How Assessment for Teaching Works | Turning Evidence Into Better Instruction Before, During and After Learning

Assessment design protocols by evidence question

Assessment quality improves when the tutor starts from the evidence question rather than from the resource. “I have a worksheet” is not an assessment design. “I need to know whether the learner can retrieve this after delay without a model” is. The evidence question determines task form, support conditions, timing and interpretation.

Baseline design

A baseline exists to make later evidence interpretable. It need not be a long pre-test. A short well-chosen sample can establish what the learner could do before an intervention. Baseline quality matters more than baseline size.

A useful baseline states: target capability, task type, support conditions, time condition, current accuracy or quality, first weak link, and any uncertainty that limits comparison. If later evidence uses a much easier task, more help or different timing, the comparison should say so rather than pretending it is the same measure.

When no clean baseline exists

Sometimes teaching has already started before the tutor realises a baseline would have been useful. Do not manufacture a pre-intervention claim. Record the current state honestly: “after two guided lessons, learner can…” and use future change from that point.

Comparability across assessments

Two scores are comparable only to the extent that they sample sufficiently similar constructs under sufficiently interpretable conditions. Identical questions are not necessary; identical questions can overstate progress through familiarity. Comparable fresh questions are usually better.

• Check whether difficulty changed.
• Check whether support changed.
• Check whether time changed.
• Check whether topic cues disappeared or appeared.
• Check whether representation changed.
• Check whether the learner had access to notes, tools or AI.
• Check whether the mark scheme rewarded the same quality.
• Check whether the number of items is large enough to support the conclusion.

Fresh-item reserve

A tutor should preserve some unseen items for later evidence. If every task is used during teaching, there is no clean way to distinguish memory of practice from learning. Fresh items can be modest in number but should target the same underlying capability.

The reserve is especially important for misconceptions, transfer, method selection and checking. Once the learner has seen the correction path, the original item becomes weak evidence of independent capability.

Assessment range: too easy and too hard are both uninformative

If every item is easy, the tutor cannot see the boundary of capability. If every item is far beyond readiness, the tutor sees only failure. Good assessment samples around the current frontier: some tasks the learner can do, some that require effort, and a small number that expose the next likely boundary.

This does not mean turning every lesson into adaptive testing. It means choosing enough range that the evidence can change the next teaching move.

Error-family assessment

Repeated errors become more useful when classified by mechanism rather than topic. In Mathematics, sign error, copying error, method selection, representation and checking are different families. In English, answer scope, evidence selection, organisation, grammar and editing differ. In Science, concept recall, mechanism, evidence use and data interpretation differ.

A tutor can sample one or two fresh items from the dominant error family to see whether the same mechanism recurs. This is more efficient than assigning another full worksheet blindly.

Assessment of partial knowledge

Wrong answers can contain strong partial knowledge. A learner may choose the correct representation but calculate incorrectly, identify the correct Science process but fail to connect cause and effect, or select relevant English evidence but answer the wrong scope.

Partial evidence changes teaching. Preserve what works and repair the earliest failing step. Re-teaching the entire topic can waste time and blur progress.

Assessment of misconceptions

A misconception is stronger than an isolated mistake. It is a stable wrong model that predicts errors consistently. To assess one, choose contrast cases where the wrong and correct models predict different answers. Ask the learner to commit before feedback.

If the learner changes answer after a cue but cannot explain the underlying relationship, the misconception may still be active. Fresh contrast cases are the acceptance evidence.

Assessment of strategy flexibility

A learner may know one valid method but fail when the method is inefficient or unavailable. Strategy assessment asks whether the learner can recognise when a different representation or method is more useful.

Do not reward method switching for its own sake. The goal is flexible control, not collecting tricks. A learner who can explain why one route is better for this task shows stronger strategic evidence.

Assessment of explanation

Explanation quality depends on the target. In Mathematics, the learner may need to justify a method or relation. In Science, a causal mechanism may be required. In English, evidence must support a claim. Assessment should define what a complete explanation contains before judging length or fluency.

Assessment of independence, help-seeking and checking

Independence is not absence of all tools or help. It means the learner owns the target under the intended support conditions. Help-seeking can also be assessed as a capability: attempt first, locate the stuck point, formulate a precise question, use the answer, and continue.

Checking should be sampled before tutor feedback. Ask the learner to identify what deserves checking, choose a verification method and act on discrepancies. A correct answer that was never checked does not reveal checking capability; an incorrect answer that the learner catches and repairs may show strong checking.

Assessment of planning and transfer

Planning is visible when the learner can turn a broad task into ordered next actions. For writing, this may be claim and structure; for a project, question, method and data; for revision, priority and order; for Mathematics, representation and subgoal selection. If the adult supplies the sequence, planning remains untested.

Transfer assessment should change one meaningful feature at a time where possible. If everything changes at once, failure is hard to interpret. Begin with moderate variation, then increase distance as capability grows.

Subject assessment protocols and worked cases

A cross-subject assessment system still needs subject-specific evidence rules. English, Mathematics, Science and Additional Mathematics do not fail in the same way, and the same surface score can imply different next teaching moves. The protocol below keeps the evidence job consistent while respecting subject structure.

English assessment protocol: reading

• Sample literal retrieval, inference, vocabulary-in-context and answer scope separately enough to locate the bottleneck.
• Compare oral understanding with written response when expression may be contaminating comprehension evidence.
• Use fresh passages after teaching; do not rely on corrected original texts.
• Record whether questions were discussed before the independent attempt.
• Retest under time only after core comprehension is stable.

A learner who reads a passage accurately but answers beyond the question scope needs a different route from a learner who cannot construct the inference. The total comprehension score may be identical. Assessment for teaching keeps those mechanisms separate.

English assessment protocol: writing

• Separate idea generation, selection, organisation, development, sentence control and editing.
• Use planning evidence to see whether collapse begins before drafting.
• Compare first draft with revision to assess feedback use.
• Use a fresh prompt to check whether the improved structure travels.
• Do not infer whole-writing mastery from one polished model-assisted piece.

If the learner generates strong ideas but cannot control paragraph purpose, the next move is not another vocabulary list. If paragraph logic is strong but sentence control collapses under time, the evidence points elsewhere. Assessment should protect strengths while locating the first weak link.

English case: inference versus answer scope

A Secondary 1 learner earns 4/8 on a comprehension set. Oral discussion shows accurate understanding of the passage. On three written questions, the learner identifies relevant evidence but adds unsupported material beyond the scope. The formative interpretation is not “weak comprehension” but Stable passage meaning + Fragile answer-scope control. The next lesson uses demand classification and one fresh passage rather than broad reteaching.

English case: composition planning

A Primary 6 learner writes vivid openings but produces rushed endings. Before assigning another full composition, the tutor assesses the planning stage: spoken story spine, written event list and five-minute plan. The learner produces too many events and no central consequence. The assessment locates narrative selection before drafting. The next move is a constrained story spine, then a fresh prompt.

Mathematics assessment protocol: arithmetic and representation

• Separate number-fact retrieval from conceptual number relationships.
• Inspect representation before calculation on contextual tasks.
• Compare numerical and word-problem forms of the same structure.
• Record whether the learner selected the operation or was cued by worksheet title.
• Use estimation or inverse checking where reasonableness matters.

