The Tutor Handbook · Volume 0124 · Series ID THB-0124
Return to The Tutor Handbook: https://edukatesengkang.com/the-tutor-handbook/
A learner is halfway through a Mathematics question when the tutor asks, “Do you want a hint?”
That sounds harmless until we ask a more important question: what is this question doing here?
If the question is part of teaching, a hint may be exactly right. If it is guided practice, a smaller cue may be useful. If it is diagnostic, the same hint can destroy the evidence the tutor needs. If it is a verification check, help may turn an independent test into another teaching episode. If it is performance rehearsal, an intervention that would never exist in the target condition may make the result impossible to interpret.
The visible task can be identical while the educational job is different.
This article owns one tutoring decision: how a tutor decides whether a task is being used for teaching, practice, diagnosis, progress monitoring or verification before deciding what help is allowed and what the result can mean.
The direct answer
Name the purpose before reading the performance.
A useful tutor can often place a task into one of five working purposes:
- Teaching: the tutor is trying to make a new idea, method or distinction available.
- Guided practice: the learner is trying the target while support is deliberately available.
- Diagnosis: the tutor is trying to distinguish competing explanations for a difficulty.
- Progress monitoring: the tutor is sampling whether a targeted capability is changing over time.
- Verification: the tutor is testing whether the capability survives with the level of independence, delay, variation or performance condition that matters.
The categories can overlap, and a lesson can move among them. The discipline is not to pretend that they are interchangeable.
The tutor should decide three things before the learner begins: what question the task is meant to answer, what support is permitted, and what conclusion the result could justify. If the purpose changes while the learner is working, say so and reset the interpretation.
The core rule is simple:
A task does not tell you what its result means. Its purpose, conditions and support history do.
Why this matters in tutoring
Tutoring compresses several educational jobs into a small amount of time. In ninety minutes a tutor may explain, model, question, practise, check homework, diagnose an error, revisit an earlier weakness and decide what comes next.
That density is useful. It also creates a measurement problem.
The tutor can accidentally teach during diagnosis, diagnose during teaching, verify with a heavily scaffolded task, or treat a supported practice success as evidence of independence. None of these actions is necessarily bad in itself. The problem appears when one activity is interpreted as though it had served another purpose.
Stanford’s National Student Support Accelerator places formative assessment and progress monitoring inside high-quality tutoring programmes, while its quality standards emphasise that data should be used to understand student needs and progress. AERO similarly frames checking for understanding as a way to determine what learners know and can do, identify gaps and adjust teaching. These sources support purposeful evidence use. They do not imply that every tutoring question should be treated as a test.
Teaching tasks: help is part of the job
When a task is being used to teach, tutor help is not contamination. It is the intervention.
The tutor may model the first step, explain a distinction, ask a leading question, compare an example and non-example, draw attention to a feature, supply a representation or work jointly with the learner.
The result of that task should therefore be interpreted modestly.
If the learner completes a question after the tutor models half of it, the completion can show that the learner followed the teaching, contributed useful thinking or can continue from the scaffold. It does not yet show that the learner could independently recognise and execute the whole route.
A good tutor does not withhold necessary teaching merely to protect a clean score.
They teach when teaching is the job, then collect different evidence later if independence matters.
Guided practice: support should be deliberate and visible
Guided practice sits between demonstration and independent performance.
The learner should be doing real work, but help remains available because the purpose is to build the capability rather than to measure it cleanly.
AERO’s scaffold-practice guidance describes scaffolds as temporary supports that should be responsive and removed when they are no longer needed. That is useful here because a scaffold has a legitimate job during practice.
Record the support level where it matters.
“Completed after one prompt to reread the condition” means something different from “completed independently”.
The support history is part of the evidence.
Diagnosis: the task is trying to separate explanations
Diagnosis has a different job.
The tutor is not primarily trying to produce a correct answer. They are trying to determine which explanation best fits the difficulty.
A learner may be failing because of missing knowledge, a misread condition, unstable notation, task language, method selection, execution, an access barrier or another cause.
A diagnostic question is valuable when different responses make those possibilities more or less plausible.
Helping too early can erase the distinction.
If the tutor says, “Remember to distribute the negative sign,” and the learner then succeeds, the tutor has taught or cued the route. They no longer know whether the learner would have noticed the sign issue independently.
