Tutor Handbook Vol 0019 | The Changed Question

The Tutor Handbook · Volume 0019 · Series ID THB-0019

The student gets the old question right.

Good.

Now change the question.

Not randomly.

Not to make it look harder.

Not to catch the learner out.

Change it because a tutor now needs to know whether the learning belongs to the student or still belongs partly to the old page.

The changed question asks whether the learner can recognise and use the same underlying capability when some of the familiar surface cues disappear.

This is Volume 0019 of The Tutor Handbook, eduKate Sengkang’s long-form practical series on the real work inside tutoring.

Volume 0018, The Reattempt, examined what a second attempt can tell us about recognition, repair and stronger learning. It ended with the next logical move:

When the old item becomes familiar, change the surface and see what travels.

The general theory of transfer already has canonical owners elsewhere in the eduKate ecosystem. The Transfer Test | What Happens When the Question Changes Shape owns the evidence concept at eduKate Sengkang. How Learning Works | Transfer owns the wider learner-runtime explanation at eduKateSG. How We Know Learning Has Really Held places transfer beside retention, practice, independence and examination performance.

This handbook does not duplicate those pages.

Its job is narrower and operational:

How does a tutor design a changed question that preserves the target relationship, changes only what is useful to change, and produces evidence that can actually be interpreted?

The tutor classification remains anchored in the Tutor Classification Model by eduKateSG, the Tutorial Class Control Tower and the Three Modes of Tuition. The Changed Question is not a new tutor class. It is a question-engineering and evidence-reading operation that every tutor class can use differently.

Quick Read

A changed question should preserve the capability being tested while deliberately changing one or more surface cues.
Changing everything at once makes failure hard to interpret. Early transfer probes should usually change one meaningful dimension at a time.
A harder question is not automatically a better transfer question. Difficulty and novelty are separate variables.
Same-item success asks whether a repair can be reconstructed. A changed question asks whether the learner can recognise the same structure when it is no longer visually announced.
The tutor should distinguish changes in numbers, wording, representation, context, order, topic label, response format and performance constraints.
If a learner succeeds only when the method is named, execution may be stronger than method selection.
If the learner fails after a harmless surface change, the original learning may have been tied to cues that the tutor did not realise were carrying performance.
Transfer evidence should be read beside support level, delay, confidence, explanation quality and the exact point where the new attempt breaks.
The changed question should be fair: do not accidentally add difficult vocabulary, unfamiliar notation, extra arithmetic or irrelevant reading load unless that is deliberately part of the test.
The long-term goal is not a student who can answer one changed question. It is a learner who can identify what remains structurally relevant when the world changes around the problem.

1. The Changed Question Is an Instrument, Not a Surprise

A weak transfer check often looks like this:

You could do that one. Let’s see if you can do this much harder one.

The new question changes the numbers, wording, diagram, topic, difficulty, response format and time pressure all at once.

The learner fails.

What did the failure mean?

Almost anything.

A good changed question behaves more like a scientific instrument. It changes enough to test the claim and preserves enough to keep the evidence interpretable.

2. Surface and Structure Must Be Separated

Every learning task contains features that matter to the underlying capability and features that merely belong to the current presentation.

Consider a ratio problem about red and blue marbles.

The colours may be surface.

The ratio relationship is structure.

Consider an English comprehension question asking why a character left the room.

The character’s name may be surface.

The requirement to infer a cause from textual evidence is structure.

Consider a Science question about evaporation from wet cloth.

The object may be surface.

The mechanism connecting moving air, particle escape and evaporation rate is structure.

The tutor cannot design a useful changed question until the tutor knows what is supposed to stay the same.

3. Ask the Invariant Question

Before changing the item, ask:

What must remain invariant for this to still be a test of the same capability?

That invariant might be:

the same mathematical relationship;
the same grammatical distinction;
the same inferential demand;
the same scientific mechanism;
the same argument structure;
the same study decision;
the same checking routine;
the same performance skill.

Once the invariant is explicit, the tutor can choose which surface feature to alter.

4. Change One Meaningful Dimension First

Early transfer probes should often change one thing at a time.

Why?

Because if the learner fails, the tutor wants to know which new demand mattered.

