Evidence Conflict | Marie Curie Series | What to Do When the Signals Disagree

Quick Read

A learner gets a strong mark but needed heavy prompting during revision.

Another learner explains the concept beautifully but cannot retrieve it after three days.

A third learner feels uncertain but works independently and accurately.

When the signals disagree, Curie does not force them into one verdict. It asks what each signal is actually measuring.

The One-Sentence Answer

Conflicting evidence should narrow the question rather than widen the label: identify which capability each signal reflects, change one condition at a time, and keep the conclusion provisional until the disagreement is explained.

Where Evidence Conflict Sits in Curie

Evidence Convergence asks when different signals support the same interpretation.

Evidence Conflict protects the other side of the system. It ensures we do not hide disagreement merely because a neat story would be easier to tell.

Disagreement Is Not a Measurement Failure

Sometimes disagreement is the most informative result.

If a learner performs well only when the topic label is visible, the conflict between familiar and mixed work reveals route dependence.

If confidence remains low while independent performance becomes strong, the conflict reveals calibration lag rather than academic weakness.

If immediate explanation is excellent but delayed retrieval is poor, conceptual access may be more fragile than classroom fluency suggests.

Common Evidence Conflicts

Signal A	Signal B	Question to investigate
High mark	Heavy prompting	How much of the route is independently owned?
Strong explanation	Weak delayed retrieval	Is the concept understood but not yet durably accessible?
High confidence	Weak transfer	Is familiarity being mistaken for general capability?
Low confidence	Strong independent execution	Has capability improved faster than self-perception?
Accurate untimed work	Fragile timed work	Which control degrades under pressure?
Good isolated skill	Poor integrated task	Does load expose an expensive lower-level operation?

Keep the Scope of the Claim Small

When signals conflict, broad labels become especially dangerous.

“Weak at Mathematics” explains almost nothing.

“Accurate once the method is selected, but mixed-question route selection is still unstable” is narrower, testable and repairable.

This is the practical value of Curie: disagreement should make the diagnosis smaller and more precise.

Change One Condition at a Time Where Possible

If we change timing, question type, support, difficulty and topic mixture all at once, a different result may still tell us very little.

1. Hold the underlying capability reasonably constant.
2. Change one meaningful condition.
3. Observe the difference.
4. Return to baseline if useful.
5. Repeat with a second condition only after the first comparison is interpretable.

This does not turn education into a laboratory experiment. It simply protects us from creating more ambiguity while trying to resolve ambiguity.

English: Strong Discussion, Weak Written Evidence

A learner may understand a passage deeply in conversation but produce a weak written answer.

That conflict should not automatically become “does not understand comprehension.” The break may sit in evidence selection, response structure, language control, memory under writing load or examination timing.

For developmental English stages, continue through the English Tutor route. Curie helps locate why two observations at the same stage disagree.

Mathematics: Good Procedures, Poor Mixed Questions

A learner may solve topic drills accurately yet fail mixed problems.

That is not contradictory once we recognise that execution and route selection are different capabilities.

For developmental Mathematics from Year 0 through University, continue through the Mathematics Tutor route; this page preserves the cross-subject distinction between what each signal measures.

Science: Memorised Language Can Conflict With Mechanism

A learner may use correct scientific vocabulary while the causal chain underneath is reversed or incomplete.

The wording signal says terminology is available. The mechanism signal says conceptual structure needs repair. Both observations can remain true.

Additional Mathematics: Correct Answer, Unstable Route

An A-Math learner can occasionally reach the correct answer through a fragile or unjustified route.

Curie therefore compares the result with method selection, algebraic legitimacy, boundary conditions and verification. The answer is evidence, but not the only evidence.

Confidence Conflict Deserves Special Care

Confidence Calibration already separates feeling ready from being ready.

Evidence Conflict strengthens that principle. If confidence and demonstrated capability diverge, do not simply decide one is “wrong.” Ask why they have drifted apart and what evidence would help them reconverge.

Repair: Turn the Conflict Into the Next Question

1. Name the two signals that disagree.
2. State what each signal actually measures.
3. Find the smallest hypothesis that could explain both.
4. Choose one discriminating task or changed condition.
5. Observe the result.
6. Update the interpretation without erasing the earlier evidence.

This keeps the evidence trail intact and makes disagreement productive.

Tutorial Is Where Conflicts Can Be Examined at High Resolution

Inside a Tutorial, the tutor can reproduce one condition while changing another: remove the prompt, delay the retrieval, change representation, mix the topic or ask for explanation before execution.

The aim is not to catch the learner out. It is to discover why two legitimate observations tell different stories.

What Tutor Should Say

The Tutor dashboard should be allowed to show unresolved disagreement:

“Your untimed work shows strong conceptual control, but retrieval becomes unreliable after delay. We do not yet have evidence that the whole capability is stable over time.”

That is more accurate than forcing a single green or red judgement.

What Parents Can Ask

• What exactly does the strong result show?
• What exactly does the weak result show?
• Did the conditions differ?
• Was more help present in one case?
• Does the disagreement repeat or was it a one-off?
• What single next check would distinguish the likely explanations?

Voyage: Development Is Often Uneven

Across the Voyage, capabilities do not all mature at the same rate. Conceptual understanding may outrun written control. Knowledge may outrun confidence. Accuracy may outrun independence.

Evidence Conflict helps us describe that unevenness without reducing it to contradiction.

Darwin: Disagreement Can Be the First Sign That the Environment Changed

The Darwin Series asks when an old method stops being sufficient. A conflict between earlier success and newer failure may be exactly the evidence that the environment has crossed a boundary.

Boundaries

• Do not hide inconvenient evidence to preserve a tidy narrative.
• Do not treat every variation as meaningful; ordinary noise exists.
• Do not combine measures that answer different questions and call them inconsistent.
• Do not use conflicting evidence as a reason to delay all action when one narrow repair is already justified.
• Keep conclusions proportionate to the evidence available.

The Core Rule of Evidence Conflict

When learning signals disagree, preserve the disagreement, identify what each signal truly measures, and use the smallest discriminating observation to discover whether the difference comes from capability, context, support, time or interpretation.

Frequently Asked Questions

Does conflicting evidence mean the tutor measured badly?

Not necessarily. Different tasks often expose genuinely different layers of the same capability.

Should the latest result replace the earlier result?

No. Ask whether the conditions changed and whether both results remain informative. New evidence should update the picture, not erase history automatically.

What is the best response to disagreement?

Narrow the claim and design one useful next observation that could separate the competing explanations.

---

Marie Curie Series · Evidence Convergence · Observed, Inferred, Unknown · Confidence Calibration · Learning Under Load · Pressure Test · Tutor

eduKate Sengkang | Small groups of up to 3 | 1.5-hour tutorials | Properly Taught Kids Shine a Bright Light Into the Future.