The Diagnostic Probe owns the narrower question of designing one discriminating task. The Task-Purpose Gate sits one level above it: first decide that diagnosis is what the task is for.
Progress monitoring: comparability matters
A progress check asks whether a targeted capability is changing over time.
That means the tutor needs enough comparability across samples to make the change interpretable.
If the learner receives three prompts this week and none next week, the raw results are not equivalent.
If the first check uses familiar rehearsed items and the second uses fresh transfer items, the difference may reflect task conditions as well as learner change.
This does not mean every progress check must be identical.
Repeated exact items can create memory effects. Better progress monitoring often preserves the target, difficulty dimensions and support conditions while varying surface features enough to keep the evidence fresh.
AERO’s progress-monitoring guidance supports checking what learners understand and can apply, then responding with additional instruction or feedback where needed. The tutor’s contribution is to keep the measurement conditions legible.
Verification: the tutor temporarily stops helping
Verification asks a stronger question.
Can the learner perform the target under the independence, delay, variation or realistic pressure that matters?
Here, answer-giving assistance must usually be removed because assistance would perform part of what is being verified.
But legitimate access support should remain where it enables access without supplying the target capability.
Verification therefore requires a support boundary, not an indiscriminate “no help” rule.
The existing Access-Support Boundary protects that distinction.
The same question can change purpose during a lesson
A task does not have one eternal category.
A tutor may begin with diagnosis.
The learner’s first attempt reveals the weak link.
The tutor then teaches.
The learner retries with support.
Later, a fresh question becomes verification.
That sequence is educationally coherent.
The error would be to treat the final supported retry on the original item as though it were an independent verification simply because it ended correctly.
Purpose transitions should be visible:
“I have seen enough to know where the problem is. I am going to teach this now.”
Later:
“Now I want to see whether you can do a fresh one without that help.”
The learner benefits from understanding the transition too.
A fictional composite case: one question, three jobs
This is a fictional composite tutoring case. It does not describe a real learner or customer.
Alicia is working on an algebraic equation.
The tutor first gives a clean example to see whether Alicia notices that the variable appears on both sides.
That first item is diagnostic.
Alicia moves one term incorrectly.
The tutor now explains the equality relationship using a balance representation. The next item is teaching.
Alicia solves a similar equation with one prompt. That item is guided practice.
At the end of the session, the tutor gives a fresh equation without announcing the method. Alicia solves it independently.
That item provides a stronger verification signal.
Four correct answers could have appeared across this sequence.
They would not mean the same thing.
Help is not one thing
Tutors often talk about “giving help” as though support were a binary switch.
But support can enter at different points:
The tutor can restate the question.
They can define a word.
They can point to a line.
They can remind the learner of a strategy.
They can tell the first step.
They can correct an execution error.
They can model the whole route.
Each support changes the evidence differently.
What matters is whether the assistance supplies part of the target being inferred.
If the target is reading an unfamiliar word independently, defining the word performs the target.
If the target is evaluating an argument and the learner has a legitimate text-to-speech accommodation, access support may not perform the evaluation.
Purpose and target jointly determine what help is compatible with the evidence.
A tutor should not optimise every task for success
Good tutoring feels helpful.
That can create a subtle failure mode: the tutor intervenes whenever the learner hesitates.
The session looks smooth. Few errors survive. Work gets completed.
But the tutor loses the very evidence needed to determine whether the learner can select, monitor and recover independently.
Some purposeful struggle is evidence.
The tutor does not need to let the learner flounder indefinitely.
They need to know whether this moment is for building capability or observing it.
The tutor can declare the evidence question
One sentence improves decision quality:
“I am using this task to find out whether…”
Finish the sentence before the learner begins.
“…whether you can identify the relevant evidence without a prompt.”
“…whether the method is stable after a week.”
“…whether the error comes from reading the condition or from the algebra.”
“…whether the new planning routine works under timed conditions.”
The sentence makes it much easier to decide what support is allowed and what the result can mean.
Purpose changes what gets recorded
Teaching records should note what was taught and how the learner responded.
Practice records should capture support level or recurring errors where they matter.
Diagnostic records should preserve observations, competing explanations and the next discriminating check.
Progress records should preserve target, relevant conditions and comparable evidence.
Verification records should make the independence or performance condition explicit.
This does not require a five-column database for every question.
It requires enough information that a later tutor can tell why the evidence was collected.
