The teacher explains the error.
The student nods.
They redo the same question perfectly while the correction is still visible.
Has learning changed?
Feedback retesting is the deliberate second measurement used to determine whether a correction changed independent performance, survived delay and generalised beyond the original example.
This is the fourth pillar beneath How Feedback Works. The master already says feedback is complete only when changed action is observed again. This page owns the receipt: what counts as evidence that the correction actually worked?
Quick Read
The first corrected attempt is useful but weak evidence because the learner may still be looking at the feedback, remembering the teacher’s wording or repeating the exact original procedure. Stronger verification comes in layers: redo the target independently, solve a parallel item, use the skill under changed wording or representation, return after a delay, mix it with competing skills and check whether the old error recurs. Retesting should match the original failure. If the problem was method selection, the retest must require selection. If the problem was explanation, the retest must require another explanation. If the problem was cue dependence, the retest must reduce those cues. A feedback loop is trustworthy when the later task shows that the learner can perform the repaired capability without the correction itself doing the work.
error → feedback → learner action → immediate independent attempt → parallel variant → delay → changed context / mixed competition → measure recurrence → close / reopen repair loop
Understanding the Comment Is Not the Same as Owning the Skill
Student says:
I see what I did wrong.
Good.
But can they produce the corrected route without being shown?
Feedback can change declarative awareness before it changes performance.
Retesting separates those states.
The First Retest Should Remove the Feedback Itself
The correction says:
Use a quotation that directly supports the claim, then explain the link.
Do not leave the sentence visible while judging whether the learner can now do it.
Hide or remove the support and ask for a new attempt.
feedback-present performance ≠ feedback-independent performance.
Redoing the Same Question Is the Lowest-Level Receipt
Same question.
No answer visible.
Learner now gets it right.
This shows the correction changed something.
But the original item itself is now a strong cue.
We need a stronger test.
Parallel Problems Test the Mechanism, Not the Memory of the Answer
Original mathematics item:
wrong method selected.
Parallel item:
same underlying relationship, different numbers and context.
If the learner selects correctly, we have stronger evidence the method-condition link improved.
Changed Representation Tests Transfer
Feedback repaired an explanation from a diagram.
Retest using a table or graph.
Feedback repaired a formula procedure.
Retest in a word problem.
Retrieval Transfer owns the broader learning mechanism. Feedback Retesting uses transfer as evidence that the repair did not remain trapped inside the original example.
Delay Is the Durability Test
Immediately after feedback, the corrected representation is highly active.
Tomorrow or next week, that activation has faded.
A delayed retest asks whether the learner can reconstruct the repaired route again.
Retrieval Strength & Storage Strength explains why delayed performance is stronger evidence than immediate fluency.
The Retest Must Match the Error Mechanism
Knowledge unavailable
Retest independent recall after delay.
Concept confusion
Retest with both concepts present and require discrimination.
Wrong method selection
Use a mixed set where the method is not named.
Execution error
Use another problem requiring the same execution step.
Writing structure
Use a new paragraph or essay where the learner must independently recreate the structural repair.
If the retest does not reproduce the original demand, it cannot verify the repair.
Retesting Too Soon Can Produce a False Pass
Teacher explains the answer.
Thirty seconds later, student repeats it.
This proves short-term access.
It does not prove durability.
The first sibling, Feedback Timing owns when the correction arrives. Feedback Retesting owns when enough distance has opened to test survival.
Retesting Too Late Can Waste the Repair Opportunity
Wait until the end of term.
The old error reappears.
Now many other topics depend on it.
A useful feedback loop normally contains:
- an immediate or near-term application;
- a later delayed check;
- continued sampling in mixed work.
One checks understanding.
One checks durability.
One checks recurrence.
Feedback Uptake Comes Before Retesting
If the learner never understood the feedback, a failed retest does not tell us the correction itself was ineffective.
The third sibling, Feedback Uptake, owns whether the learner understood and acted on the signal.