Mathematics assessment protocol: problem solving

• Check representation before arithmetic where the task is contextual.
• Use mixed questions to test method selection.
• Inspect the first wrong line, not only the final answer.
• Include one changed-context problem after repair.
• Use checking evidence before adult correction.

A learner can be Stable in arithmetic and Blocked in translating relationships. If the tutor responds to every word-problem error with more arithmetic practice, the assessment has failed to separate capabilities.

Mathematics assessment protocol: algebra

• Probe prerequisites in dependency order: signed operations, equality, expansion or factorisation, equation control.
• Distinguish execution with topic cues from selection in mixed work.
• Check whether shorthand hides invalid transformations.
• Use fresh equations after feedback.
• Add timed and mixed conditions only after structure is stable.

Mathematics case: Primary 5 fraction representation

A learner solves unlike-denominator calculations accurately but fails two context questions. A short assessment shows the learner can compare quantities once the whole is identified but chooses the wrong reference whole when the story changes. The next teaching move is a contrast set of changing-whole problems, not more common-denominator practice. Fresh context after delay becomes the acceptance check.

Mathematics case: Secondary 2 algebra

A learner loses marks across factorisation and algebraic fractions. The tutor orders a short dependency probe: signed operations, expansion, common-factor extraction, factorisation, cancellation of factors. The first repeated failure is common-factor extraction. The assessment has done its job when the next lesson begins there rather than reteaching all algebraic fractions.

Science assessment protocol: concepts and mechanisms

• Separate term recall from relationship understanding.
• Ask for cause → mechanism → observation where appropriate.
• Use changed apparatus or context to test transfer.
• Distinguish observation, inference and conclusion.
• Check whether answer language is the bottleneck rather than concept knowledge.

Science assessment protocol: investigations

• Ask the learner to define question, changed and measured variables, and relevant controls.
• Check the measurement rule before data collection.
• Require a conclusion proportional to the evidence.
• Ask for one limitation tied to mechanism, not vague “human error”.
• Retest investigation reasoning in a different context.

Science case: keywords without mechanisms

A Primary 6 learner writes the expected terms in open-ended responses but does not earn the mark. A three-item probe shows accurate vocabulary and process identification but incomplete causal chains. The next move is not keyword drilling. It is explanation architecture: what changed, what mechanism connects the change, and what observation follows. Fresh apparatus tests whether the relationship transfers.

Science case: graph overclaim

A learner reads every plotted value correctly but writes a conclusion broader than the measured range. The assessment separates data reading from claim calibration. The tutor uses a second graph with a different pattern and asks what can and cannot be concluded. The learner’s Science vocabulary was never the primary weak link.

Additional Mathematics assessment protocol

• Probe algebraic dependencies before attributing failure to the current A-Math chapter.
• Separate identity or formula recall from selection.
• Check interpretation alongside technique in functions, graphs and calculus.
• Use mixed multi-stage questions to test integration.
• Use full-paper simulation only after dependencies are sufficiently stable.

A-Math case: differentiation that is really algebra

A Secondary 4 learner differentiates cleanly but loses marks simplifying the derivative. Chapter scores make calculus look weak. Step-level evidence shows the calculus is Stable while algebraic simplification is Fragile. Assessment for teaching routes the repair to algebra and protects calculus practice from unnecessary repetition.

A-Math case: identity selection

A learner can reproduce several trigonometric identities from memory but cannot decide which one reduces an unfamiliar equation. Retrieval is Stable; method selection is Fragile. The next assessment mixes identity candidates and asks the learner to state what form would make the equation solvable before manipulating symbols.

Assessment in revision cycles

Revision should use assessment to allocate time, not merely to produce scores. Blocked prerequisites get repair. Fragile skills get retrieval and spacing. Stable capabilities get variation. Transfer-Ready capabilities get mixed integration. Examination-Ready capabilities receive selective maintenance.

A revision timetable that ignores learner state can overinvest in favourite topics, repeatedly attack weak topics without diagnosing them, or allow strong topics to decay because they receive no return.

Assessment after a revision block

At the end of a revision block, do not ask only “how much did we cover?” Ask what changed in fresh performance. If the learner completed forty questions but the same weak link remains, volume did not solve the problem.

Assessment and workload capacity

A learning plan can fail because the timetable cannot sustain it. Assessment should sometimes include the weekly system: missed work, start times, fatigue, recovery and ability to complete independent practice. A correct academic intervention still needs enough capacity to run.

Assessment and attendance

Attendance is exposure, not evidence of learning. A learner can attend every lesson and remain prompt-dependent. Another can miss a session yet maintain capability independently. Use attendance as context, not outcome evidence.

Assessment and lesson satisfaction

A learner or parent may find a lesson helpful, clear or enjoyable. Those reactions matter for fit and engagement but do not prove learning. Pair satisfaction with fresh performance before making instructional conclusions.

Assessment and tutor judgement

Professional judgement is necessary because evidence is incomplete. Good judgement is not guesswork; it makes assumptions explicit, seeks disconfirming evidence and updates when fresh work contradicts the model.

The Capability Profile is useful precisely because it keeps judgement provisional and bounded.

The evidence review meeting

A review meeting should not become a recital of marks. Its job is to decide what the evidence now justifies. A useful sequence is: state the target, show the evidence, name the support conditions, identify the first weak link, state what remains uncertain, choose the next move, and define the fresh retest.

In a parent–tutor setting, this keeps the conversation on learning rather than personality. In a tutor-team setting, it keeps professional judgement inspectable. In a student review, it gives the learner a route they can understand and increasingly carry.

Assessment should create a decision point in the lesson plan

A lesson plan is more adaptive when it contains at least one place where evidence can change what happens next. For example: if all three learners retrieve the prerequisite independently, move to transfer; if one learner remains Fragile, branch that learner into supported practice; if all three are Blocked, pause the planned extension and repair the prerequisite.

This does not require scripting every possible response. It requires planning enough that assessment can matter.

The next-move decision tree

Continue, change, reduce or stop: assessment as the release system

Assessment should not only identify when learners need more help. It should identify when help can change, reduce or stop. A mature teaching system contains its own release logic.

Continue

Continue when the intervention is changing fresh evidence but the learner still needs the current support. The important sign is not attendance or satisfaction; it is improving learner-owned performance.

Change

Change when a fair trial produces little independent improvement, the support never fades, or new evidence shows the wrong weak link was targeted. Check implementation fidelity before abandoning a sound route.

Reduce

Reduce when the learner succeeds with less prompting, after delay and under relevant variation. Reduction can mean fewer sessions, lighter homework, less feedback, narrower scope or more independent time.

Stop

Stop special intervention when ordinary school and home learning can maintain the capability and the profile can be retired. Define a return condition rather than keeping support forever “just in case”.

Implementation fidelity: did we run the intervention we think we ran?

Before declaring that a teaching move failed, check whether it was actually implemented. Did the planned retrieval happen? Were prompts faded? Was the fresh item truly fresh? Did homework occur under the recorded conditions? Did the tutor change multiple variables at once?