Purpose and delayed checking
Immediate success after teaching is useful.
It tells the tutor whether the explanation and guided attempt were at least accessible in the moment.
It is not the same as delayed evidence.
The Evidence Freshness Window protects that distinction.
If the tutor wants to know whether learning persists, a later fresh task should be planned for that purpose.
The lesson can therefore end with two different truths:
“The learner can now perform the method with guided practice.”
“Independent retention has not yet been verified.”
That is not pessimism. It is accurate sequencing.
Changed-condition checking
A capability may work only under one surface form.
Verification can therefore change one condition while preserving the underlying target.
Change the numbers.
Change the context.
Remove the topic label.
Change the representation.
Introduce a competing method.
But do not change everything at once unless integrated performance is the actual target.
A changed-condition failure should still be interpretable.
Three learners make purpose discipline more important
In a three-student tutorial, one learner’s teaching can destroy another learner’s diagnosis.
Alicia answers aloud.
Beatrice hears the route.
Ciara has not yet attempted the problem.
If the task is practice, discussion may be useful.
If the tutor is trying to observe each learner’s independent method selection, the same discussion creates answer leakage.
The tutor can protect first attempts, then open the group discussion after the evidence is collected.
The task did not change.
The allowed interaction changed because the purpose changed.
Parent requests can accidentally change task purpose
A parent may reasonably ask, “Can you make sure the homework is correct before submission?”
If the tutor’s plan was to use the homework as diagnostic evidence, correcting every error changes its purpose.
The tutor can explain:
“I can help your child learn from this work, but I want to preserve the first attempt long enough to see what they can currently do. After that, we can correct it together.”
That is more precise than refusing help in the name of independence.
It tells the parent what the first attempt is for.
Accessibility does not disappear during verification
A common mistake is to interpret “independent” as “without any support of any kind”.
That can remove legitimate access conditions and make the task measure a barrier rather than the intended capability.
The tutor should distinguish:
access support that enables the learner to encounter the target;
instructional scaffolding that helps the learner perform the target;
answer-giving assistance that performs part of the target.
The correct boundary depends on what is being verified.
A task can be too clean to teach from and too contaminated to measure from
Tutors sometimes feel trapped between two extremes.
If they help, they contaminate evidence.
If they do not help, they waste learning time.
The solution is sequencing.
Collect the small amount of evidence needed for the decision.
Then teach.
Later, verify with a fresh sample.
The diagnostic moment does not need to consume the whole lesson.
When to abandon a task as evidence
Sometimes the tutor has already contaminated the evidence.
They gave a hint.
A peer revealed the strategy.
The learner saw the answer key.
The task wording itself gave away the method.
Do not pretend the item is still clean.
Use it for learning.
Then reserve a new item for the later evidence job.
The existing Fresh-Item Reserve exists for exactly this reason: some questions should remain untaught and unrehearsed until their evidence value is needed.
Purpose and the Tutor Classification functions
The Tutor Classification Model is useful here when it changes the decision.
A Class 1 Explainer function naturally works inside teaching.
A Class 3 Diagnostic Tutor function needs moments where evidence is protected before explanation begins.
A Class 5 Performance Coach function may need verification under realistic execution conditions.
The classes are functions, not permanent human rankings.
The same tutor may move among them in one lesson.
Task purpose helps explain why that movement requires different support rules.
Repair, Alignment and Frontier
The three tuition modes also change what task purpose dominates.
In Repair, the tutor may use diagnosis to locate the first weak link, teaching to rebuild it and verification to see whether the repair holds.
In Alignment, monitoring may ask whether the learner can meet current curriculum demand under the right conditions.
In Frontier, teaching and verification may deliberately explore whether a learner can transfer established capability into more advanced work.
There is no fourth “Performance” mode. Performance is a tutoring function and condition that can appear within the three established modes.
AI-generated questions still need a declared purpose
Generative AI can produce ten questions in seconds.
That does not make them a coherent evidence set.
A tutor must still decide:
Is this item teaching, practice, diagnosis, monitoring or verification?
Does the wording accidentally cue the method?
Is the answer valid?
Is the difficulty appropriate?
Does the question test the intended capability or a hidden demand?
AI can increase item supply.
It cannot outsource the educational purpose.
A practical purpose strip
Before a consequential task, the tutor can write four short fields:
- Purpose: teaching / practice / diagnosis / monitoring / verification.
