MATHEMATICS DIAGNOSIS · FIND THE EARLIEST WEAK LINK
Do not diagnose the chapter. Diagnose where the route first becomes unreliable.
A low mark, a difficult chapter or a loss of confidence is a signal. It tells us that something is wrong. It does not yet tell us what is wrong.
At Bukit Timah Tutor, Mathematics diagnosis means tracing the learner’s work backwards until we find the earliest important concept, representation, decision or prerequisite that is no longer carrying the later work reliably. We then repair that point, reconnect it to the current syllabus and test whether the repair survives when the question changes, support reduces and time passes.
Current failure → gather evidence → narrow the possibilities → probe → locate the earliest unstable connection → repair → reconnect → verify.
Diagnosis is a process. A test is one possible instrument.
There is no rule that every student must begin with the same standard paper. Sometimes a recent school script already exposes the problem. Sometimes the tutor needs to watch the student attempt a question. Sometimes a conversation reveals a useful history but not the mechanism. Sometimes a short set of deliberately chosen questions is needed to distinguish between two competing explanations.
Conversation
Level, syllabus, recent results, repeated difficulty, school sequence, study behaviour and what the student experiences when the work becomes difficult.
Evidence
Marked papers, homework, corrections and the student’s own written working. The first wrong line is often more informative than the final wrong answer.
Probe
A few questions chosen to discriminate between possible causes. The purpose is not to generate a score. It is to make the hidden break easier to see.
Observation
How the learner reads, begins, hesitates, chooses a representation, executes, checks, responds to a prompt and recovers after an error.
These are not four competing diagnostic systems. They are four sensors feeding the same question: where does the route first become unreliable?
The visible mistake may have several different causes.
Two students can produce the same wrong answer and require different teaching. That is why the intervention should follow the cause rather than the label on the chapter.
| Possible weak link | What it can look like | What the tutor needs to distinguish |
|---|---|---|
| Concept | The idea itself is not understood. | Can the learner explain what the operation or relationship means? |
| Representation | The student cannot move reliably between words, diagrams, symbols or equations. | Does the difficulty disappear when the same idea is represented differently? |
| Prerequisite | The current chapter exposes an earlier dependency. | Which earlier skill must be stable before the current method can work? |
| Retrieval | The student once knew the method but cannot bring it back when needed. | Is the knowledge absent, or merely unavailable without a cue? |
| Recognition | The student can solve a familiar form but cannot identify when the same idea applies elsewhere. | Can the learner recognise the structure when the surface changes? |
| Method | The intended route is known but the procedure is incomplete or incorrectly sequenced. | Where does the method first depart from a valid route? |
| Execution | The mathematics is known but signs, substitutions, algebra or working break during performance. | Is this conceptual failure or performance leakage? |
| Load | The student becomes slow, overloaded or loses track during multi-step work. | Does accuracy return when the task is decomposed or time pressure is removed? |
| Transfer | Single-topic practice works; mixed or unfamiliar questions fail. | Can the learner select and adapt the idea without the topic label? |
| Examination control | The student knows the Mathematics but loses marks under paper conditions. | Is the limiter retrieval, timing, reading, decision-making, checking or recovery? |
A diagnostic probe should discriminate, not merely measure.
Suppose a student sees 2(x + 3) = 14 and writes 2x + 3 = 14. Calling this “an algebra weakness” is too broad. A better diagnosis asks which explanation predicts the error.
- Can the student explain what 2(x + 3) represents?
- Can the student expand it correctly when explicitly asked to expand?
- Does the same error occur with 3(x + 5)?
- Can the student identify the mistake in somebody else’s working?
- Does the problem disappear when variables are replaced with numbers?
- Can the learner distribute correctly in isolation but fail only while solving an equation?
Each question changes the evidence. The aim is to eliminate explanations until the smallest useful repair becomes clear.
The smallest useful repair is not necessarily the smallest error. It is the earliest unstable connection whose repair restores the largest amount of downstream work.
A diagnosis is provisional until reality answers back.
The tutor’s first interpretation can be wrong. So diagnosis should not end when a plausible explanation is found. It should make a prediction: if this is the important weak link, repairing it should improve these later behaviours.
Repair
Teach or rebuild the suspected weak connection directly.