If the route was not implemented consistently, the assessment cannot fairly judge the intervention. At the same time, an intervention too complicated to run consistently may be a poor fit for the learner’s real week.

Intervention cost belongs in the decision

An intervention can improve one score and still cost too much time, travel, sleep, attention or family bandwidth. Assessment for teaching should consider whether the gain is large enough and durable enough to justify the cost.

A ten-minute retrieval routine that stabilises a high-leverage prerequisite may be more valuable than a two-hour weekly programme that duplicates school work without increasing independence.

Assessment load: do not test the learner out of learning

Constant checking can fragment practice, create performance pressure and reduce time available for teaching. Assess where the result can change a decision. Once the route is clear, let the learner work long enough for the intervention to take effect before measuring again.

Common assessment-for-teaching failure mode 1: score first, mechanism never

The tutor records 63% and immediately assigns more of the same topic. No one checks which errors produced the score. The fix is item-level mechanism analysis before prescribing practice.

Failure mode 2: every error becomes a diagnosis

Normal noise becomes a growing deficit list. Use repeated patterns and decision relevance. One unusual mistake can be monitored without creating a profile.

Failure mode 3: the corrected task becomes the evidence

The learner produces a perfect correction after feedback and adults declare mastery. Use a fresh task later. Corrections are intervention evidence, not acceptance evidence.

Failure mode 4: hidden support

Parent prompts, peer answers, AI hints or tutor cues are invisible in the record. Later comparisons are misleading. Make support conditions visible without shaming the learner.

Failure mode 5: broad labels from narrow samples

Two algebra questions become “weak algebra”; one awkward oral response becomes “poor communication”. Match the breadth of the claim to the breadth of the evidence.

Failure mode 6: easy tasks masquerade as progress

Scores rise because task difficulty falls. Record difficulty changes and use comparable fresh tasks before claiming improvement.

Failure mode 7: hard tasks masquerade as regression

The learner moves from familiar to transfer tasks, scores lower and is told they got worse. Interpret the change in construct demand before reading the percentage.

Failure mode 8: the fastest learner becomes the standard

Small-group comparison replaces criterion-based interpretation. Track each learner against the target and support conditions, not the quickest peer.

Failure mode 9: assessment becomes surveillance

Every hesitation, emotion and error is recorded. Keep only evidence that changes educational decisions. Learners need space to practise without being continuously profiled.

Failure mode 10: no exit condition

The intervention continues because it is working, but nobody defines what success would look like without it. Add a reduction or retirement condition from the start.

A one-page formative assessment record

Capability: ______
Decision this evidence should improve: ______
Task: ______
Conditions: ______
Support: ______
Observation: ______
First weak link / strong stable component: ______
Current learner state: ______
Uncertainty: ______
Next teaching move: ______
Fresh retest: ______
Reduce/return condition: ______

Parent version of the assessment record

Parents usually need five lines: what we checked, what the learner could do, what help was needed, what we are changing next, and when we will review. That is enough to coordinate without turning the child into a case file.

Student version of the assessment record

The learner should hear: what already works, where the route breaks, what they will try next, what support will be available, and what fresh task will show progress. Assessment should increase agency rather than merely deliver judgement.

Tutor-team version of the assessment record

Professional continuity needs slightly more detail: task sample, support level, error pattern, interpretation, uncertainty and planned retest. Another tutor should be able to understand the route without inheriting a permanent label.

The assessment handover to the Capability Profile

When several observations must be assembled across tasks, use the Capability Profile. Assessment for Teaching asks what the current evidence supports; the Capability Profile holds the bounded learner model that guides multiple later decisions.

Advanced evidence boundaries: what ordinary tutoring data can and cannot tell us

Assessment for teaching happens in a noisy environment. Tasks differ, learners have good and bad days, school demands change, and support is rarely perfectly standardised. The goal is not to pretend ordinary tutoring data has laboratory precision. The goal is to make claims proportionate to the evidence and choose the next educational move responsibly.

Reliability: would the result roughly repeat?

One response can be informative but unstable. Reliability improves when a pattern recurs across fresh tasks, time points or representations. The tutor does not need formal reliability coefficients to use the idea. Ask whether the conclusion depends on one item or persists under reasonable resampling.

If one fraction problem is wrong and the next four are independently correct, “weak fractions” is a poor conclusion. If the same representation failure appears across varied tasks, the evidence is stronger.

Validity: does the evidence support the claim we want to make?

A score can be reliable and still measure the wrong thing. A learner may consistently fail a dense Mathematics worksheet because reading load dominates. A repeated result does not prove the mathematical capability is weak. Validity asks whether the task actually supports the intended interpretation.

Sampling error in ordinary classroom language

A few questions sample only a small part of a subject. The smaller the sample, the more cautious the claim should be. “Two of three inference items were weak” is more honest than “inference is poor” when only three items exist.

Large decisions—changing pathways, adding intensive support, stopping a subject—deserve broader and more varied evidence than small decisions such as choosing tomorrow’s practice task.

Subscore precision can be an illusion

Digital platforms and test reports often display fine-grained categories. A subscore based on two items may look precise because it has a percentage and label, but the underlying evidence may be thin. Read item counts and task demand before building a capability profile from the subscore.

Triangulation: use different evidence sources for the same decision

Confidence grows when independent evidence sources point in the same direction. A school script, a fresh tutor task and delayed homework retrieval can together support a stronger interpretation than three near-identical tuition worksheets.

Triangulation does not mean collecting everything. Choose evidence sources that add genuinely different information: school performance, fresh independent work, delayed return, transfer task, or paper conditions.

School evidence and tuition evidence can legitimately disagree

Tuition tasks may be more scaffolded, more diagnostic, more focused or more difficult than school work. School tests may be timed and mixed while tuition practice is untimed and topical. Different results may reflect different conditions rather than one setting being “wrong”.

Compare the task conditions before comparing the marks. Then design one bridge task that isolates the suspected condition: mixed versus topical, timed versus untimed, supported versus independent, familiar versus changed context.

Trend data needs stable enough measurement

A line of scores looks persuasive, but trends can be distorted when task difficulty, marking, subject mix or support changes. Record major changes in the assessment conditions so the trend is not interpreted as pure learner growth or decline.

Sometimes a lower score on a harder transfer task represents progress in the assessment system because the learner has moved from familiar execution into a more demanding capability. Interpretation should follow construct demand, not graph direction alone.

Regression to the mean and unusually good or bad results

Extreme performances are often followed by more ordinary ones even without intervention. After one exceptionally low or high score, avoid designing a major learning route until other evidence confirms the pattern. This is especially important when the paper was unusual, the learner was unwell, or the sample was small.

Assessment after an intervention does not prove causation

If performance improves after tuition, the improvement is real evidence about current performance. It does not automatically prove that tuition caused the whole change. School teaching, independent practice, maturation, easier tasks and other influences may contribute.

For ordinary educational decisions, strong causal proof is rarely necessary. What matters is whether the current route is producing useful fresh evidence and whether the learner is becoming more independent. Keep causal language modest.

Response to intervention is evidence, but only under the intervention conditions

If a learner improves immediately after a scaffold, that tells us the scaffold unlocks performance. It does not yet show the learner can perform without it. If performance stays strong as the scaffold fades and after delay, the interpretation changes.