For example:

same relationship, different numbers;
same relationship, different nouns;
same mechanism, different diagram;
same inference, different passage;
same grammar point, different sentence;
same method, topic label removed.

Later, once the learning becomes more stable, the tutor can change several dimensions together.

5. Different Numbers Are the Smallest Useful Change in Many Mathematics Tasks

Changing numbers can remove answer memory while preserving the mathematical form.

But even number changes require thought.

If the original problem used small integers and the new problem introduces awkward fractions, the tutor has changed both the surface and the arithmetic load.

If arithmetic is not the target, keep arithmetic difficulty comparable.

The transfer question should test the intended relation, not create an accidental second test.

6. Different Nouns Can Reveal Whether the Learner Learned the Story or the Relationship

A learner solves a ratio question about sweets.

Now use buses and passengers.

If the underlying relation remains equivalent, the tutor learns whether the learner can abstract away from the familiar object story.

This is a very gentle transfer test.

It is also surprisingly revealing for students who have overlearned one textbook pattern.

7. Different Wording Can Test Meaning Rather Than Keyword Matching

Some learners recognise a method because a familiar phrase announces it.

Remove the phrase.

Keep the relationship.

In Mathematics, “find the rate of change” might become a contextual statement that requires the learner to identify a changing quantity.

In English, “Why did…” might become “What prompted…”

In Science, “state why the rate increased” might become “explain how the changed condition produced the observed difference”.

If performance collapses, the student may have learned the wording cue more strongly than the underlying demand.

8. Different Representation Can Be a Stronger Transfer Test

The same relationship can be represented as:

words;
a table;
a diagram;
a graph;
an equation;
a physical setup;
a paragraph;
a worked solution.

Changing representation can reveal whether the learner owns the relationship or only the original format.

eduKate Sengkang already has a subject-specific Science owner for this narrower job: How to Test Whether a PSLE Science Correction Transfers to a New Representation. This Tutor Handbook volume keeps the cross-subject tutor rule: change representation only when the tutor knows which underlying relation must survive.

9. Removing the Topic Label Tests Selection

A worksheet titled Simultaneous Equations gives away part of the decision.

A page titled Cause and Effect Comprehension gives away part of the decision.

A Science practice sheet grouped under one concept gives away part of the decision.

Remove the topic label and the learner must perform an additional operation:

What kind of problem is this, and what knowledge is relevant?

For PSLE Science, the detailed subject-specific route already exists in How to Practise PSLE Science Without Topic Labels So You Learn to Choose the Concept Yourself.

10. Change the Order to Test Whether Sequence Is Carrying Performance

Students can learn the order of a worksheet without learning the decision structure.

Question 1 is always expansion.

Question 2 is always factorisation.

Question 3 is always solving.

Reorder the same kinds of items.

Now the learner has to read each task rather than inherit the method from position.

11. Change the Response Format Carefully

Moving from multiple choice to open response can be an excellent transfer test.

It also changes response demand.

A learner may recognise the correct idea among options but struggle to generate it independently.

That difference is useful evidence.

But a tutor should not interpret the extra difficulty as purely conceptual. Generation, language production and organisation may now matter too.

12. Change Context Without Accidentally Changing Knowledge

A new context can reveal flexible learning.

It can also introduce new background knowledge.

If a Primary learner understands a heat concept but the changed question uses an unfamiliar industrial setting, failure may partly reflect context comprehension rather than Science.

The tutor must know whether unfamiliar context is the intended variable.

13. A Harder Question Is Not Automatically a Better Changed Question

Transfer and difficulty are related but not identical.

A question can be different but equally difficult.

A question can be harder while remaining highly familiar.

A question can be easier yet reveal transfer because the usual cue has disappeared.

Do not confuse novelty with difficulty, and do not confuse difficulty with depth.

14. Preserve Difficulty When You Want a Clean Transfer Comparison

If the original item and changed item have roughly comparable computational, reading and reasoning load, the comparison becomes easier to interpret.

Attempt 1 succeeded under the familiar surface.

Attempt 2 failed under a changed surface with similar load.

Now cue dependence becomes a plausible explanation.

15. Deliberately Increase Difficulty Only When That Is the Next Claim

Once basic transfer is stable, the tutor may deliberately increase:

number of steps;
information density;
number of competing methods;
abstraction;
time pressure;
required explanation;
distance from the original context.