- Target: what capability is being built or observed?
- Allowed support: what help can enter without defeating the job?
- Safe claim: what could this performance legitimately show?
That is enough.
There is no need to formalise every classroom interaction.
Use the strip when the distinction changes a route, progress claim, handoff or parent report.
Purpose comes before difficulty
Tutors often begin evidence design by asking, “How hard should the question be?”
That is too early.
A very easy item can be a strong diagnostic if it cleanly tests a prerequisite. The same item can be useless as final verification of an advanced capability. A demanding transfer problem can be excellent evidence of generalisation and poor evidence for locating the first step of a basic misconception.
Difficulty only becomes meaningful relative to the task’s job.
Likewise, a short question is not automatically shallow and a long question is not automatically rigorous. What matters is whether the item creates an opportunity to observe the target distinction without irrelevant demands dominating it.
This is especially important when a learner is changing quickly. The tutor may keep raising difficulty because the learner succeeds, when what is actually needed is a delayed check. Or the tutor may keep repeating a familiar hard format when a changed surface would reveal whether the method transfers.
Start with the evidence question.
Then choose the difficulty, support, delay, variation and scoring rule that make that question answerable.
Failure modes
The tutoring purpose gate prevents several recurring errors.
The helpfulness reflex. The tutor intervenes whenever the learner hesitates, even when hesitation is the evidence.
The score reflex. Every task receives a mark even when the task was primarily for teaching.
The supported-mastery error. Guided practice success is reported as independent mastery.
The contaminated diagnosis. A hint removes the very distinction the tutor was trying to observe.
The access purge. Legitimate access support is removed in the name of “true independence”.
The endless test. The tutor keeps withholding teaching to protect evidence long after enough evidence has been collected.
The practice-as-proof error. Repeated success on rehearsed items is treated as strong evidence of transfer.
The peer-leakage error. Group discussion destroys independent evidence before it is collected.
The AI item pile. Many generated questions are mistaken for a purposeful assessment design.
Research limits
The distinctions in this article are compatible with formative assessment, scaffolding and progress-monitoring guidance, but they should not be presented as a validated five-category tutoring taxonomy.
Stanford’s National Student Support Accelerator provides research-based, research-informed and emergent quality standards for tutoring programmes, including formative assessment and data use. Its standards are programme guidance, not proof that this exact Task-Purpose Gate produces a particular student outcome.
AERO’s practice guides support checking understanding, responding to learner needs and using scaffolds deliberately. These are school-facing resources. They do not automatically validate a three-student private-tuition protocol in Singapore.
EEF’s tutoring guidance is also appropriately cautious: tutoring is positive on average but not every study shows positive impact, so implementation should be monitored and evaluated.
The framework here is therefore a practical evidence-governance model for tutoring. It separates actions that are educationally useful from conclusions those actions can justify.
Sources and further reading
National Student Support Accelerator, Tutoring Quality Standards: https://nssa.stanford.edu/tqis/quality-standards
National Student Support Accelerator, Tutor Training Toolkit Rationale and Usage Guide: https://nssa.stanford.edu/tutor-training-toolkit/rationale-usage-guide
Australian Education Research Organisation, Monitor Progress: https://www.edresearch.edu.au/guides-resources/practice-guides/monitor-progress
Australian Education Research Organisation, Scaffold Practice: https://www.edresearch.edu.au/guides-resources/practice-guides/scaffold-practice
Education Endowment Foundation, Making a Difference with Effective Tutoring: https://educationendowmentfoundation.org.uk/education-evidence/effective-tutoring
Education Endowment Foundation, A School’s Guide to Implementation: https://educationendowmentfoundation.org.uk/education-evidence/guidance-reports/implementation
The final return
Imagine the tutor asking the original question again.
“Do you want a hint?”
There is nothing wrong with the question.
What matters is whether the tutor knows what happens to the evidence if the learner says yes.
During teaching, help can be the whole point.
During practice, help can be deliberately faded.
During diagnosis, help can destroy a distinction.
During monitoring, changing help can break comparability.
During verification, answer-giving help can turn the test back into teaching.
A tutor does not become less helpful by protecting these boundaries.
They become more precise about when to help, why to help and what a successful response means after the help arrives.
That precision makes tutoring kinder to the learner and harder on unsupported claims.
Before deciding whether to hint, teach, wait, score or move on, decide what job the task is doing.