No response to intervention also needs interpretation

When a learner does not improve, ask whether the weak link was identified correctly, the intervention was implemented, the learner had enough opportunity, the task was comparable, and the duration was sufficient. “Did not work” is a starting question, not the final diagnosis.

Floor and ceiling effects

An assessment that is too difficult pushes many results toward zero and hides differences in partial knowledge. An assessment that is too easy pushes results toward full marks and hides growth. Choose tasks near enough to the learner’s range that change can be observed.

Missing evidence should sometimes stay missing

Not every capability needs immediate measurement. If the decision does not depend on it, avoid unnecessary testing. “Not enough evidence” is a valid state. Create a targeted opportunity later only when the information would change the route.

Assessment fairness does not mean identical conditions

Fair evidence aims to preserve the target while giving the learner a valid opportunity to demonstrate it. Legitimate access conditions may differ. What matters is whether the interpretation matches the actual conditions and whether instructional scaffolds are distinguished from access.

Assessment and response modality

Oral, typed, handwritten and visual responses can reveal overlapping but not identical capabilities. If the target is conceptual understanding, a modality comparison can isolate expression barriers. If the target includes formal written performance, oral evidence cannot simply replace it.

Assessment and language load

Subject tasks often carry language demand. To separate language from subject reasoning, compare equivalent structures with adjusted wording or ask the learner to represent the task nonverbally. Then restore the integrated school demand later. The purpose is diagnosis, not permanent removal of language.

Assessment and working memory load

A learner may know every component but fail when too many elements must be held and coordinated at once. Compare isolated component evidence with integrated performance. The next move may be chunking, external representation or integration practice rather than reteaching each component.

Assessment and fluency

Speed needs interpretation. Slow correct work may reflect deliberate strategy, weak retrieval, inefficient method or overchecking. Fast work may reflect fluency or impulsivity. Observe process and accuracy together before setting a speed target.

Assessment and test anxiety claims

A learner may report feeling anxious and perform worse under test conditions. The tutor can record the difference without making a clinical diagnosis. Compare untimed subject capability with timed paper behaviour and route the educational problem appropriately. Health or psychological concerns exceed the tutor’s role and should not be inferred from marks alone.

Assessment and motivation claims

Low effort on one task does not establish a stable motivation problem. Examine difficulty, clarity, previous failure, workload and opportunity. If task behaviour changes when the weak link is repaired, the original “motivation” interpretation was incomplete.

Assessment and AI assistance

AI can change the construct by supplying explanation, planning, wording or verification. Record the AI role. If independent performance matters, use a later fresh task with reduced or no AI contribution. Assisted output remains valuable learning evidence; it is simply a different condition.

Assessment and privacy

Collect only the learner information needed for the educational job. A capability record usually needs task evidence and support conditions, not speculative personal explanations. Share the minimum necessary for continuity and retire stale records.

Assessment language that preserves uncertainty

• Prefer “current evidence suggests…” to “the learner is…”.
• Prefer “on these tasks…” to “always…”.
• Prefer “needed a strategic prompt…” to “couldn’t do it”.
• Prefer “not yet tested after delay…” to “mastered”.
• Prefer “the first weak link appears to be…” when evidence is provisional.
• Prefer “under the current access conditions…” when supports matter.
• Prefer “fresh retest planned…” to treating the corrected original as proof.

The strength of the claim should match the strength of the evidence

A small piece of evidence can justify a small action. One clear error may justify tomorrow’s contrast example. A broad claim about readiness, long-term progress or support reduction requires wider and more varied evidence. This proportionality rule is the centre of responsible assessment for teaching.

Practical workflows for assessment-led teaching

Workflow 1: the learner gets the question wrong during teaching

Do not correct immediately. First decide whether the wrong answer is informative. Ask what the learner thought the task required or inspect the working. If the first weak link is visible, teach that link. If the cause is uncertain, use one discriminating question or contrast item. Then reserve a fresh attempt.

Workflow 2: the learner gets the question right after a hint

Record success-with-hint. Do not upgrade the learner state yet. Fade the hint on the next fresh item. If the learner succeeds independently after delay, the interpretation strengthens. If the learner waits for the same hint, the support dependency is part of the evidence.

Workflow 3: the learner is correct but very slow

Inspect the method before prescribing speed work. Slow performance may reflect careful checking, weak retrieval, inefficient strategy, unfamiliar notation or overthinking. Identify the time-consuming step, then decide whether fluency, strategy or examination pacing is the actual target.

Workflow 4: the learner is fast but inaccurate

Compare normal pace with a deliberately controlled attempt. If accuracy improves sharply, pacing or monitoring may be the lever. If the same misconception persists, slowing down alone will not repair it. Use error-family evidence before telling the learner to “be careful”.

Workflow 5: the learner does well in tuition but poorly in school

Compare conditions: topic-labelled versus mixed, untimed versus timed, tutor prompts versus independent work, familiar materials versus school formats. Build one bridge task that changes the suspected condition. Do not assume school performance is unfair or tuition performance is artificial until the condition is tested.

Workflow 6: the learner does poorly in tuition but well in school

The tutor may be using transfer tasks, fresh questions or diagnostic probes that are intentionally harder than school homework. If school evidence is strong and stable, the tutoring difficulty may represent extension rather than failure. State that explicitly so the learner does not interpret lower tuition scores as regression.

Workflow 7: marks drop suddenly

Before intensifying support, check task demand, paper format, syllabus transition, health/attendance context, subject dependencies and whether the lower mark is one unusual result or a repeated pattern. A sudden drop is a signal for diagnosis, not an automatic instruction to add hours.

Workflow 8: marks improve quickly

Celebrate the evidence but check what changed. Was the task easier, support heavier, topic familiar, or capability genuinely stronger? Use fresh delayed and changed-condition work before declaring the profile complete.

Frequently asked questions

Is formative assessment just frequent testing?

No. Formative assessment is evidence used to improve learning or teaching. A test that produces a mark and no instructional response is not formative merely because it happens often. A short conversation, exit item or draft can be formative if the evidence changes the next move.

Should every lesson include an assessment?

Every lesson contains informal evidence, but not every moment needs a formal check. Place deliberate checks where the result can change the route: prerequisites, first application, support fading, transfer, end-of-lesson planning or delayed return.

How many questions are enough?

Enough to support the decision. One item can raise a question. Several varied fresh items may support a provisional pattern. Broad readiness or support decisions need stronger evidence. There is no universal item count independent of task complexity and consequence.

Should I assess before teaching a new topic?

Use a short prerequisite check when prior knowledge will change the teaching plan. Do not pre-test exhaustively merely to produce a baseline score. The assessment should save teaching time by locating what is already secure and what needs repair.

What if the learner guesses correctly?

If reasoning matters, ask for the basis or use a fresh parallel item. Do not interrogate every correct answer. The need for further evidence depends on the decision being made.

What if the learner knows the answer but cannot explain it?

Separate execution from explanation. Some capabilities require explanation; others do not. If explanation is part of the target, assess it directly. If the target is fluent retrieval, forcing long explanations can distort practice.

Should marks be hidden during formative assessment?

