A learner gets the answer right.

That sounds like good evidence.

But what if the learner remembers the item?

What if the tutor used the same question during correction?

What if a parent supplied the first step?

What if the worksheet groups every problem by method, so the chapter title effectively tells the learner what to do?

The performance is still real.

But its meaning has changed.

Training contamination occurs when prior exposure, repeated items, hints, answer memory, coaching or environmental cues make observed performance look stronger than the learner’s independent capability would be on a fresh task.

---

Quick Read: Correct Does Not Always Mean Independent

Observed Success → Ask What Support or Familiarity Entered → Remove Contamination → Use Fresh Comparable Task → Retest Independently

Common contamination sources include:

• repeating the exact same item;
• remembering a previous answer;
• seeing the worked solution before retesting;
• receiving a method hint;
• having the chapter title identify the technique;
• parental prompting;
• tutor cueing;
• practising the exact test form repeatedly;
• memorising model answers without transfer.

Contamination Is Different From Support

Support is not bad.

Worked examples, hints and guided practice are legitimate teaching tools.

Contamination becomes a problem when supported performance is interpreted as though it were independent performance.

Mira gets a quadratic correct after the tutor says, “Try factorisation.”

Useful learning event.

Weak evidence of independent method selection.

Label the condition honestly.

Then retest fresh.

Repeated Exposure Can Inflate Scores

Research on repeated test-item exposure has long shown that prior exposure can contaminate score interpretation by producing gains specific to repeated items rather than equivalent gains on new items. One study indexed in PubMed compared repeated and unique questions across certification-exam attempts and found evidence that repeated exposure could inflate performance on common items.

The setting is adult professional certification, not school tuition, so the size of the effect should not be transferred directly.

The principle is still important:

If the learner has seen the exact item, success may reflect item memory in addition to the capability we hoped to measure.

The Same-Question Trap

Mira gets Question 4 wrong.

The tutor explains it.

Mira repeats Question 4 correctly.

Good.

But the answer may still be supported by memory of the correction.

Now use a fresh parallel item.

This is why Training Retests insists that correction is not verification.

Hidden Method Cues

A worksheet titled “Factorisation Practice” tells the learner more than the question does.

It removes method recognition from the task.

That is fine when factorisation execution is the target.

It becomes contamination if the tutor later concludes that Mira can independently choose factorisation in mixed conditions.

The same issue appears in English.

A worksheet titled “Inference Practice” tells Jonas what kind of thinking to use.

A section titled “Variables” tells Nadia what feature of the Science investigation deserves attention.

When recognition is part of the target, remove the label in the retest.

Tutor Prompt Contamination

The tutor asks:

What evidence supports your inference?

Jonas then finds the correct line and answers well.

The performance contains evidence-selection support.

Record it.

Later use the Training Cue Hierarchy to reduce prompts and determine whether the cue has become internal.

Parent Prompt Contamination

At home, support is often invisible in the final worksheet.

The parent says:

• “Check the sign.”
• “Read the last sentence again.”
• “Remember what your tutor taught yesterday.”

The finished page looks independent.

It was not.

This does not make home help wrong.

It means the family should not use the final page alone to infer mastery.

Model-Answer Contamination

Jonas studies a model paragraph.

Then receives a nearly identical prompt and produces excellent writing.

That may reflect genuine learning.

It may also reflect memorised surface structure.

Change the topic.

Change the audience.

Preserve the writing function.

If the architecture survives, the model has transferred.

Science Contamination

Nadia sees the same investigation three times.

By the third attempt she knows which variable the teacher previously circled.

The repeated setup is useful for correction but weak evidence of independent variable reasoning.

Change the apparatus while preserving the experimental logic.

Now test whether the rule travels.

Contamination and Training Comparability

A contaminated performance should not be compared naively with a fresh independent one.

Use Training Comparability.

Supported and independent conditions are different.

Repeated and fresh items are different.

Named-method and mixed-method tasks are different.

Contamination and Training Baselines

If the baseline itself is contaminated, later improvement can be misjudged.

Suppose Mira’s starting baseline uses familiar questions she has already corrected.

Her starting performance may be inflated.

Later genuine improvement may look smaller than it really is.