Reconnect
Return the repaired capability to the current chapter and school work.
Change the surface
Use a different-looking question so copied familiarity cannot masquerade as understanding.
Reduce support
Remove prompts and see whether the student can carry the decision personally.
Delay and return
Check again later. A repair that works only immediately after explanation is not yet stable.
If the predicted improvement does not appear, we revise the diagnosis. That is an important part of the philosophy: the learner’s later performance is allowed to contradict our first explanation.
The complete loop
- Signal — notice the visible problem.
- Gather evidence — conversation, work, papers and history.
- Narrow — identify competing explanations.
- Probe — ask only the questions needed to discriminate between them.
- Locate — identify the earliest important unstable connection.
- Repair — intervene at that point.
- Reconnect — return the repair to current Mathematics.
- Verify — change the question, reduce support and revisit later.
- Re-diagnose when necessary — if reality does not behave as predicted, change the explanation rather than defending it.
What this is — and what it is not
This is a structured teaching and mathematical error-analysis process. It uses evidence to decide where teaching should begin and whether a repair has worked.
It is not presented as a standardised psychometric assessment, clinical diagnosis or independently validated diagnostic test. A short diagnostic worksheet can be useful, but its role is to provide evidence for teaching decisions—not to turn one score into a complete description of a learner.
Diagnosis should make the next piece of work smaller, clearer and more justified. If it does not change what we teach next, it has not yet done enough.
DIAGNOSIS IS A HELP FUNCTION
Do not diagnose farther than HELP requires.
Diagnosis exists to improve the next teaching decision. If two remaining explanations would lead to the same safe intervention, more probing may add load without adding useful HELP. If they require different interventions, use the smallest discriminating probe that can separate them.
Before intervening, classify the current struggle as PRODUCTIVE, FRAGILE, STUCK, MISDIRECTED, OVERLOADED or BLOCKED. Productive struggle may require no help yet. Fragile progress may need only a cue. A blocked or misdirected route may justify explicit repair.
Diagnosis is not the destination. Better HELP is the destination.
THE DIAGNOSTIC CONSTITUTION · FIVE RULES
Diagnose only far enough to make the next teaching decision better.
Diagnosis is useful only when it reduces uncertainty enough to choose a smaller, better-justified intervention. More testing is not automatically more accurate.
1 · Minimum sufficient evidence
Use the least amount of evidence needed to distinguish the explanations that would lead to meaningfully different teaching. If a marked paper already exposes the break clearly, a second full test may add load without changing the decision.
2 · Confidence, not certainty
Describe a weak link as observed, strongly supported or still suspected. Do not turn a provisional explanation into a permanent label for the learner.
3 · Intervention as a test
A repair should make a prediction. If this weak link matters, repairing it should improve a defined downstream behaviour. Teaching therefore becomes part of diagnosis rather than something that begins only after diagnosis ends.
4 · Stop rule
Stop probing when the remaining uncertainty would not change the next safe teaching action. Begin the repair, observe the response and learn from what happens next.
5 · Reopen rule
Reopen the diagnosis when the repair does not generalise, the same failure returns, a new contradiction appears or the learner requires unexpectedly high support. Reality has supplied new evidence.
We are not trying to know everything about the student before teaching. We are trying to know enough to make the next intervention more correct—and to keep that intervention correctable.
The practical confidence ladder
| State | Meaning | What we do next |
|---|---|---|
| Signal only | We know a problem is visible, but not yet why. | Gather targeted evidence. |
| Suspected | One explanation fits, but plausible alternatives remain. | Use a discriminating probe or observation. |
| Supported | Several pieces of evidence converge on the same weak link. | Begin a bounded repair and predict the change. |
| Verified in use | The repair survives changed questions, reduced support and some delay. | Reconnect, automate and extend. |
| Contradicted | The expected improvement does not appear. | Reopen the diagnosis and revise the explanation. |
This prevents two opposite errors: under-diagnosis, where every problem becomes “practise more,” and over-diagnosis, where the learner is repeatedly tested even though the next useful teaching action is already clear.
DIAGNOSTIC RESOLUTION · NODE OR EDGE?
Sometimes the object is known. The connection is what fails.