Not necessarily. The key is what the learner does with the evidence. Marks can help track performance but should not crowd out mechanism, feedback and fresh action. For some tasks, qualitative evidence is more useful than a percentage.

Can homework be trusted as evidence?

Yes, if support conditions are understood. Homework completed with notes, parent help or AI can still show learning, but it does not automatically show independent performance. Use a fresh in-lesson item when independence matters.

Can self-assessment replace teacher assessment?

No. Self-assessment is a capability that improves through calibration with external evidence. Use it to increase learner judgement, not to remove professional assessment before the learner is ready.

Can peer assessment be accurate?

It can be useful for narrow teachable jobs. Ask peers to identify one criterion, evidence or discrepancy rather than assign broad grades. The learner should remain responsible for accepting, rejecting or using the feedback with reasons.

Should I use full papers early?

Only when the evidence question needs integrated paper performance. If a prerequisite is Blocked, full papers may repeatedly expose the same issue without repairing it. Use topic or dependency assessment first, then restore mixed and timed conditions later.

How do I assess transfer?

Change wording, representation or context while preserving the underlying capability. Start with moderate variation. Use more distant transfer only after the learner succeeds on nearer cases. Record which dimensions changed so failure remains interpretable.

How do I assess examination readiness?

Use mixed, timed and sufficiently authentic paper conditions after subject capability is stable enough. Observe pacing, selection, checking and recovery. One full paper is evidence, not a permanent verdict.

What if assessment makes the learner anxious?

Reduce unnecessary stakes and make the purpose clear where possible. Teaching assessments should help choose the next move, not constantly rank the learner. If anxiety or health concerns exceed ordinary educational support, tutors should not diagnose them from performance.

Should strong learners still be assessed?

Yes, when the result will affect extension, transfer or support reduction. Do not create endless testing simply because the learner is strong. Use assessment to identify the next meaningful boundary, then let the learner work.

What if the evidence is contradictory?

Compare conditions rather than averaging blindly. One setting may be timed, another scaffolded; one task may be familiar, another transfer. Design a bridge task that isolates the difference. Keep the profile provisional until the contradiction is better understood.

When should assessment records be deleted?

When they no longer serve continuity or current decisions. Keep the minimum educationally necessary record. Retire old weakness profiles so they do not become permanent identity.

A seven-question review before changing instruction

• What capability did the task actually sample?
• What support and access conditions were present?
• Is the pattern repeated or could it be noise?
• Where is the earliest weak link?
• What evidence contradicts my first interpretation?
• What is the smallest teaching change justified now?
• What fresh evidence would make me change my mind?

The WEP120 operating standard

Formative assessment and the next teaching move follow one loop: define the decision → define the target → gather bounded evidence under known conditions → interpret cautiously → choose the smallest warranted instructional response → give the learner a chance to act → retest fresh → update, reduce or change support.

The public owner is this page. Adaptive Teaching owns the changed instructional response; Assessment Evidence owns the broader limits of what tests can tell us; the Capability Profile holds learner-specific synthesis; Practice & Review and Learning Has Held provide the acceptance cycle. No new generic formative-assessment URL is needed.

Assessment casebook across the school journey

The same assessment logic should survive changing school stages. What changes is the capability being sampled, the task demand, and how much independence the learner is expected to carry. These cases show how a total result becomes a teaching decision without turning the result into identity.

Primary 1: decoding versus familiar-text memory

A learner reads a class reader fluently but hesitates on a fresh decodable sentence. Assessment separates repeated-text fluency from fresh decoding. The tutor samples several unfamiliar regular words under ordinary reading conditions. If the learner succeeds after a first-sound prompt but not independently, the current evidence is Fragile decoding rather than a global reading weakness. The next move is short fresh-word retrieval and later sentence reading, not more rereading of the familiar book.

Primary 2: arithmetic facts versus number relationships

A student answers addition facts rapidly but cannot explain why 8 + 7 can be decomposed as 10 + 5. The tutor should preserve both pieces of evidence. Fact retrieval may be Stable while decomposition and relational number sense are Fragile. The next task uses ten frames and equations, then removes the representation. Speed drills alone would improve the capability that is already strong.

Primary 3: Science observation versus inference

A learner writes “the plant is unhealthy” when asked to record what can be observed. Rather than marking the entire process-skill question wrong and moving on, the tutor samples two different observation contexts. If the learner repeatedly mixes interpretation with observation, the next move teaches the evidence boundary. A changed context later tests whether the rule travelled beyond plants.

Primary 4: comprehension literal meaning versus inference

A learner finds explicit details accurately but gives vague inference answers. The tutor assesses whether the learner notices the clue, connects clues across sentences, or simply writes an unsupported plausible idea. The next teaching move depends on which step fails. One “inference” mark cannot identify the mechanism by itself.

Primary 5: fraction, ratio and percentage dependencies

A broad mark drop appears across three multiplicative topics. Instead of creating three separate remediation plans, the tutor samples one fresh item from each topic and notices the learner repeatedly misidentifies the reference whole. The assessment suggests a shared dependency. The next teaching move repairs reference-quantity reasoning across contrast cases, then retests each topic. If errors diverge later, the profile can split.

Primary 6: topical competence versus PSLE integration

A student solves percentage questions correctly on a topic sheet but misses similar structures in a mixed paper. Assessment should not downgrade percentage knowledge automatically. The evidence suggests method selection or paper integration is weaker than execution. A fresh mixed section becomes more informative than another percentage worksheet, and Examination Craft may become the next owner.

Secondary 1: subject gap versus transition load

A learner’s first report is lower across several subjects. Assessment begins with subject scripts, but also checks homework start time, task tracking, CCA load and independent planning. If knowledge probes are mostly stable while missed tasks and late starts are common, the next teaching move belongs to Learning Runtime. Adding three subject tuition programmes would solve the wrong problem.

Secondary 2: algebra that works only when method is named

A learner completes factorisation exercises accurately but fails algebraic fractions because factorisation is not recognised as the first move. The formative assessment removes topic headings and mixes expansion, factorisation and simplification. Execution remains strong; selection fails. The next lesson teaches structure recognition and contrast, not more factorisation execution.

Secondary 3: multiple weak marks, one high-leverage dependency

Several subjects fall at mid-year. Assessment resists turning the report into one emotional verdict. English answer scope, Mathematics algebraic control and Science mechanism explanation are sampled separately. Algebra proves to be the highest-leverage dependency because it affects several current and future topics. The next plan prioritises it while other subjects receive narrower maintenance rather than simultaneous intensive intervention.

Secondary 4: A-Math content versus full-paper control

A learner performs strongly on chapter tests but loses large marks in prelim papers. Item analysis shows content accuracy early in the paper, then time loss around one hard item and rushed late work. The assessment routes the problem to leave-and-return, pacing and checking. Relearning every A-Math chapter would create volume without addressing the mechanism.

Cross-stage: the learner who always needs confirmation

A learner produces correct work but looks at the tutor after every line. Assessment asks whether the subject capability or self-verification is weak. When the tutor delays confirmation and asks for one checking method, the learner can verify many answers independently. Subject performance may already be Stable while checking confidence and confirmation dependence remain Fragile. The next move is support fading, not harder subject work.

