Feedback is useful only when it changes what the learner can do next.
“Wrong.” “Careless.” “Revise this.” “Good job.” These comments may communicate something, but they do not necessarily improve the learner’s mathematical state. At the other extreme, a full worked solution can remove so much uncertainty that the learner no longer has to diagnose, choose or recover anything themselves.
Mathematical feedback therefore has a difficult job. It must reduce the right uncertainty without taking away the thinking that the learner still needs to do.
How Mathematical Feedback Works is BTT’s public feedback layer. It does not replace How Mathematics Diagnosis Works, the Mathematics HELP Runtime, How Mathematical Practice Works, How Mathematical Transfer Works, How Independent Mathematics Works, or the existing Diagnostic and Feedback Technology in Mathematics. Those remain current owners. This page explains the general rule that connects them:
Feedback should reveal enough to restore useful mathematical movement, but not so much that it becomes the movement.
Feedback is not the same as correction
A correction tells the learner what is wrong or supplies what is right. Feedback goes further: it helps determine what should happen next.
| Response | What it tells the learner | What may still be missing |
|---|---|---|
| “Wrong answer” | The final result is not accepted | Where the first break occurred |
| Correct solution shown | A valid route exists | Whether the learner can reconstruct it |
| “Check your signs” | A likely error class | Which sign decision failed and why |
| “Your first three lines are correct” | The trustworthy boundary | What decision should be inspected next |
| “What quantity is 100% here?” | A discriminating question | The learner still owns the next step |
The strongest feedback often narrows the search space rather than giving the answer.
The first job: preserve what is already correct
When a solution fails, learners often assume the whole attempt is useless. Good feedback should identify the last trustworthy point.
Suppose a learner sets up a correct equation, expands correctly, then makes a sign error while collecting terms. The useful message is not “Your algebra is wrong.” Most of the algebra may be right.
A stronger response is:
Your equation and expansion are valid. Recheck the transition from line 3 to line 4: what happened to the negative term?
This does three things at once. It protects correct knowledge, identifies the boundary where trust ends, and gives the learner a next action without completing it for them.
The second job: distinguish symptom from cause
A wrong answer is a symptom. Feedback becomes more useful when it responds to the mechanism that produced it.
The same final error can come from different causes:
- a prerequisite is missing;
- a representation was misread;
- a relationship was misunderstood;
- a correct method was executed inaccurately;
- the wrong method was selected;
- a restriction was ignored;
- task language was misinterpreted;
- or examination pressure disrupted otherwise available Mathematics.
That is why feedback and diagnosis are linked. If the reason for failure is unclear, the next response should often be a probe rather than an explanation.
Use How Mathematics Diagnosis Works when a recurring error needs to be traced to the first useful weak link.
The feedback ladder
BTT can treat feedback as a ladder. Start with the smallest intervention that could plausibly restore productive movement. Escalate only when the evidence says the smaller response was insufficient.
| Level | Feedback move | Example |
|---|---|---|
| 1 | Wait | Give the learner time to detect the issue independently |
| 2 | Verify boundary | “Everything is correct up to this line.” |
| 3 | Ask | “What quantity does this percentage refer to?” |
| 4 | Reframe | “Can you draw or tabulate the relationship first?” |
| 5 | Cue | “Compare the units on both sides.” |
| 6 | Hint | “The gradient connects change in y to change in x.” |
| 7 | Model one local move | Demonstrate only the broken transition |
| 8 | Teach | Rebuild the concept if smaller feedback cannot restore it |
This mirrors the logic of the existing Mathematics HELP Runtime: minimum justified help first, larger intervention only when necessary.
Feedback should point forward, not merely backward
Looking backward tells us what failed. Looking forward tells the learner what to do next.
Compare these two comments:
“You forgot to use the correct base.”
“Before calculating, mark the quantity that represents 100%. Then rebuild the percentage relationship.”
The second comment creates an action that the learner can reuse in another question.
Good feedback therefore produces a small piece of future control.
