The Tutor Handbook · Volume 0129 · Series ID THB-0129
A tutor notices an error while the learner is still working. There are at least three plausible moves. Correct it immediately. Let the learner finish and return to it later. Or create a short opportunity for the learner to notice and revisit the work before receiving the answer. Each can be right. Each can also be damaging when used as a reflex.
The decision is not “Is immediate feedback good?” or “Is delayed feedback better?” Educational research does not support a universal timing rule that can be lifted intact into every three-student tuition situation. The useful question is narrower: what information does the learner need now for the next action to remain educationally productive, and what information is better withheld briefly because the learner still has a useful chance to detect, test or repair the issue?
This volume calls that judgement the Feedback Timing Gate. It does not replace the broader owner How Feedback Works in Teaching, nor The Feedback Receipt or The Correction Load Gate. Those own what feedback is, whether it changes the next attempt and how much correction to deliver. This article owns one decision: when to intervene.
Why timing changes the educational job of feedback
The same sentence from a tutor can do different work depending on when it arrives.
“Check the sign before the bracket” said one second after a learner writes the line can stop an error from propagating through six more steps. That may protect working memory and prevent needless practice of a wrong procedure. The same prompt, however, can also remove the learner’s opportunity to monitor the line independently. If every sign error is caught at once, the learner may become accurate only while the tutor is present.
At the other extreme, waiting until the end can preserve independent performance but allow a simple misconception to contaminate the rest of the task. A learner may spend ten minutes building on a premise that the tutor could have corrected in twenty seconds. The delayed feedback has not strengthened independence; it has merely allowed unproductive error to accumulate.
There is therefore no virtue in delay by itself. The educational value lies in the work that the delay makes possible. A short delay can create space for self-monitoring, comparison, retrieval or error detection. A long delay can create forgetting, confusion or unnecessary practice. An immediate correction can protect the learning route; an immediate correction can also steal the very monitoring process the learner needs to own.
The gate asks what the learner would do during the interval. If the answer is “nothing useful except continue from a false premise,” intervene. If the answer is “inspect, compare, test or complete a thought that reveals whether they can monitor independently,” a brief delay may be valuable.
Start with task purpose, not a house rule
A tutor should never use one timing policy for teaching, practice, diagnosis and verification.
During initial teaching, immediate corrective guidance can be appropriate because the goal is to build an accurate representation and prevent the learner from practising a newly introduced idea incorrectly. During guided practice, the tutor may still intervene relatively quickly, but should increasingly create room for the learner to detect problems. During independent practice, constant live correction can convert an independent attempt into guided work. During diagnosis, feedback before the evidence has done its discriminating job can destroy the information the tutor was trying to collect. During verification, the tutor generally needs to preserve the target conditions until the attempt is complete.
This is why The Task-Purpose Gate must sit upstream. A timing rule that is excellent for instruction can invalidate a diagnostic check. A delay that is useful for verification can be cruel and pointless during first exposure.
The first question is therefore not “How long should I wait?” It is “What job is this attempt doing?” Once that is clear, timing can be judged against the purpose rather than against a fashionable slogan.
A three-part timing test
Before intervening, the tutor can ask three questions.
Will the error poison what comes next? Some errors are upstream. If left alone, they make the rest of the task uninterpretable or force the learner to practise an invalid route repeatedly. These favour earlier intervention.
Does the learner still have a realistic route to notice or test the error? If the learner has criteria, a representation, a checking method or enough knowledge to inspect the work, a short self-correction window can be useful. If the missing knowledge is precisely what would be required to notice the error, waiting becomes theatre.
What evidence do I need from this attempt? If the tutor needs to know whether the learner can catch the issue independently, intervening immediately destroys that evidence. If the evidence has already been collected and the next priority is learning, the tutor can teach.
These questions turn timing into a causal decision rather than a stopwatch setting.