Cross-stage: “I don’t know” as an ambiguous assessment signal

“I don’t know” can mean missing knowledge, unclear task interpretation, fear of committing, low retrieval or learned prompt dependence. A useful assessment asks for the first thing the learner does know, allows a short independent attempt, and changes one condition if needed. The phrase itself should never be entered as a diagnosis.

What these cases have in common

• The total mark starts the investigation but rarely finishes it.
• The first weak link matters more than the most visible downstream error.
• Support conditions change what success means.
• Fresh evidence is stronger than a corrected original.
• Delayed and changed-condition evidence determine whether learning held.
• Assessment can justify less support as well as more support.
• The next owner may be a subject hub, Learning Runtime, Tutor System, Examination Craft or recovery.

Assessment should shrink uncertainty, not expand the learner’s problem list

A good assessment process leaves the learner with a smaller, clearer next job. It does not transform one bad paper into ten simultaneous deficits. If the evidence review produces a longer and more alarming list without prioritisation, the system has collected information but has not yet improved teaching.

Assessment across the lesson timeline

Assessment for teaching becomes easier to manage when each phase has a different evidence job. The tutor does not need to collect everything at once.

Before the lesson: prerequisite and retrieval evidence

Use a short fresh task to see what returned from previous learning and whether the planned lesson assumes knowledge that is not currently available. The result can shorten the lesson, trigger one prerequisite repair or confirm that the class is ready to move forward.

During explanation: interpretation evidence

Ask the learner to predict, compare or complete one step. The goal is to see whether the representation and relationship are becoming available—not to certify independent mastery while the explanation is still doing much of the work.

During guided practice: support-level evidence

Record what level of support unlocks the route. A strategic prompt and a full worked example are different evidence conditions. Use guided practice to discover how much support is currently productive and which support can begin to fade.

During independent practice: ownership evidence

Give enough uninterrupted work for the learner to choose, monitor and recover. The tutor should resist correcting too quickly. Independent practice is where hidden prompt dependence, checking habits and method-selection capability become visible.

At the end: exit evidence

Use one fresh item that contains the decisive lesson move. The exit task should inform the next lesson, not merely repeat the easiest practice item. If the learner still needs the scaffold, record that and plan the next fading step.

Between lessons: durability evidence

Homework, retrieval or the next lesson’s opening task can show whether the capability survived time. Do not assume that a strong exit task guarantees next-week access.

The six-week evidence calendar

The calendar is an example, not a fixed programme. Some capabilities move faster; others need longer. The logic is what matters: baseline → learning → delay → variation → integration → review.

Three-student orchestration: assessment without turning the class into three private lessons

A small group of three allows shared teaching and individual evidence to coexist. The tutor can use one target, one model and one discussion while still sampling each learner’s first attempt and exit performance.

• Shared prerequisite: all three answer independently before discussion.
• Shared model: tutor explains one relationship and asks each learner for a different representation or reason.
• Branch point: evidence decides whether one learner gets prerequisite repair while others practise transfer.
• Peer comparison: learners compare methods after independent evidence has been preserved.
• Exit check: each learner gets a fresh item matched to the same underlying capability.

Assessment should tell the tutor when shared instruction remains efficient and when individual branching becomes necessary. Equal headcount does not require identical tasks at every moment.

Contradictory evidence: do not average away the problem

A learner may score highly at school and struggle in tuition, or perform strongly in discussion and weakly in writing. These are not necessarily contradictions to be averaged. They may reflect different constructs or conditions.

When evidence conflicts, list the differences: task type, support, timing, familiarity, modality, difficulty and learner state. Then create one bridge task that changes the most plausible condition while holding the target as stable as possible.

Disconfirming evidence makes assessment stronger

Once the tutor has a hypothesis—“the learner’s problem is algebraic sign control”—look deliberately for a case where that explanation might fail. If sign errors disappear under a different representation but method-selection errors remain, the profile should change. Good assessment tries to prove itself wrong before becoming a strong conclusion.

When to stop assessing and teach

Assessment has diminishing returns. Once the missing relationship is sufficiently clear, teach it. Continuing to ask diagnostic questions after the decision is already obvious wastes learner attention and can feel like interrogation.

The stopping rule is simple: when additional evidence is unlikely to change the immediate teaching move, switch from diagnosis to instruction. Return to assessment on fresh work after the learner has had a real chance to learn.

When to stop teaching and assess

The reverse boundary matters too. Tutors can keep explaining because teaching feels useful even after the learner is ready to act. Once the model is clear enough, remove the explanation and create a fresh attempt. Otherwise the tutor never discovers whether learning is independent.

Assessment after feedback

Feedback should change the next attempt, not simply produce agreement. Ask the learner to use the feedback on a fresh item where the same principle matters. If the learner can identify and apply the principle without being reminded, the evidence is stronger than a polished correction.

Assessment after modelling

After a worked example, use a near case to check whether the learner understood the structure, then a changed case to see whether the idea can transfer. Avoid using an identical template and interpreting successful imitation as flexible capability.

Assessment after scaffolding

Reduce one support at a time when possible. If performance remains strong, continue fading. If the learner returns to Blocked, restore the smallest support needed and identify whether the scaffold was carrying part of the target.

Assessment after retrieval practice

Retrieval practice is itself learning, so later assessment should not simply count the repeated retrieval attempts as independent confirmations. Use a delayed fresh item and, where relevant, a changed context.

Assessment after transfer practice

If the learner succeeds across several changed contexts, ask whether the capability now survives mixed work where the relevant method must be selected among competitors. Transfer is stronger when the learner recognises when to use the capability, not only how to use it once cued.

Assessment after full-paper practice

A full paper should generate two outputs: a score and an error/decision map. The map should separate content gaps, retrieval, method selection, time loss, checking, question-order decisions and recovery. The next revision plan should follow the map, not merely the percentage.

The minimum evidence package before a major route change

• At least one recent authentic piece of work.
• At least one fresh task chosen to test the suspected mechanism.
• Support and access conditions recorded.
• The first weak link identified provisionally.
• One piece of evidence that could contradict the interpretation.
• A defined next intervention and fresh retest.
• A review point for continue, change, reduce or stop.

Major route changes deserve more than one bad mark or one adult impression. The package does not need to be bureaucratic; it needs to be sufficient for the decision.

Assessment handover between tutors

When a learner changes tutor, hand over current high-leverage evidence rather than a history of every mistake. Include target capability, task samples, support conditions, current state, what was tried, what changed and the next planned retest. The receiving tutor should verify the model rather than inherit it as truth.

Assessment handover from Primary to Secondary

Old labels should not cross the transition unquestioned. “Good at Maths”, “weak writer” or “careless” may no longer describe the learner under new subject and workload conditions. Handover should preserve portable evidence: what is stable, what is fragile, what support is still needed and what new environment should retest.

Assessment handover into examination season

As examinations approach, narrow evidence gathering to decisions that can still change performance. Protect strong capabilities, repair high-leverage gaps, check retrieval, and integrate paper control. Do not reopen every old profile merely because the examination is important.

Assessment handover after a bad result

Compare the result with the pre-exam model. Which predicted fragilities appeared? Which stable capabilities failed? Which strengths survived? Which new pattern emerged? If the result has destabilised ordinary learning, recovery may need to happen before normal assessment cycles resume.