Feedback should sometimes ask, not tell
A well-chosen question can be more useful than a long explanation because it preserves the learner’s need to retrieve and decide.
Useful feedback questions include:
- What does this symbol represent here?
- Which line is the last one you still trust?
- What changed between these two steps?
- What condition must be true before this formula applies?
- Can you represent the same relationship another way?
- What would you expect the answer to be approximately?
- Which unit should the final quantity have?
- If your answer were correct, what else should also be true?
The best question depends on the diagnosis. A generic question such as “Are you sure?” can create anxiety without creating information.
Feedback must not create confirmation dependence
Frequent immediate feedback can accidentally teach the learner to outsource verification.
If a tutor confirms every line, the learner may stop developing internal checks. If a parent marks each homework answer the moment it is written, the learner may stop asking whether the result is plausible. If AI instantly evaluates every step, the learner may never learn to hold uncertainty long enough to inspect their own work.
As capability grows, feedback should sometimes be delayed.
A useful progression is:
- Immediate feedback during first acquisition.
- Feedback after one completed step or subproblem.
- Feedback after an independent attempt.
- Feedback after self-checking.
- Feedback only after a changed question or mixed set.
- Feedback after an examination section or full paper.
The goal is not to withhold useful information. It is to give the learner enough space to become part of the feedback system themselves.
Feedback must make a repair testable
A correction is not complete because the learner says, “I understand now.” The feedback should produce a prediction that can be tested.
If the feedback identifies a weak percentage base, the next question should change the surface but preserve the same base-selection demand. If the feedback identifies a sign problem, the learner should face another question where sign handling is embedded rather than isolated. If the feedback identifies a graph–equation translation problem, the learner should translate in the opposite direction.
This is where feedback hands off to How Mathematical Transfer Works.
The feedback loop
Observe → Locate → Classify → Respond → Retry → Change → Verify → Fade.
Each step has a different job:
- Observe: collect the learner’s actual working.
- Locate: identify the first point where trust breaks.
- Classify: decide what kind of break is plausible.
- Respond: give the minimum useful information.
- Retry: let the learner repair the original task.
- Change: alter the question enough to test ownership.
- Verify: check whether the predicted improvement appears.
- Fade: reduce feedback if the learner can now self-monitor.
The loop can reopen whenever new evidence contradicts the first interpretation.
Feedback for a correct answer can still matter
Correct answers are not always evidence of a stable route.
A learner may guess correctly, use an inefficient method, rely on an invalid assumption that happened not to matter, or reach the right answer after an unrecognised error cancellation.
Useful feedback on a correct answer may ask:
- Why is this method valid?
- Can you show another route?
- Which step carries the main mathematical idea?
- Would this still work if the numbers changed?
- What restriction have you used implicitly?
- How would you verify the result?
Feedback is therefore not only an error service. It can deepen and test successful Mathematics too.
Feedback for Primary Mathematics
Primary feedback should preserve the connection between numbers and meaning.
- Ask what each number represents.
- Ask the child to show the relationship with a model or diagram.
- Use estimation to challenge implausible answers.
- Distinguish operation mistakes from misunderstanding the story.
- Ask what changed when the unknown moves.
- Encourage self-checking before revealing correctness.
The aim is to prevent “right/wrong” from becoming the only information the child receives about Mathematics.
Feedback for Secondary Mathematics
Secondary feedback should increasingly expose mathematical structure.
- Mark the last trustworthy algebraic line.
- Identify whether an equivalence-preserving step failed.
- Ask the learner to translate between graph and equation.
- Distinguish a prerequisite error from a current-topic error.
- Ask whether the method’s conditions actually hold.
- Require enough working that feedback can target a decision rather than only a final answer.
Use How SEC Mathematics Works as the existing Secondary stage owner.
Feedback for Additional Mathematics and JC
At higher levels, feedback should increasingly address strategy, assumptions and mathematical judgement.
- Why was this representation chosen?
- Which function or quantity should be defined first?