Immediate feedback is most useful when the route is about to become unproductive
Imagine a learner practising a newly taught method for solving simultaneous equations. On the first independent line, the learner copies a coefficient incorrectly. If the tutor waits through the entire solution, most of the remaining work may be mathematically coherent but irrelevant because it follows the wrong equation. The learner could spend several minutes practising arithmetic on a corrupted problem.
Here, a brief immediate prompt may be the higher-value move: “Compare that copied line with the original equation.” This does not supply the solution. It protects the route from a transcription error that would otherwise swamp the target practice.
The same logic applies in writing when a learner has misunderstood the task demand so completely that every paragraph is being built for the wrong audience, or in science when a variable has been misidentified and the rest of an experimental plan therefore answers a different question. Early intervention prevents a local mistake from consuming the whole practice opportunity.
AERO’s Monitor progress, updated 14 May 2026, explicitly frames checking for understanding as a way to identify gaps and respond with additional instruction, guidance or feedback where necessary. That school-oriented guidance does not tell a private tutor exactly when to interrupt, but it supports the broader principle that feedback should be responsive enough to keep instruction aligned with learner need.
Delay is useful only when it creates productive learner work
A tutor may deliberately wait when the learner is still in a position to detect the problem. The interval should have an educational purpose.
For example, a learner finishes a paragraph with a pronoun whose reference is ambiguous. Instead of immediately marking the phrase, the tutor asks the learner to read the paragraph once as an editor and underline any sentence where a reader might have to guess who or what is being referred to. If the learner catches the ambiguity, the tutor has learned something important: production and monitoring are not identical capabilities. If the learner misses it, the tutor now has evidence about the monitoring gap and can teach from there.
The wait has value because the learner is doing a defined act of inspection. “I will stay silent and see if they magically notice” is not the same thing.
EEF’s current Metacognition and Self-Regulated Learning guidance, second edition published 13 November 2025, describes explicit support for planning, monitoring and evaluating learning. The associated evidence base concerns school teaching and does not validate a particular delay interval in tuition. It does, however, support designing moments in which learners actively monitor their own work rather than receiving continuous external regulation.
The self-correction window is not a guessing game
A brief revisit before feedback can be powerful, but tutors can misuse it by repeatedly saying “Are you sure?” until the learner abandons a correct answer or starts guessing.
A useful self-correction window has a legitimate basis. The tutor may direct attention to a criterion the learner is expected to own: “Check whether every term has the same unit.” Or ask for a comparison the learner has already been taught to make: “Test your answer against the diagram.” Or invite a full reread under a stated purpose: “Read it once for tense consistency.”
The tutor should not use facial expressions, tone or repeated scepticism as hidden answer signals. If the learner has no reasonable method for checking, the tutor should teach rather than stage a ritual of doubt.
This distinction protects confidence in a subtle way. Learners need to know that being asked to check does not necessarily mean the answer is wrong. Otherwise “check again” becomes a coded correction, and the tutor is still supplying the answer through social signalling.
Composite case: the sign error that should be interrupted
The following case is fictional and uses constructed details.
Alicia is working through an algebra problem in a three-student tutorial. The target is not sign manipulation; the target is choosing and executing a substitution route. She correctly selects the substitution, but while copying the second equation she changes -3y to +3y.
The tutor has several options. Waiting until the end would preserve an untouched attempt, but almost every later line would now be generated from corrupted data. The eventual wrong answer would say little about substitution. The tutor therefore intervenes early: “Before you continue, compare the equation you copied with the question.”
Alicia spots the sign and repairs it. The tutor records this as a transcription correction during practice, not as evidence that the full solution was independent. On a later fresh item, Alicia works without interruption.
The timing protected the target practice while keeping the claim honest. Immediate feedback was useful not because immediate is always better, but because the error was incidental to the target and would have made the remaining work educationally expensive.
Composite case: the inference that should be allowed to stand briefly
Beatrice answers a reading-comprehension question with a plausible inference but weak textual support. The tutor can already see the problem. Instead of correcting it at once, the tutor asks Beatrice to finish the next two questions and then run a “support check”: for each inference, identify the words that make it more likely than a competing interpretation.