The final implementation checklist

• Decision: what decision will this evidence improve?
• Construct: what capability is actually being sampled?
• Opportunity: did the learner have a fair chance to demonstrate it?
• Support: what prompts, tools, peers, AI or access conditions were present?
• Freshness: is this a new task or a corrected/rehearsed one?
• Pattern: does the result repeat across enough evidence for the claim?
• Mechanism: where is the first weak link?
• Action: what is the smallest teaching change justified?
• Acceptance: what fresh evidence will show the change held?
• Release: what evidence will justify reducing or stopping support?

The final WEP120 acceptance protocol

WEP120 is complete when assessment reliably feeds a next teaching move and the next teaching move feeds a fresh evidence cycle. The route is evidence question → bounded task → known conditions → interpretation → smallest instructional response → learner action → fresh retest → state/support update.

This page remains the public canonical owner for assessment used to improve teaching. Adaptive Teaching owns the changed response; Questioning owns live evidence elicitation; Feedback owns learner use of information; the Capability Profile holds learner-specific synthesis; Accessible Tasks protect target validity; Practice & Review and Learning Has Held supply the acceptance evidence. The Tutor Handbook retains professional technical gates. No competing generic formative-assessment article is needed.

From parent symptom to assessment question

Parents usually arrive with a symptom, not a construct. The assessment job is to translate the symptom into a bounded question that can change the next move. The table below does not diagnose the learner. It shows what useful evidence might be gathered first.

The symptom crosswalk is an entrance, not a diagnosis

The same symptom can come from different mechanisms. The table therefore points to an evidence question rather than a label. Start with the smallest useful sample, record support conditions, and allow the result to redirect the owner.

For the full parent symptom library, use Start Here. Once the evidence becomes learner-specific across several tasks, use the Capability Profile. Once the intervention is selected, hand off to the subject owner, Learning Runtime, Tutor System, Examination Craft or recovery route as appropriate.

A final rule for parent-facing assessment

Translate worry into a testable question without turning the child into the problem. “My child is careless” becomes “Which error repeats and under what conditions?” “My child is weak” becomes “Which capability fails first, with how much support?” “My child needs more tuition” becomes “What teachable bottleneck would additional support solve, and what fresh evidence would let us reduce it later?”

Assessment review windows: how much recent evidence should count?

A review should read enough recent evidence to see a pattern without allowing stale history or one dramatic lesson to dominate. The right window depends on the capability. A newly repaired algebra step may need several checks over two weeks. A stable reading capability may need much less frequent review. Examination execution may be reviewed after each full paper during the final phase.

The review window should shorten when capability is volatile and lengthen when evidence is stable. Old failures remain relevant only if they still change the current decision.

When to retire an assessment profile

Retire a profile when special monitoring no longer changes the learner’s route. A capability that survives delay, variation and the relevant performance conditions can return to ordinary review. The history may remain minimally for continuity, but it should stop acting like a current weakness.

A closure note can be simple: “Ratio representation profile retired after three fresh changed-context checks across two weeks and stable mixed-paper performance. Reopen only if the same representation failure repeats.”

When to reopen a profile

Reopen when the same high-leverage weak link reappears across fresh evidence, a transition creates a genuinely new demand, a support condition changes, or a previously stable capability fails under a relevant new context. Do not reopen merely because one isolated error resembles an old problem.

The assessment governance checklist

• Owner: which page or tutor system owns this educational decision?
• Purpose: what decision will the evidence improve?
• Construct: what capability is actually being sampled?
• Opportunity: did the learner have a fair chance to demonstrate it?
• Conditions: what timing, support, access, tools, peers or AI were present?
• Freshness: is this new evidence or a rehearsed/corrected item?
• Pattern: is there enough repeated evidence for the breadth of the claim?
• Mechanism: what is the earliest weak link or strongest stable component?
• Uncertainty: what can the current evidence not yet establish?
• Action: what is the smallest next teaching move justified?
• Acceptance: what fresh evidence would show the move worked?
• Release: what evidence would justify reducing, stopping or retiring support?

The evidence-to-action sentence

A useful formative record can often be compressed into one sentence: “On [task] under [conditions/support], the learner demonstrated [capability] but the first weak link appears at [mechanism]; [uncertainty] remains; next we will [teaching move] and check it with [fresh retest].”

Example: “On four mixed fraction problems without a model, the learner calculated accurately once the whole was identified but selected the wrong whole in two changed-context items; only two transfer cases were sampled; next we will contrast changing-whole structures and retest on a fresh problem after three days.”

The smallest useful assessment

Assessment should be no larger than the decision requires. If the uncertainty is whether a learner remembers one prerequisite, one or two fresh items may be enough to decide whether to proceed. If the decision is whether to reduce a long-running intervention, evidence should be broader: delayed, varied and independently produced.

This proportionality protects teaching time. The purpose of assessment is to improve learning, not to consume it.

The strongest useful assessment is often not the hardest

A hard question can produce a dramatic failure without clarifying why. A carefully chosen contrast pair or prerequisite item may provide far more diagnostic information. Difficulty should serve the evidence question, not the tutor’s desire to challenge.

Assessment should preserve what is already strong

When a learner struggles, adults often focus so heavily on deficits that strong capabilities disappear from the plan. Assessment should state what is stable and should therefore be maintained rather than retaught. This reduces workload and protects learner confidence with accurate evidence.

Assessment should expose hidden support without shaming it

Prompts, parent reminders, peer explanations, AI assistance and reference materials can all be legitimate parts of learning. Record them because they change the interpretation, not because help is a failure. The next question is whether the learner should continue using that support, fade it, or carry more of the step independently.

Assessment should protect legitimate access

Progress does not require removing access conditions that are not the learning target. A learner can become more capable while using stable legitimate supports. The assessment job is to keep target, access and instructional scaffold distinct enough that growth is interpretable.

Assessment should make future teaching simpler

The best evidence shrinks the space of plausible next moves. “Weak in English” creates many possibilities. “Passage meaning is stable; answer scope is fragile on inference questions” creates a focused teaching plan. “Careless in Maths” is vague. “Sign errors recur only when equality transformations are compressed” is actionable.

Assessment should eventually transfer to the learner

A mature learner begins performing parts of the assessment loop internally: noticing uncertainty, checking work, comparing evidence, choosing practice, asking precise questions and deciding when a skill needs another return. The tutor’s external assessment gradually becomes learner self-regulation.

This is the deepest release condition. The learner no longer waits for an adult to identify every weak link because they can recognise enough of the evidence to make a useful next move.

WEP120 final operating contract

Assessment for teaching is complete only when evidence changes what happens next and the next move is later checked on fresh work. The canonical cycle is: decision → target → task → conditions → observation → interpretation → smallest response → learner action → fresh retest → update or release.

Questioning supplies live evidence. Adaptive Teaching changes instruction. Feedback gives the learner an action. Accessible Tasks preserve the target under fair conditions. The Capability Profile holds learner-specific synthesis. Practice & Review chooses the next task. Learning Has Held checks durability and transfer. Examination Craft owns paper integration. Assessment for Teaching connects those owners without duplicating them.

