How Learning Diagnosis Works | From Visible Difficulty to the First Useful Weak Link

Direct Answer: Learning diagnosis works by starting with what we can actually observe, generating more than one plausible explanation, then using the smallest useful test to separate those explanations before deciding what to teach or practise next. A weak score, slow answer or repeated mistake is a symptom. Diagnosis asks what mechanism produced it.

The simplest definition of learning diagnosis

Learning diagnosis is the disciplined process of finding the earliest useful point where independent performance breaks.

In one line: Do not repair the loudest symptom until you know what produced it.

The core mechanism

OBSERVE → BOUND THE PROBLEM → GENERATE HYPOTHESES → FIND WHAT WOULD DISTINGUISH THEM → RUN A SMALL TEST → UPDATE → LOCATE THE FIRST USEFUL WEAK LINK → INTERVENE → RETEST

This matters because the same visible outcome can be produced by very different causes. Two students can both score 55%. One may have weak foundational knowledge. Another may know the content but fail under mixed-topic selection. A third may lose marks through time pressure or misreading. A fourth may understand the topic but depend on cues that disappeared in the examination.

The Education Endowment Foundation describes diagnosis as identifying a problem or barrier so that teaching can be adapted to what pupils are actually bringing to the classroom. That is close to the central eduKate rule: find before you fix.

The seven jobs of good learning diagnosis

1. Start with an observation, not a label

“Careless,” “weak,” “lazy,” “not a Math person” and “poor memory” are interpretations, not observations. A better starting point is narrower: the learner omitted units three times; solved routine algebra correctly but chose the wrong method in mixed questions; could explain the concept with notes but not retrieve it the next day.

Specific observations keep diagnosis correctable. They also prevent one performance from becoming an identity.

2. Separate the visible error from the earlier cause

The line where an answer becomes wrong is not always the point where learning first failed. A final calculation error might have begun with a poor representation. A weak composition might begin with idea selection rather than grammar. A Science explanation may fail because the learner cannot connect evidence to mechanism even though the facts are known.

Finding the First Weak Link in Learning makes this distinction explicit.

3. Hold multiple explanations at once

Premature diagnosis creates premature intervention. If a student is slow, possible explanations include weak retrieval, excessive working-memory load, uncertainty about the route, careful checking or insufficient fluency. The observation “slow” does not choose among them.

Good diagnosis therefore keeps a small hypothesis set alive until evidence separates it.

4. Use a discriminating test

A discriminating test is not necessarily a formal test. It is the smallest observation that would make one explanation more plausible than another.

If we suspect retrieval rather than understanding, close the notes and ask for the concept. If we suspect representation, present the same relationship as a diagram instead of words. If we suspect support dependence, remove one hint. If we suspect timing, compare untimed and timed performance.

The Curie Series Discriminating Test is built around this exact move.

5. Distinguish state from capability

A learner can know something and fail to show it under particular conditions. Fatigue, pressure, distraction, unfamiliar format and timing can alter expressed performance. Conversely, strong performance under heavy support can exaggerate independent capability.

Bolt 20 — State Is Not Ability protects diagnosis from turning one state-dependent result into a global conclusion.

6. Keep uncertainty visible

Sometimes the evidence is insufficient. That is not failure. “We do not yet know whether this is retrieval or representation” is often a stronger statement than a confident but unsupported diagnosis.

Curie’s Observed, Inferred, Unknown separates what was directly seen from what we are inferring and what remains unresolved.

7. Retest the mechanism, not only the score

An intervention is not proven because the next mark rose. The task may have been easier, the student may have remembered the format, or support may have increased. After repair, test the suspected weak operation again and then examine whether the improvement survives a fresh problem.

This is where diagnosis joins Practice, Feedback and Transfer.

What learning diagnosis is not

Diagnosis is not a label. “Weak student” compresses too much.
Diagnosis is not one score. A score is evidence about one performance under particular conditions.
Diagnosis is not guessing the most common cause. Common explanations still need evidence in this learner.
Diagnosis is not maximum testing. More data is useful only when it can change the decision.
Diagnosis is not permanent. Learner states change, so useful models must update.

Five common diagnostic confusions

Visible symptom	Possible explanations	Small useful test
“Forgets everything”	Weak retrieval, cue dependence, shallow encoding, insufficient spacing	Delayed retrieval with and without familiar cues
“Careless mistakes”	Weak checking, overload, poor representation, speed pressure	Slow one task down and require explicit verification
“Can do homework, cannot do tests”	Support dependence, transfer failure, mixed-topic selection, pressure	Fresh unsupported mixed set under gradually more realistic conditions
“Understands but cannot explain”	Shallow relational understanding, vocabulary limits, retrieval weakness	Ask for a causal explanation without notes, then probe the missing link
“Needs constant reminders”	Task ambiguity, weak first-action control, planning dependence, attention capture	Make the goal explicit and transfer one bounded decision

How diagnosis works across subjects

English: A weak answer may originate in vocabulary, reading accuracy, inference, evidence selection, sentence control, organisation or task interpretation. The visible writing problem is not always a writing-origin problem.

Mathematics: A wrong answer may originate in representation, concept, relationship, fluency, method selection, execution or checking. The final arithmetic line can be downstream of a much earlier modelling failure.

Science: A weak response may originate in factual knowledge, system reconstruction, variables, evidence, mechanism, explanation or examination interpretation. Re-teaching facts will not repair every Science error.

For parents: what should I bring to a diagnosis?

Bring evidence that preserves the learner’s actual performance: recent school papers, marked work, teacher comments, examples of repeated errors and information about what support was present. A single total score is useful, but the written work often shows where the route changed.

Our parent guide What Should Parents Bring to a Tuition Consultation? explains why the artifact is often more useful than the story we tell about the artifact.

How do we know the diagnosis was useful?

It explains more of the observed pattern than competing explanations.
It predicts where performance should improve after a targeted change.
The targeted intervention improves the suspected mechanism.
The improvement survives a fresh task.
The diagnosis can be revised when new evidence disagrees.
It produces a better next action than the undifferentiated symptom did.

The complete diagnosis chain

OBSERVE → DESCRIBE WITHOUT LABEL → DECOMPOSE THE PERFORMANCE → HOLD COMPETING EXPLANATIONS → FIND THE SMALLEST DISCRIMINATING TEST → UPDATE → LOCATE THE FIRST USEFUL WEAK LINK → REPAIR → RETEST → TRANSFER → UPDATE AGAIN

Frequently asked questions

Can one test diagnose a learner?

Usually only partially. One test can reveal useful patterns, but its conclusions are bounded by the content sampled, task conditions, marking and the learner’s state. Strong diagnosis combines the work itself with targeted follow-up observations.

Why not just give more practice first?

More practice is sensible when the weak operation is already known. When it is not, volume can exercise the wrong thing. Diagnosis helps make practice more selective.

Is diagnosis only for weak students?

No. Strong learners also have bottlenecks, hidden dependencies and transfer limits. Diagnosis can identify the next constraint even when overall performance is high.

Can a diagnosis be wrong?

Yes. That is why good diagnosis remains provisional and testable. New evidence should be able to correct the model.

Research bridge

For a broader evidence perspective on diagnosing pupil needs and adapting teaching, see the Education Endowment Foundation discussion of diagnosis and adaptive teaching.