Retesting evaluates the result of that uptake.
Feedback Bandwidth Affects What We Retest
Three corrections were prioritized.
Retest all three at once and failure becomes ambiguous.
Sometimes it is better to verify one mechanism at a time first, then combine them in a richer task.
The second sibling, Feedback Bandwidth, owns the size of the original correction packet.
Recurrence Is Stronger Evidence Than One Failure
One sign error appears once after a week.
Could be a random slip.
The same sign error appears across four later tasks.
The correction did not stabilise the mechanism.
Track error recurrence across tasks rather than treating each paper as isolated.
A Correction Can Work Locally and Fail Globally
The student fixes one paragraph perfectly after feedback.
Next essay repeats the old weakness.
The feedback produced local imitation but not a transferable writing routine.
Retesting should therefore move:
same item → parallel item → changed context → later task.
Teacher Presence Can Become a Cue
The learner performs correctly only when the teacher stands beside them.
The teacher’s question, face or reminder may be functioning as a retrieval cue.
Independent retesting should gradually remove:
- teacher prompts;
- rubric highlights;
- worked solutions;
- feedback language;
- chapter labels.
The learner must carry more of the control loop internally.
Delayed Retesting Improves Confidence Calibration
After feedback, ask:
How likely are you to avoid this error next week?
Then test next week.
The gap between prediction and actual recurrence improves the learner’s model of what “fixed” really feels like.
How Confidence Works remains the calibration owner.
Retesting Is Not Punishment for Being Wrong
If students experience every retest as a penalty, they may hide errors or avoid challenging feedback.
Retesting should be framed as verification:
We made a repair. Now we need evidence that it holds.
This normalises correction as an engineering loop rather than a moral judgement.
Retest Difficulty Should Be Diagnostic
Too easy:
the learner passes because the new item is almost identical.
Too hard:
new complexity introduces unrelated failure.
Good:
the retest changes enough to require independent reconstruction while preserving the same target mechanism.
The Retest Can Be Smaller Than the Original Task
A full essay revealed weak claim-evidence linkage.
You do not need another full essay immediately.
A short paragraph can test the corrected relationship efficiently.
Later, a full essay verifies integration.
Retesting should use the smallest instrument that can answer the current verification question.
A Feedback Receipt Should Name the Before and After State
Before:
Chooses cosine rule whenever three lengths appear; does not check whether opposite angle/side conditions fit.
Feedback:
Compare available information with method conditions before selecting.
Retest:
5 mixed triangle items, no method labels, 4/5 correct selections; one competitor error remains.
Next action:
Contrast sine vs cosine conditions; retest after 3 days.
The feedback loop now has evidence.
A Repair Can Graduate From Active Feedback
After several clean delayed/mixed performances, the correction no longer needs explicit teacher commentary every time.
It moves into normal maintenance and self-checking.
Revision Stopping Rules owns the broader graduation logic.
Repeated Failure Means Reopen Diagnosis, Not Repeat the Same Comment Forever
Feedback given.
Retest fails.
Feedback repeated.
Retest fails again.
Do not assume the learner needs a louder version of the same message.
Reopen the chain:
- Was the diagnosis wrong?
- Was feedback not understood?
- Was the skill prerequisite missing?
- Was the retest misaligned?
- Is interference causing recurrence?
- Was there insufficient practice?
Retesting Builds Self-Regulation When Learners Predict the Result
Before retest:
What error are you watching for?
After retest:
Did the old error return? If not, what did you do differently?
The learner starts generating internal feedback instead of waiting for the teacher to rediscover every pattern.
A Better Feedback-Retesting Model
feedback → learner explains change → immediate independent application → parallel variant → delayed retrieval → changed representation / mixed competition → recurrence log → close loop or reopen diagnosis
A 30-Lens Feedback Retesting Audit
- Original error: what mechanism failed?
- Feedback: what correction was provided?
- Uptake: did learner understand it?
- Support: is feedback removed during retest?