When Beatrice returns, she changes the first answer herself because she cannot locate evidence. That tells the tutor something different from an immediate correction. She may be capable of evaluating evidence when prompted to monitor, even if she does not yet initiate that monitoring automatically.
The next instructional job is therefore not simply “teach inference.” It may be to make the support check part of the learner’s own routine and later test whether it occurs without prompting.
The delay created evidence about self-regulation. If the tutor had corrected the answer immediately, that information would have been lost.
Composite case: delayed feedback that became neglect
Ciara is learning a new sentence structure and repeatedly applies the same incorrect punctuation rule. The tutor believes in independence and decides not to interrupt. Ciara completes fifteen sentences. At the end, the tutor marks the repeated error and explains the rule.
This is not a triumph of productive delay. Ciara did not have the knowledge needed to detect the mistake, and the repeated practice strengthened a faulty routine. A better route would have been to stop after the first or second recurrence, teach the distinction, then give fresh items.
The lesson is important because “let them struggle” can sound intellectually respectable while being pedagogically lazy. Productive struggle requires a problem the learner can meaningfully work on with available knowledge. Repeatedly performing a rule that has not been understood is not automatically productive.
Feedback timing in a three-learner group
Small-group tutoring adds a complication: feedback to one learner changes the information available to the others.
If the tutor immediately corrects Denise aloud, Emily and Faith may hear the answer before completing their own attempts. Their work is no longer an independent sample. The timing decision therefore has a group dimension.
One option is silent or private marking while others work. Another is to let all three commit to an answer before collective feedback. Another is to choose a different but equivalent follow-up item for learners who were exposed to the correction. The appropriate method depends on whether the current task is teaching, practice or evidence collection.
The Peer Answer Leakage Boundary owns the broader leakage problem. Here the key point is that “immediate” for one learner can mean “premature” for another. A tutor should time feedback not only against the recipient’s need but also against the group’s unfinished thinking.
What must be corrected immediately
Some situations justify very fast intervention even when a tutor usually prefers learner monitoring.
Correct promptly when continuing would create a material safety issue, a serious factual misconception with harmful implications, or an inappropriate educational claim that should not be rehearsed as true. Correct promptly when a wrong premise makes the rest of the practice meaningless. Correct promptly when the task is explicitly guided practice and the learner is rehearsing a newly taught routine in a way likely to consolidate the wrong sequence. Correct promptly when the learner is visibly stuck in a loop with no new information being generated.
“Promptly” does not always mean “give the answer.” A well-chosen cue can stop the unproductive route while preserving some learner thinking. The least intrusive useful intervention is often preferable to a complete solution.
Educational advice should remain within educational scope. If a question becomes medical, psychological, safeguarding-related or otherwise requires professional expertise beyond tuition, the tutor should not improvise a domain diagnosis under the banner of feedback.
What can often wait
Feedback can often wait until a natural boundary when the current error does not corrupt the rest of the work and the learner can benefit from completing the thought.
Examples include a stylistic weakness in a first draft when the current target is idea generation; a non-critical arithmetic slip during a diagnostic task whose purpose is to observe route selection; an imprecise word choice that can be addressed during a later editing pass; or a minor notation issue when interrupting would break a complex chain of reasoning.
Waiting until a boundary can reduce fragmentation. It allows the tutor to respond to the pattern rather than to every local imperfection. It can also reveal whether the learner repeats, notices or compensates for the error later.
But delayed feedback should have a return path. “I will deal with that later” is only a plan if later is defined. The tutor may mark the point, note it briefly, or schedule a revision pass. Otherwise delay easily becomes omission.
Feedback latency and memory
The longer the delay, the greater the risk that the learner no longer remembers what they were thinking when the error occurred. This matters because feedback is often most useful when it can connect outcome to decision.