Now the question is not merely “does it transfer?”

It becomes “how far does it transfer before the learner needs another support or representation?”

16. Transfer Is Not One Distance

Researchers have long warned that “near” and “far” transfer can hide many different changes. Barnett and Ceci’s classic taxonomy argued that transfer varies along several dimensions, including knowledge domain, physical context, temporal context, functional context, social context and modality.

For a tutor, the practical lesson is simple:

Say what changed.

“She transferred” is too vague.

“She used the same proportional reasoning when the numbers and objects changed, but not when the information moved from a table into prose” is much more useful.

17. Same Question Success Should Usually Come First

If the student cannot reconstruct the repaired idea on the original item, a changed question may be premature.

This is why Volume 0018 comes before Volume 0019.

The sequence is:

repair → reattempt → changed question.

If the repair itself is still unstable, changing the surface produces noisy evidence.

18. But Do Not Over-Practise the Old Item Before Changing It

The opposite danger is repetition until the old item becomes a script.

The learner can now reproduce the page beautifully.

The tutor has learned nothing about portability.

Once the learner can reconstruct the repair independently and explain the changed operation, move the question.

19. The Changed Question Tests Recognition of Structure

On the old question, the learner knows that the question was previously difficult.

That knowledge is itself a cue.

On the changed question, the learner has to detect the relevant relationship again.

This is why a learner may know how to execute a method once named and still fail transfer.

The missing capability may be recognition or selection.

20. Selection and Execution Must Be Separated

A learner sees the changed question and stalls.

The tutor names the method.

The learner completes it accurately.

This is not “doesn’t understand”.

It is more precise:

Execution is available. Independent method selection is not yet reliable under the changed surface.

That diagnosis leads to different practice.

21. The Prompt Ladder Works on Changed Questions Too

If the learner stalls, climb support gradually:

Wait.
Ask the learner to restate the question.
Ask what is known and unknown.
Ask what is similar to something seen before.
Direct attention to one structural feature.
Name the broad method family.
Offer a parallel comparison.
Reteach if necessary.

The smallest prompt that restores progress becomes part of the transfer evidence.

22. Ask What Stayed the Same

After the learner solves the changed item, ask:

What was actually the same as the old problem?

This question forces abstraction.

A good answer is not:

They both had numbers.

It is something like:

In both questions I knew the whole and one ratio relationship, so I could represent the same proportional structure even though the story changed.

Now the learner is beginning to name the invariant.

23. Ask What Changed

The second metacognitive question is:

What changed, and which changes did not matter?

This helps the learner separate signal from decoration.

That distinction is one of the deepest skills behind transfer.

24. A Learner Can Fail Because the Wrong Similarity Was Noticed

Changed questions create a new problem:

Which earlier example should this remind me of?

Students sometimes select examples by surface similarity.

Both questions mention speed, so they must use the same method.

Both passages mention disappointment, so they must ask the same inference.

Both Science questions show plants, so they must use the same concept.

A tutor should teach the learner to compare relationships, not just objects.

25. Comparing Two Examples Can Strengthen Transfer

Research on analogical encoding has shown that comparing cases can help learners abstract a shared relational structure more effectively than studying examples separately. Gentner, Loewenstein and Thompson’s work is a classic demonstration of this principle.

For tutoring, that suggests a powerful move:

Put the old and changed question side by side and ask what relationship both require.

eduKate Sengkang has a Science-specific companion for this operation in How to Compare Two Worked PSLE Science Examples to Find the Reasoning Pattern They Share.

26. Contrast Can Be More Useful Than Another Explanation

Suppose two questions look similar but require different methods.

Do not immediately explain both again.

Ask the learner to find the discriminating feature.

What is true in Question A that is not true in Question B?

Now the changed question trains classification boundaries, not just procedure repetition.

27. Similar-Looking Questions With Different Concepts Are Essential Later

Once a learner can transfer within one method family, the tutor should eventually test resistance to false similarity.

For PSLE Science, a detailed owner already exists: How to Tell When Two Similar-Looking PSLE Science Questions Need Different Concepts.