One last rule

If an assessment result does not make teaching, practice, support or learner judgement more accurate, collect less data and return attention to learning.

Choosing assessment resources from the evidence job

Worksheets, school papers, assessment books, teacher-made questions, oral prompts, digital quizzes and AI-generated tasks can all provide useful evidence. None is automatically the best assessment resource. Choose the resource whose design matches the evidence question.

If the tutor needs to know whether a learner can select a method, a topic-labelled assessment book page may be weak evidence because the heading supplies the category. If the tutor needs delayed retrieval, a short fresh question can be stronger than a full paper. If the issue is examination integration, a school-style mixed paper becomes more relevant than isolated practice.

School worksheets

School worksheets are valuable because they reflect the learner’s actual curriculum and classroom methods. They may also contain repeated examples, teacher cues or familiar formats. Use them to understand school expectations, then add fresh evidence where independence or transfer is the decision.

Assessment books and question banks

Large question banks are useful for sampling cases and building practice. They can create false confidence when every page is sorted by topic and method. Mix selected questions deliberately when method selection matters, and check that difficulty or wording has not changed the intended construct.

Past and practice papers

Full papers are strong evidence for integrated examination behaviour but inefficient tools for locating every prerequisite. Use them to reveal patterns, then zoom into the mechanism with smaller diagnostic tasks. After repair, return to a fresh mixed section or paper to test whether the capability rejoined the whole system.

Tutor-made questions

Tutor-made questions are powerful because they can isolate one suspected weak link or create a controlled contrast. Their weakness is hidden bias: the tutor may unintentionally make the question easier, harder or too similar to the explanation. Reserve some items before teaching and ask whether another tutor would agree that the target stayed the same.

AI-generated questions

AI can draft variants quickly, but the tutor must still verify correctness, syllabus fit, difficulty, distractors, representation and answer key. A fresh-looking AI item can still be structurally identical to the model. Use human judgement to decide whether it truly tests retrieval, transfer or method selection.

Retest design after the teaching move

The retest should be chosen when the intervention is planned, not improvised after the learner appears to improve. Knowing the acceptance evidence in advance prevents the tutor from choosing an easy success case after the fact.

Retest failure is information, not proof the learner did not try

If the learner fails the fresh retest, compare the failure with the original profile. Did the same weak link return? Did the intervention fix one step but expose another? Did the support fade too quickly? Did the task move from familiar to transfer conditions? The failed retest should sharpen the model rather than trigger a generic instruction to work harder.

Retest success should change the route

When the learner succeeds independently on fresh work, the teaching system should respond. Reduce unnecessary repetition, increase spacing, vary the context or retire the special profile. Continuing the same intervention after acceptance evidence can create dependence and waste learning time.

The final quality test for Assessment for Teaching

A strong assessment-for-teaching system can answer four questions at any review point: What are we trying to know? What evidence do we actually have? What is the smallest justified next move? What fresh evidence will tell us whether that move worked?

If those answers remain clear from baseline through retest and release, assessment is serving teaching rather than becoming a parallel bureaucracy of marks, categories and records.

Assessment evidence for start, continue, change, reduce and stop decisions

Families and tutors often make support decisions from marks, attendance or anxiety. Assessment for Teaching improves those decisions by asking what capability actually needs support, whether the current intervention changes fresh evidence, and whether the learner is carrying more of the route over time.

Start support

Start when there is a repeated teachable weak link, ordinary school/home learning is not reliably repairing it, and the proposed support has a defined job. A low score alone is not enough. The evidence should show what the support will teach and what later fresh performance would justify reducing it.

Continue support

Continue when fresh evidence is improving but the learner still needs the current support level. Look for reduced prompting, stronger delayed retrieval, better transfer or more reliable paper performance. Attendance and completed worksheets are exposure evidence, not sufficient continuation evidence.

Change support

Change when a fairly implemented route produces little independent improvement, when the identified weak link proves wrong, or when the intervention solves one problem while creating another. Before changing tutor or programme, decide whether the teaching function, task design, frequency, support level or owner is what actually needs to change.

Reduce support

Reduce when the learner succeeds with fewer instructional prompts and can reproduce the capability after delay or variation. Reduction can mean fewer sessions, narrower subject scope, lighter homework, less correction or more independent work before feedback. It is a positive evidence-led decision, not withdrawal of care.

Stop support

Stop special intervention when ordinary learning can maintain the capability and continued support no longer changes the route. Record the closure evidence and a return condition. A repaired weakness should not remain active forever merely because it once mattered.

The release question

At every review, ask: what can the learner now do without us that previously required us? That question turns assessment from a system for finding deficits into a system for transferring control.

The final canonical rule

Assessment for teaching should leave the learner with a clearer next move and, over time, fewer unnecessary supports. Evidence that never changes teaching is record keeping. Teaching that never returns to fresh evidence is assumption. The canonical loop keeps both accountable to the learner’s actual work.

Assessment should become learner-owned over time

The mature endpoint of assessment for teaching is not a learner who waits for adults to measure every weakness. It is a learner who can increasingly notice evidence, locate uncertainty, choose a checking method, recognise when a skill needs another return, and ask for help precisely enough that support does not replace the first attempt.

That transfer can begin with small routines. Before feedback, the learner predicts which answer is least secure. After feedback, the learner states the rule or relationship that changed. Before a retest, the learner explains what condition will be different. After the retest, the learner compares the result with the previous profile and proposes the next practice type.

Self-assessment is strongest when it remains calibrated against real performance. A learner may feel confident and be wrong, or feel uncertain and be correct. The job is not to manufacture confidence; it is to build accurate self-trust from evidence. Over time, the learner should become better at distinguishing “I have not learned this”, “I know this but cannot retrieve it reliably”, “I can do it on familiar work”, and “I can use it under changed or examination conditions”.

The tutor can support this transfer by making assessment language explicit without turning every lesson into a formal review. Questions such as “What evidence do you have?”, “What support did you need?”, “What changed from last time?”, “What should you test next?”, and “What can you now do without me?” gradually become internal questions.

When learners can make those distinctions responsibly, assessment stops being something done to them and becomes part of independent learning control. That is the strongest evidence that the teaching system is preparing to reduce its own role.

Assessment closure: know when enough evidence is enough

A teaching system can become less accurate when it keeps collecting evidence after the decision is already clear. Once the learner state, first weak link and next move are sufficiently supported, stop diagnosing and let the learner learn. Additional testing should wait until the intervention has had a real chance to change performance.

The closure rule protects both time and dignity. The learner should not feel permanently under examination, and the tutor should not confuse more data with better judgement. Record the current interpretation, name what remains uncertain, choose the next teaching move, and define the fresh retest. Then return attention to instruction, practice and independent work.

When the retest arrives, reopen the evidence question rather than defending the old conclusion. Stronger fresh evidence can confirm, narrow, split, overturn or retire the profile. Assessment for teaching remains useful precisely because it is temporary, revisable and subordinate to learning.

One final operating principle

The best assessment leaves less uncertainty and more learner control. It should help the tutor teach more precisely, help the learner understand what to do next, and make unnecessary support easier to remove. When evidence becomes clearer but the next move remains vague, the assessment process is not yet finished.