- Same item: can original task now be completed?
- Parallel item: does mechanism survive a new example?
- Wording: can the prompt change?
- Representation: can form change?
- Selection: must learner choose among methods?
- Interference: is nearest competitor present?
- Delay: does performance survive time?
- Latency: is the corrected response available quickly enough?
- Accuracy: is the new route correct?
- Explanation: can learner state why it is correct?
- Transfer: can the correction generalise?
- Recurrence: does old error return later?
- Frequency: how many clean performances?
- Confidence: does learner predict stability accurately?
- Teacher cue: is performance dependent on teacher presence?
- Rubric cue: does highlighted feedback still carry the answer?
- Test size: is the retest larger than necessary?
- Difficulty: is it diagnostic rather than unrelatedly hard?
- Immediate check: was correction applied once?
- Delayed check: did it survive after activation faded?
- Mixed check: does it survive competition?
- Paper integration: does it survive full-task conditions?
- Graduation: can explicit feedback stop?
- Re-entry: what recurrence reopens repair?
- Diagnosis reopen: when should repeated failure trigger a new hypothesis?
- World return: does the corrected behaviour reappear independently when the learner later needs it, without the original comment present?
Laboratory 1: Same → Parallel → Delayed
After one correction, use three retests: the same item without feedback, a parallel item, then a delayed item several days later. Compare what each layer reveals.
Laboratory 2: Retest the Mechanism
Choose one recurring error and write a retest that reproduces the deciding mechanism without copying the original surface form.
Laboratory 3: Recurrence Ledger
Track one repaired error over five later tasks. Record clean performances and recurrences. Decide when the repair can leave active feedback and when it must reopen.
For Primary Readers
If someone shows you how to spell a word, writing the same word once is a start. Writing it correctly tomorrow inside a sentence shows much more.
For Secondary Readers
Explain why redoing the exact corrected question is weaker evidence than solving a parallel problem after a delay and design a three-stage feedback retest.
For Advanced Readers
Model feedback retesting as causal verification of a proposed intervention. The retest must preserve the target mechanism while progressively removing correction-associated cues and introducing temporal or contextual variation, allowing observed performance change to be attributed more plausibly to learning rather than immediate imitation.
Common Misconceptions
- “If the learner can redo the question, the feedback worked.” The same item can provide strong memory cues.
- “A delayed retest is unnecessary once immediate performance is correct.” Immediate success can reflect high temporary accessibility.
- “Retesting means giving another full assignment.” Use the smallest task that can verify the target mechanism first.
- “Any new question tests transfer.” The retest must preserve the relevant underlying demand.
- “Repeated failure means repeat the same comment.” It may require a new diagnosis.
Research Corridor
- Education Endowment Foundation — Teacher Feedback to Improve Pupil Learning — feedback should be followed by opportunities for pupils to use it.
- US Institute of Education Sciences — Organizing Instruction and Study — retrieval and spaced learning as durable performance checks.
- Bjork Learning and Forgetting Lab — delayed retrieval and performance-versus-learning distinctions.
- eduKateSG — How Feedback Works.
Frequently Asked Questions
How should teachers check whether feedback worked?
Ask the learner to perform the corrected capability again without the feedback visible, then use a parallel or changed item and later a delayed retest.
Why not just repeat the same question?
Because memory for the original question and correction can support performance. A parallel item provides stronger evidence that the underlying mechanism changed.
What if the error returns later?
Reopen diagnosis. Determine whether the correction was not durable, the feedback was misunderstood, interference emerged or the original diagnosis targeted the wrong mechanism.
Final Thought: Feedback Needs a Receipt From the Future
The correction is not proven when the learner can follow it once. It is proven more convincingly when the learner later reconstructs the better action after the comment, teacher and original example have disappeared.
FEEDBACK · FOUR PILLAR LEGS
Return to How Feedback Works, or continue through Feedback Timing, Feedback Bandwidth and Feedback Uptake. Return to the How X Works Hub.