The Mathematics Knowledge Warehouse now distinguishes node failures from edge failures. A node failure means the object itself—such as fraction magnitude, equality or function meaning—is unstable. An edge failure means two capabilities may each work alone but the learner cannot reliably connect them—for example, arithmetic distribution → algebraic expansion, ratio → gradient, or graph behaviour → derivative meaning.
The diagnostic engine should therefore ask two different questions: “Does the learner know this object?” and “Can the learner carry the relationship from this object into the next one?”
Do not reteach both rooms when only the corridor between them is broken.
DIAGNOSTIC ENGINE · EXECUTABLE PROBES
A suspicion becomes useful only when we know what evidence could separate it from another explanation.
The Diagnostic Probe Bank gives the engine stable question IDs such as ALG-DIST-001, FUNC-GRAD-001 and STAT-INF-001. Each probe owns a discrimination job: candidate hypotheses, what to observe, how different responses change the route, the smallest repair, a verification item and a stop rule.
This prevents diagnosis from becoming an improvised interview. The AI can record which uncertainty it was trying to reduce, what the learner actually did, and why the next question or repair is justified.
A probe earns its place only if a different answer can lead to a different next action.
DIAGNOSTIC ENGINE · VALIDATION HARNESS
A diagnosis can be useful without pretending to be scientifically validated.
The engine now separates research support for the method from evidence for the BTT implementation. Graph edges, probe IDs and interventions each carry their own evidence status. Agreement, transfer, retention and contradiction are tracked separately.
A diagnosis should therefore be expressed as a current best hypothesis with confidence, linked to the evidence that supports it, the intervention it predicts should help, and the condition that would force the diagnosis to reopen.
If the predicted improvement does not appear, confidence falls. The framework does not get to rescue itself by inventing a new hidden explanation without new evidence.
Differential diagnosis | One symptom, several possible causes
A diagnosis becomes useful when it separates explanations that would lead to different teaching decisions. “Weak in algebra” is not yet a diagnosis. The student may not understand equality, may know the concept but lose signs under symbolic load, may retrieve the wrong method, may misread the question or may perform accurately in untimed work but collapse under paper conditions. Each cause suggests a different intervention.
| Visible symptom | Possible cause | Small discriminating probe |
|---|---|---|
| Cannot start a word problem | Concept gap, representation gap, question-reading gap | Ask the learner to describe the quantities and draw or symbolise the relationship without solving it. |
| Correct method, repeated sign errors | Execution control, symbolic overload, weak checking | Give the same algebra with shorter lines and require verification after each transformation. |
| Can do chapter exercises but fails mixed work | Recognition or transfer gap | Mix two familiar methods and remove topic labels. |
| Understands when prompted, cannot recall later | Retrieval or retention weakness | Return after delay without notes or a model answer nearby. |
| Untimed work is strong, test work collapses | Paper control, pacing, pressure, retrieval access | Use a short timed set while keeping mathematical difficulty constant. |
Case 1 | The student who says “I understand in class”
This statement can describe several states. The learner may genuinely understand an explanation but be unable to retrieve it later. The learner may follow a teacher’s sequence without being able to generate the sequence independently. The learner may recognise a method after the first step is shown but not before. Or the learner may be using “understand” to mean that the explanation felt familiar.
A useful probe removes one layer of support at a time. First ask the student to explain the idea without solving. Then ask for the first step only. Then use a changed example. Finally return after delay. The point is not to catch the learner out. It is to identify which part of independence has not yet transferred.
Case 2 | The student labelled careless
“Careless” often describes the final appearance of an error, not its mechanism. Some mistakes really are lapses of attention. Others occur because working memory is overloaded, notation is compressed, the learner does not know which steps are risky, or checking is too vague to target the likely failure points.
Compare short and long versions of the same Mathematics. If accuracy is high on short work but collapses as several transformations accumulate, the intervention should improve line discipline and local checking rather than repeat the concept from the beginning. If the same misconception appears even in simple work, the concept or representation is the stronger suspect.
Case 3 | The student whose marks suddenly fall after a transition
A transition such as P6→Secondary 1, Secondary 2→3 or Secondary 4→JC changes the load placed on earlier Mathematics. A learner who looked secure may have relied on a level of prompting, chapter familiarity or arithmetic support that is no longer available. The correct response is not automatically “work harder.” Diagnose what the new stage is asking the old system to do.