- What domain or restriction must be preserved?
- Is the chosen route efficient enough for the examination context?
- What does the derivative, integral, vector or probability result mean?
- What independent check is available?
The feedback target moves from local execution toward whole-solution control.
Feedback under examination conditions
Examination feedback should separate mathematical knowledge from examination execution.
After a paper, classify lost marks before deciding what to practise:
| Lost mark type | Possible next action |
|---|---|
| Knowledge absent | Relearn or repair prerequisite |
| Method not recognised | Interleave and transfer |
| Method known but execution failed | Stabilise mechanics and checking |
| Question misread | Task-language and representation practice |
| Time ran out | Pacing, sequencing and method efficiency |
| Answer form lost marks | Command-word and final-response discipline |
| Correct work overwritten | Verification and decision discipline |
Mathematics Examination Craft remains the current owner for the paper-performance layer.
Feedback from parents
Parents do not need to become Mathematics tutors to give useful feedback.
Instead of immediately correcting the Mathematics, a parent can ask:
- Which part are you sure about?
- Where did you first become uncertain?
- What is the question asking for?
- Can you check whether your answer is reasonable?
- What would you try before asking for the full solution?
This preserves the parent’s role as a supporter without requiring them to carry the mathematical route.
For the wider decision layer, use Parents’ Guide to Mathematics | What Should I Do Next?.
Feedback from tutors
In a three-student tutorial, the value of small-group teaching comes partly from feedback resolution. The tutor can see enough working to respond to the actual break rather than deliver one general explanation to a large group.
That visibility should produce a chain:
See the work → identify the state → choose the smallest useful response → return the problem → observe what changed.
The tutor should not become the student’s permanent error detector. As the learner improves, more of the checking loop must be handed back.
Feedback from AI and digital tools
AI can generate explanations, hints and evaluations quickly. That speed is useful, but it creates a control problem: the tool can provide more information than the learner should receive.
Useful AI feedback should ideally preserve several boundaries:
- Do not reveal a full solution when one discriminating question is enough.
- Separate checking from generation.
- Ask the learner to show their current working first.
- Identify the first break rather than rewrite everything.
- Make the learner retry after feedback.
- Use a changed question to test whether the repair transferred.
- Allow an AI-WITHOUT condition when independence is being measured.
BTT’s existing Diagnostic and Feedback Technology in Mathematics remains the technology-specific owner.
When feedback should stop
A response system can become so active that it prevents the learner from experiencing enough uncertainty to build judgement.
Feedback can be reduced when the learner can:
- identify the last trustworthy step independently;
- classify common error types;
- choose a useful check;
- recover without immediate confirmation;
- solve a changed question after repair;
- and explain why the corrected method is valid.
This is where feedback hands off to How Independent Mathematics Works.
Where this page routes next
| Need | BTT owner |
|---|---|
| I need to identify the first weak link. | How Mathematics Diagnosis Works |
| I need the correct level of intervention. | Mathematics HELP Runtime |
| I need to practise the repaired capability. | How Mathematical Practice Works |
| I need to test a changed question. | How Mathematical Transfer Works |
| I need to validate whether the repair held. | BTT Mathematical Lab |
| I need to know when feedback can reduce. | How Independent Mathematics Works |
| I need examination-specific feedback. | Mathematics Examination Craft |
| I need technology-specific feedback design. | Diagnostic and Feedback Technology in Mathematics |
The destination of feedback
The best feedback does not create a learner who waits for increasingly precise corrections.
It creates a learner who increasingly performs those functions internally: notices a mismatch, locates the break, asks a discriminating question, chooses a repair, tries again, checks the result and decides whether outside help is actually necessary.
Feedback is complete when part of the feedback loop has become the learner’s own mathematical judgement.
BTT feedback principle: preserve what is correct, locate the first useful break, respond with the minimum information that changes the next action, make the learner retry, test the repair on changed work, delay and fade feedback as self-monitoring strengthens, and keep returning control to the learner.