A comment returned days later may still be useful for stable features such as structure, recurring grammar patterns or conceptual misunderstandings, but the tutor may need to reconstruct the learner’s reasoning. “What were you trying to do here?” becomes important. Without that reconstruction, the tutor can end up correcting the product while missing the process that produced it.
The practical implication is not “feedback must always be immediate.” It is that delayed feedback should preserve enough trace of the attempt for learner and tutor to reconnect the comment to the decision. This might be the original working, a short note, a marked decision point or a saved draft. The Record Minimum sets the broader proportionality rule.
Timing is different from correction load
A tutor can give very little feedback immediately or a great deal later. Timing and amount are separate variables.
This matters because a tutor who realises that continuous interruption is harmful may swing to the opposite extreme and store up twenty comments for the end. The learner then receives an unusable wall of feedback. That is a load problem, not a timing solution.
A good timing decision therefore passes responsibility to the Correction Load Gate. Once the tutor decides “now,” the next question is how much information the learner can use. Once the tutor decides “later,” the same question remains. Delaying ten corrections does not make ten corrections easier to absorb.
The two gates should work together: time the intervention so that it protects useful thinking, then size the feedback so that it can change the next attempt.
Timing can also be wrong because the tutor is reading emotion instead of learning need
There is another trap in live tuition: the tutor times feedback according to discomfort rather than instructional purpose. A learner frowns, hesitates or looks disappointed, and the tutor rushes in to rescue. Another learner appears calm and confident, so the tutor waits even though the work is drifting badly. Neither expression is a reliable measure of what feedback timing should be.
The tutor should remain humane without treating visible ease as competence or visible strain as evidence that help must arrive immediately. A learner may be thinking hard and productively. Another may be confidently applying a false rule. The timing gate therefore returns to observable features of the task: whether the route remains productive, whether the learner has a usable checking method, whether the attempt is supposed to produce independent evidence, and whether delay is generating information or merely extending error.
This is especially important with younger learners and learners who have had repeated experiences of failure. A tutor can preserve dignity by explaining the routine in advance: “Sometimes I will stop you quickly because the next step depends on this. Sometimes I will let you finish because I want you to have a chance to check your own work. Silence does not mean I think you are wrong, and a prompt does not mean you have failed.”
That framing reduces the social guessing game around tutor behaviour. It also makes the instructional contract clearer. Feedback timing becomes part of how responsibility is transferred, rather than an unpredictable reward or punishment tied to the tutor’s facial expression.
A timing decision should remain reversible
When the tutor is unsure, prefer a timing move that keeps the next decision open. A brief cue can stop destructive drift without supplying the whole route. A short note can preserve an error for later discussion without interrupting the current chain of thought. A fresh follow-up can restore evidence lost by an earlier prompt.
This matters because live tutoring rarely offers perfect information. The tutor does not need certainty before acting, but should avoid making one ambiguous moment carry more weight than it can. Good timing is often a small, reversible intervention followed by another chance to observe.
The timing gate across Repair, Alignment and Frontier
In Repair mode, earlier feedback is often useful when the tutor is rebuilding a fragile component and wants accurate practice. Yet the tutor should still create small monitoring opportunities so the learner does not become dependent on live correction.
In Alignment mode, the tutor balances current curriculum flow with error correction. Feedback may be grouped at natural boundaries so the learner can complete school-relevant work without constant interruption, unless an upstream error would make the rest invalid.
In Frontier mode, more delay can sometimes be productive because the learner is exploring an unfamiliar problem and needs room to generate, test and reject ideas. Premature correction can collapse exploration into imitation. Still, frontier work is not an excuse to leave the learner in indefinite confusion. The tutor monitors whether new information is being generated.
These are functions of the tuition job, not permanent learner classifications, and they do not create a fourth mode called Performance.
A practical timing matrix
The following is a decision aid, not a validated scale.
Intervene now when the error is upstream, repeated under a newly taught rule, safety-relevant, or likely to consume substantial practice on a false premise.
Create a short revisit first when the learner has a plausible checking method and the monitoring process itself is part of the learning goal.