The cross-subject tutor rule is:

Transfer is not only recognising similarity. It is also refusing the wrong analogy.

28. Retrieval and Transfer Are Related but Not the Same

Retrieval asks whether knowledge can be brought back from memory.

Transfer asks whether that knowledge can be selected and used under changed conditions.

Pan and Rickard’s meta-analysis of test-enhanced learning found that retrieval practice can produce transferable learning, but the size and reliability of transfer depend on what kind of transfer is being tested and how retrieval is designed.

The tutoring implication is important:

Do not assume that remembering the old answer means the learner can recognise the idea under a new surface.

29. Delay and Change Ask Different Questions

Delay changes time.

A changed question changes surface or context.

These can be combined, but the tutor should know what each contributes.

Test	Main question
Same item, immediate	Did the repair become usable now?
Same item, delayed	Can the learner retrieve and reconstruct it later?
Changed item, immediate	Can the learner recognise the structure when the surface changes?
Changed item, delayed	Can the learner both retrieve and transfer after the original lesson has faded?

Do not collapse these into one vague claim that “the student knows it”.

30. The Changed Question Should Usually Come Before the Mixed Set

A changed question still allows the tutor to preserve one target capability.

A mixed set adds a new decision:

Which method, concept or reasoning route is relevant among several competing possibilities?

That is a later and stronger selection test.

The progression becomes:

same item → changed item → mixed item → performance condition.

31. Changed Questions Should Not Become Trick Questions

A tutor can accidentally become too clever.

The changed question is designed with obscure wording, hidden assumptions and unusual formatting.

The tutor feels that this tests “deep understanding”.

It may instead test tolerance for tutor-designed traps.

Transfer questions should be unfamiliar enough to reduce cue dependence and fair enough that failure still means something useful.

32. Fairness Requires Controlling Language Load

If a Mathematics learner understands the target relationship but the changed question contains unfamiliar vocabulary, the tutor may be measuring reading.

If reading is not the target, simplify the language or define the new word.

If reading under unfamiliar wording is deliberately part of the test, then keep it.

The difference is intentionality.

33. Fairness Requires Controlling Notation Load

Changing a diagram is one thing.

Introducing notation the learner has never seen is another.

If notation is unfamiliar, failure may not reveal whether the underlying concept transferred.

Teach or define the notation first unless notation transfer is the intended claim.

34. Fairness Requires Controlling Arithmetic Load

A changed algebra question that suddenly requires awkward arithmetic may overload a learner whose algebraic structure is actually stable.

Keep arithmetic approximately comparable when the target is algebraic transfer.

Later, restore realistic arithmetic complexity if the real performance environment demands it.

35. Fairness Requires Controlling Background Knowledge

An English argument question may be structurally similar but placed in a topic the student knows nothing about.

If background knowledge now becomes essential, the transfer distance has changed in more than one way.

That may be desirable later.

Early in diagnosis, keep the change cleaner.

36. Fairness Does Not Mean Making Every Question Easy

A fair transfer question can be difficult.

Fairness means the tutor can explain what is being demanded and why the evidence would matter.

Difficulty without interpretability is noise.

37. A Changed Question Can Reveal Prompt Dependence

On familiar work, the student succeeds whenever the tutor says:

Remember what we did last time.

The phrase itself may be carrying selection.

On a changed question, with no reminder, the learner does not recognise the relationship.

The tutor has discovered that the prompt is still external.

The canonical eduKate Sengkang owner for this broader condition is Prompt Dependence | When Correct Work Is Not Yet Independent.

38. A Changed Question Can Reveal Template Dependence

Students can learn:

the shape of a model answer;
the visual layout of a worked solution;
a paragraph template;
a memorised Science phrase;
a fixed sequence of algebraic moves.

The template may be useful scaffolding.

The changed question reveals whether the learner can reconstruct the reasoning when the template no longer fits perfectly.

39. A Changed Question Can Reveal Negative Transfer

Sometimes prior learning actively pushes the learner toward the wrong method.

The new question looks similar enough to trigger an old routine but differs in one decisive condition.

This is valuable evidence.

The tutor should not merely say, “Read carefully.”

Ask which feature triggered the old method and which feature should have blocked it.

40. Error Location Matters More Than Final Accuracy

A changed question can be wrong for a better reason.