Check whether earlier objects remain available under the new representation. Can the learner handle algebra without consuming all attention? Can graphs be interpreted rather than copied? Can a problem be translated before calculation begins? If the dependency is secure, investigate retrieval, workload and examination conditions instead.
False positives | When the earliest visible weakness is not the cause
Diagnosis can go wrong in both directions. A tutor may see slow arithmetic and assume it causes algebra failure when the student can in fact manage arithmetic well enough for the current task. Or a tutor may see an algebra mistake and reteach algebra even though the student actually misread a condition. The earliest weakness in history is not automatically the earliest relevant weak link.
The correct standard is functional: does this suspected weakness change the present performance? If improving the suspected dependency produces no change in the target task, the diagnosis should be reopened. Intervention itself becomes evidence.
Choose the smallest intervention that can test the explanation
- State the hypothesis. Example: “The student knows quadratic methods but does not recognise when factorisation is useful.”
- Predict what should improve. If recognition is the issue, a short contrast set should help more than another long explanation.
- Intervene narrowly. Teach the distinguishing cue and compare several cases.
- Retest on changed questions. Do not reuse the exact training examples.
- Reduce support. The student must identify the cue independently.
- Return later. If the improvement disappears immediately, retention may be part of the problem.
If the predicted change occurs, confidence in the diagnosis rises. If it does not, do not defend the label. Reopen the possibilities.
Evidence hierarchy | What should carry the most weight?
Different evidence answers different questions. A school result shows performance under one assessment. Marked working can reveal where validity first breaks. A student explanation shows current meaning. A short probe can separate two plausible causes. A delayed return tests retention. A changed surface tests transfer. No single source should carry more certainty than it deserves.
- Best for mechanism: the learner’s written working and live reasoning.
- Best for transfer: changed questions that preserve structure.
- Best for retention: delayed retrieval without immediate rehearsal.
- Best for examination control: short controlled timed tasks and full-paper evidence when appropriate.
- Best for history: school scripts, homework patterns, teacher comments and the learner’s own account.
A diagnosis should change teaching immediately
If the diagnosis does not alter the next teaching decision, it may be too vague. “Weak foundations” is only useful when it identifies which foundation and how it blocks present Mathematics. “Careless” is only useful when it identifies which control breaks and how to test it. “Needs confidence” is only useful when we know what successful independent action would rebuild confidence.
The next lesson should therefore look different because of the diagnosis: a different representation, a narrower dependency, a contrast set, a slower symbolic sequence, an untimed retrieval task, a timed access task or a changed surface that tests transfer.
When to stop diagnosing and teach
Diagnosis is not an end in itself. Stop when the evidence is sufficient to make a better next decision. A learner does not need an exhaustive profile before receiving help. If one short probe shows that fraction equivalence is blocking percentage work, repair it and see what changes. If the repair works, continue. If it does not, reopen the diagnosis.
This stop rule protects time and dignity. The learner is not a puzzle to be analysed indefinitely. Diagnosis exists to improve teaching.
Parent communication | Explain the mechanism, not a label
A useful parent update should say what evidence was observed, what explanation is currently most plausible, what will be tried next and what change would count as evidence that the repair is working. This is more informative than “needs more practice” or “weak in Maths.”
For example: “She can solve routine simultaneous equations accurately, but mixed questions show that she does not reliably recognise when two relationships should be modelled together. We are working on representation and recognition rather than reteaching elimination from the beginning. We will know the repair is holding when she can build the equations from changed contexts without a prompt.”
Exit criteria | A repair is not complete because one worksheet is correct
- The learner can explain the relevant relationship.
- The learner can execute the method without step-by-step prompting.
- The learner can recognise the structure when the surface changes.
- The improvement remains after delay.
- The repaired capability reconnects successfully to current schoolwork.
- The learner can detect or recover from a plausible error.
These exit criteria move diagnosis toward independence. The goal is not to prove that teaching occurred. It is to show that the learner can now carry more of the Mathematics alone.
