mechanism is clear, then use How to Check Whether Learning Recovery Is Working to test whether the next independent attempt actually changes.
Apply the diagnosis in the material the learner is actually studying: Primary English · Secondary English · Primary Mathematics · Secondary Mathematics · Additional Mathematics · Science.
After the repair, use a fresh task rather than the original worked example or correction as the evidence of change.