Wait to a natural boundary when the error is local, does not corrupt later work, and interruption would damage a larger chain of reasoning or composition.
Hold feedback until the attempt is complete when the purpose is diagnosis or verification and help would alter the evidence.
Return later with reconstruction when the work needs distance—for example, a full writing draft—but preserve enough of the original reasoning to make the later comment intelligible.
The categories are not rigid. One task can move between them as its purpose changes.
What tutors should record
Timing decisions only need recording when they affect how the evidence should be interpreted or what happens next.
A useful note might say: “Prompted sign check on Q2 during guided practice because transcription error would invalidate substitution route. Do not count Q2 as independent. Q5 completed fresh without prompt.” Another might say: “Held inference feedback until evidence-check pass; learner self-corrected two of three. Next session test whether evidence check is initiated without cue.”
These records distinguish supported success from independent success and preserve why the tutor intervened when they did.
They also prevent a later tutor or parent from reading a corrected page as if it were an untouched sample.
What parents should hear
Parents often see marked work and ask why the tutor did not correct every mistake immediately. The answer should be concrete rather than ideological.
A tutor can explain: “Some errors I stop quickly because they would make the rest of the work unhelpful. Others I leave briefly because your child has a real chance to detect them using a checking method we are trying to make independent. If I correct everything the instant it appears, I can make the page look better while learning less about what your child can monitor alone.”
That explanation makes the purpose visible. It also reassures parents that delayed feedback is not inattention.
Delayed and changed-condition checks
A timing decision should eventually be tested by what the learner can do without the tutor’s timing support.
If immediate prompts were used during repair, later work should remove or fade them. If self-correction windows were used, later checks should see whether the learner initiates the relevant monitoring without being told. If end-of-task feedback was used, a fresh task should show whether the feedback changed the next performance rather than only the corrected page.
This is where the Feedback Receipt becomes essential. Good feedback is not proved by a good conversation about feedback. It is shown by a better next attempt under conditions appropriate to the target.
Research boundaries
The research base on feedback is large, but “feedback” covers many designs, ages, subjects, timings and information types. AERO’s Formative assessment: Practice guide, last updated 7 July 2026, summarises evidence that formative assessment and well-used feedback can support learning, while also noting contextual variation. It cites a large feedback meta-analysis and systematic reviews, but those do not establish one universal delay interval for a Singapore private tutor.
EEF’s metacognition guidance supports explicit planning, monitoring and evaluating, but it is guidance for school contexts and should not be converted into a claim that withholding feedback is always metacognitively superior. NSSA’s Tutoring Quality Standards distinguish research-based, research-informed and emergent recommendations; their formative-assessment standard supports using evidence to inform future sessions, not a precise second-by-second feedback protocol.
The Feedback Timing Gate is therefore a research-informed tutoring decision framework, not a validated measurement tool. Its purpose is to prevent timing from becoming habit.
The return: feedback should arrive when it can still change the right thing
The best moment for feedback is not automatically the earliest moment or the latest moment. It is the moment at which the information can improve the learner’s next useful decision without unnecessarily taking that decision away.
Sometimes that is now. Sometimes it is after one more line. Sometimes it is after the learner has checked. Sometimes it is after the full attempt. The tutor’s job is to know what the interval is buying.
If the delay creates productive monitoring, use it. If the delay only multiplies confusion, intervene. If immediate feedback protects the route, give it. If immediate feedback would perform the learner’s monitoring for them, create space first.
Return to The Tutor Handbook | Complete Series Index.
Research and guidance consulted
- Australian Education Research Organisation — Monitor progress, published 26 February 2024 and last updated 14 May 2026.
- Australian Education Research Organisation — Formative assessment: Practice guide, last updated 7 July 2026.
- Education Endowment Foundation — Metacognition and Self-Regulated Learning, second edition published 13 November 2025.
- National Student Support Accelerator — Tutoring Quality Standards, revised 12 November 2025.