Boundary | What Mathematics Diagnosis should not claim
This framework is educational diagnosis of mathematical learning and performance. It is not medical, psychological or clinical diagnosis. It should not infer intelligence, motivation, neurological conditions or mental health from Mathematics work. When a concern goes beyond educational evidence, the appropriate qualified professional remains the correct route.
Within Mathematics, uncertainty should remain visible. A diagnosis is a working explanation supported by evidence and tested through teaching—not a permanent label attached to the student.
Measure the repair, not the amount of teaching
A diagnosis is useful only if the resulting intervention changes performance in the predicted direction. This means progress should be measured against the suspected mechanism. If the problem was representation, can the learner now move between words, diagrams and symbols? If it was recognition, can the learner identify the method without a chapter label? If it was retrieval, is the idea still available after delay? If it was examination access, does the Mathematics remain available under controlled time?
Counting completed worksheets is therefore weak evidence. A large amount of practice can coexist with an unchanged failure mechanism. Better evidence is smaller and more specific: fewer prompts, cleaner first steps, better transfer, improved retention and successful reconnection to current schoolwork.
When should a diagnosis be reopened?
- The targeted repair produces little or no change in the target task.
- Performance improves only on examples that closely resemble the teaching examples.
- The learner can execute immediately after teaching but cannot retrieve the method later.
- The same symptom reappears in another topic with a different mathematical object underneath it.
- Untimed performance improves but timed performance does not, suggesting an access or paper-control layer.
- A supposedly weak prerequisite proves reliable under a direct probe, meaning the cause lies elsewhere.
Reopening a diagnosis is not failure. It is evidence that the first explanation was incomplete. A good diagnostic system is willing to update itself.
Five questions parents can ask without turning home into another classroom
- What kind of mistake keeps repeating?
- Does the student understand the idea when there is no time pressure?
- Can the student begin independently, or only after a hint?
- Does the improvement survive a changed question and a delay?
- What will the tutor look for to decide that the repair is complete?
These questions keep attention on mechanism and evidence rather than blame. They also make progress easier to discuss because the family knows what change should become visible.
Frequently asked questions about Mathematics diagnosis
Does every student need a formal diagnostic paper?
No. A recent school script, a short live attempt or a focused probe may already contain enough evidence. Use the least intrusive tool that can distinguish the plausible causes.
How far back should repair go?
Only as far as the earliest relevant weak link. Earlier Mathematics matters when it blocks the current task, not simply because it came first chronologically.
Can a student have more than one weak link?
Yes, but teaching still needs sequence. Start with the constraint most likely to improve the current system, then reassess. Trying to repair everything at once can make progress impossible to interpret.
What if the diagnosis changes after two weeks?
That can be a sign of better evidence. The first repair may expose a second layer that was previously hidden. Update the explanation rather than defending the first label.
The final diagnostic question: what can the learner now do without us?
A repair is strongest when the learner no longer needs the same external decision that was previously supplied by the tutor. If representation was taught, can the student now choose a representation? If recognition was trained, can the student identify the structure without a label? If checking was modelled, can the student decide where checking is worth the time?
This is why independence is part of diagnosis rather than an optional final stage. A learner can look improved while support is still doing invisible work. The diagnosis is better resolved when the target capability survives changed questions, reduced prompting and delayed return. That is the point at which the Mathematics—not the tutor’s scaffolding—is carrying the next step.
The best diagnosis eventually becomes unnecessary because the learner can perform the repaired decision independently.
When diagnosis needs an experiment: route into the BTT Mathematical Lab. Diagnosis keeps ownership of learner-state interpretation; MathLab is called only when competing explanations need a discriminating probe, controlled intervention or validation test before the state can be updated responsibly.
Mathematics routes: Mathematics Hub · Curriculum Overview · Complete Article Directory
State, feedback and repair
Before and after diagnosis: Find My Mathematics State, When Mathematics Slips, Mathematical Feedback, Secondary Diagnostic Progression.
Continue from diagnosis: Find My Mathematics State · Mathematics HELP Runtime · complete Mathematics directory.
Mathematics system route: Mathematics Hub · How Mathematics Works · Knowledge Warehouse · Learning Library · Examination Craft.
World Mathematics route: use the World Mathematics Atlas after diagnosis to locate the matching knowledge object, school route, examination or recovery path.