Use Training Baselines with sufficiently fresh tasks.

Contamination and Measurement Noise

Contamination can produce score spikes that look like learning.

This links to Training Measurement Noise.

One unexpectedly high result should trigger inspection.

Was the task easier?

Was it familiar?

Was there more support?

If yes, the score may still be encouraging, but its meaning is narrower.

Contamination Can Also Hide Learning

This is less obvious.

If both baseline and retest use heavily coached familiar items, the learner may look stable even though independent capability has improved elsewhere.

Fresh tasks can reveal growth that contaminated measures fail to distinguish.

Use Fresh Parallel Items

The simplest contamination control is a fresh task with the same underlying training target.

Change:

• numbers;
• surface story;
• passage;
• apparatus;
• sentence content;
• option order.

Preserve the underlying rule.

This reduces answer-memory advantage while keeping the retest interpretable.

Use Delay

Immediate performance is especially vulnerable to temporary access from recent teaching.

Return later.

The correct delay depends on the learning goal.

The principle is that durable capability should not depend entirely on the correction still being warm.

Use Reduced Support

Record the smallest cue required.

Then progressively reduce it.

If success remains, the learner—not the cue—is increasingly carrying the performance.

Mira’s Contamination Check

Mira gets four quadratic questions correct during a factorisation lesson.

The worksheet title names factorisation.

The tutor wants to know whether method selection improved.

So the retest mixes factorisation, completing the square and quadratic-formula cases without labels.

Now recognition and selection are back inside the task.

Jonas’s Contamination Check

Jonas improves a composition after reading a model essay.

The tutor gives a new prompt with a different topic and audience.

Jonas must recreate the structural functions without copying the original language.

That fresh task separates deeper learning from surface imitation.

Nadia’s Contamination Check

Nadia has repeated the same plant experiment several times.

The tutor changes to an unfamiliar heat experiment preserving the same change–measure–control logic.

If Nadia identifies the structure independently, the learning has escaped the original item.

Do Not Ban Repetition

Repeated items can be useful for fluency, error correction and immediate consolidation.

Contamination is not an argument against repetition.

It is an argument against using repeated-item success as the only proof of general capability.

Do Not Ban Hints

Hints can restore productive thinking.

Use them.

But distinguish:

• performance with cue;
• performance after cue;
• performance later without cue.

These are different evidence states.

Do Not Confuse Familiarity With Fluency

The learner answers quickly because the exact question is familiar.

That is not necessarily the same as fast access to the underlying rule.

Use fresh equivalent cases to distinguish item familiarity from capability fluency.

A Parent Contamination Audit

• Has my child seen this exact question before?
• Was the answer recently corrected?
• Did someone provide a hint?
• Does the worksheet title reveal the method?
• Can the child perform on a fresh item?
• Can the child perform later without help?
• Are we calling supported performance independent mastery?

A Tutor Contamination Audit

• What information did the learner receive before this attempt?
• Has the learner seen the item or solution?
• Did my prompt remove the target decision?
• Is the task blocked in a way that names the method?
• What fresh parallel item will verify the capability?
• What later retest will reduce immediate-memory effects?
• Am I recording support honestly?

The Deeper Idea: Good Evidence Needs Independence From the Evidence-Producing Process

Teaching changes performance.

That is the point.

But the closer a measurement sits to the teaching event, the more carefully we should ask what part of the performance came from the learner and what part came from the immediate teaching environment.

Training contamination does not make supported success fake. It reminds us that proof of independent capability requires a fresh opportunity where the support, answer memory and item familiarity are no longer doing the work.

Research Foundations

A useful empirical anchor is the study The Impact of Repeated Exposure to Items, which examined score inflation associated with prior exposure to repeated certification-exam questions. The context differs from school tuition, but it provides direct evidence for the measurement problem created by repeated item exposure. This article also relies on the broader assessment literature on practice effects, test security, comparability, validity and retesting. The instructional boundary is simple: repeated items and hints remain useful learning tools, but verification should eventually move to fresh comparable tasks under appropriately independent conditions.

Continue Through How Training Works

Read this with Training Retests, Training Comparability, Training Baselines, Training Cue Hierarchy and Training Measurement Noise.