On the old item, the learner selected the wrong concept immediately.

On the changed item, the learner selects the right concept and makes a later execution error.

The final answer is still wrong.

Transfer of selection may still have occurred.

Read the first meaningful failure.

41. Class 0 · Homework Helper: Change the Situation, Not Just the Worksheet

For a Class 0 tutor, the transferable capability may be operational.

The student learned to begin English homework using a three-step start routine.

Now test the same routine on Mathematics homework.

The changed question is not academic content.

It is whether the initiation routine travels to another subject.

42. Class 1 · Explainer: Change the Example and Ask for the Same Idea

An Explainer should not finish with:

Do you understand?

Use a changed example.

Ask the learner to explain the same relation without reusing the tutor’s exact words.

The changed question tests whether the explanation became an idea rather than a sentence.

43. Class 2 · Drill Builder: Vary Without Losing the Target

Drill can produce fragile fluency if every question looks the same.

A skilled Drill Builder varies:

numbers;
order;
representation;
surface wording;
location of the critical step;
presence or absence of obvious cues.

But variation remains controlled enough that the target skill is still visible in the data.

44. Class 3 · Diagnostic Tutor: Change One Variable to Discriminate Between Hypotheses

A Diagnostic Tutor may suspect two explanations:

the learner only recognises familiar wording;
the learner lacks the underlying concept.

Keep the concept demand constant.

Change only the wording.

If performance collapses, cue dependence becomes more plausible.

The changed question becomes a diagnostic experiment.

45. Class 4 · Route Designer: Build a Transfer Gradient

The Route Designer can arrange changes from near to progressively less familiar:

same numbers, new wording → new numbers, same structure → new representation → topic label removed → mixed context → timed performance.

This creates a transfer gradient.

The tutor can now see where the capability stops travelling smoothly.

46. Class 5 · Performance Coach: Change the Surface Inside Real Constraints

A Performance Coach eventually asks whether transfer survives:

time pressure;
mixed papers;
mark constraints;
fatigue;
uncertain topic identity;
unfamiliar phrasing.

The changed question becomes less like a laboratory probe and more like the real examination environment.

47. Class 6 · Learning Architect: Teach the Learner to Build Their Own Changed Questions

A mature learner should eventually be able to ask:

What part of this question is structural?
What is merely surface?
How could I alter the surface without changing the underlying mechanism?
What changed version would expose whether I actually understand it?
What would count as evidence that I am overfitted to the example?

Now the learner can generate part of the transfer test independently.

48. Repair Mode: Change the Downstream Problem After the Weak Link Is Rebuilt

Suppose fractions were the hidden prerequisite breaking ratio work.

After fraction repair, do not only test more fractions.

Return to ratio through a changed downstream problem.

The question is whether the repaired prerequisite now releases later learning.

49. Alignment Mode: Change the Surface Inside Current School Work

In Alignment mode, the strongest changed question may come from school.

A new teacher worksheet, quiz or class assignment naturally changes wording and context.

This prevents tutoring from proving transfer only on tutor-designed materials.

50. Frontier Mode: Change the Assumption, Not Just the Numbers

Strong learners need more than cosmetic variation.

Ask what happens if:

one assumption is removed;
the representation is incomplete;
two methods become possible;
a counterexample appears;
the learner must justify why the usual method still applies;
the familiar method no longer applies.

Frontier transfer tests the limits of the learner’s model.

51. Mathematics: Change the Story but Preserve the Relationship

Alicia solves:

Three red counters for every five blue counters.

Now the tutor changes the surface:

A recipe uses three cups of one ingredient for every five cups of another.

If Alicia recognises the same multiplicative relation, the concept is beginning to travel beyond the classroom object story.

52. Mathematics: Change Representation to Expose Structural Understanding

Denise solves an algebraic relationship from an equation.

Now show the same relationship in a graph or verbal condition.

If Denise only succeeds once the equation is written for her, symbolic execution may be stronger than representation building.

The changed question reveals the missing bridge.

53. Mathematics: Remove the Chapter Signal

A learner completes ten differentiation questions on a differentiation worksheet.

That demonstrates useful execution.

Now place one differentiation problem among algebra, functions and coordinate geometry.

Selection becomes visible.

That next step belongs increasingly to the mixed-set territory of Volume 0020, but a single unlabeled changed item can begin the transition.

54. English Comprehension: Change the Passage but Preserve the Relationship Demand

Beatrice has learned to distinguish cause from description.

Do not keep asking about the same passage.

Give a new passage with different characters and events but another causal inference.

Then ask:

What relationship is this question requesting before you search for evidence?

If Beatrice can identify the relationship under new content, the reasoning routine is becoming portable.

55. English Writing: Change the Topic but Preserve the Writing Operation

A writing learner improves paragraph causality in one composition.

Do not ask them to reproduce the same paragraph.

Give a different topic and ask for the same operation:

clear claim;
relevant evidence;
explicit connection;
controlled conclusion.

The topic changes.

The writing architecture remains.

56. Vocabulary: Change the Sentence, Preserve the Meaning Boundary

A learner memorises that reluctant means unwilling.

Recognition is useful.

Now place the word in several sentences where nearby words such as hesitant, resistant and indifferent would produce different meanings.

The changed question tests whether the learner knows the boundary strongly enough to select the word appropriately.

57. Science: Change the Object but Preserve the Mechanism

Faith explains evaporation from wet cloth.

Now use a shallow dish of water.

The object changes.

The underlying mechanism remains available.

If Faith can rebuild condition → concept → mechanism → outcome, the learning has moved beyond one memorised answer sentence.

58. Science: Change the Question While Testing the Same Mechanism

eduKate Sengkang already has a detailed subject-specific guide for this exact construction problem: How to Turn a PSLE Science Explanation Into a New Question That Tests the Same Mechanism.

The Tutor Handbook keeps the wider tutoring principle:

Change the presentation enough to require recognition again, but preserve the mechanism strongly enough that success and failure remain interpretable.

59. Science: Increase One Reasoning Demand at a Time

A Science transfer question can accidentally change six things at once.

The subject-specific owner How to Increase PSLE Science Practice Difficulty by Changing One Reasoning Demand at a Time gives the detailed PSLE route.

The general tutor rule travels across subjects:

If you want to know which new demand caused the failure, do not introduce every new demand simultaneously.

60. Primary Learners Need Visible but Gentle Variation

Younger learners may interpret a changed surface as an entirely new task.

The tutor can make transfer explicit:

The story looks different. I want you to find what part is actually the same.

This turns transfer from a hidden test into a learnable comparison habit.

61. Secondary Learners Need Increasingly Unannounced Variation

By Secondary school, the tutor should gradually reduce warnings that a transfer test is coming.

The learner must increasingly recognise the underlying structure without being told that the new question is “the same type”.

This matters because real assessments rarely announce which previous example should be copied.

62. JC and Advanced Learners Need Boundary Testing

At higher levels, transfer should include knowing when a model stops applying.

The tutor can ask:

Which assumption makes this method valid?
What changes if that assumption is removed?
Can a second method solve the same problem?
Which representation exposes the relation most clearly?
What counterexample would break an overgeneralised rule?

Transfer now includes model selection and model limits.

63. The Three-Pax Class Creates Natural Changed Questions

In a three-student group, the tutor can give three surface variants of the same underlying task.

Each learner works independently first.

Then the class compares:

what looked different;
what relationship stayed the same;
which cues were useful;
which cues were misleading;
which representation made the structure easiest to see.

This is one of the strongest uses of a small group.

The group is not three students racing toward one answer.

It becomes three views of the same learning structure.

64. Do Not Let Peer Comparison Reveal the Structure Too Early

If students discuss before independent attempt, one learner may announce the method and erase the selection evidence for the others.

A useful sequence is:

independent attempt → record method choice → compare → explain invariant → reattempt another variant.

Now collaboration supports transfer without replacing it.

65. AI Can Generate Changed Questions—but the Tutor Must Own the Invariant

AI systems can generate many variants quickly.

That is useful only if the generated question still tests the intended capability.

A generated variant may accidentally:

change the concept;
introduce ambiguity;
raise reading difficulty;
require outside knowledge;
alter the correct method;
create multiple valid answers;
remove the very relationship the tutor meant to preserve.

The tutor therefore needs to validate the invariant before using AI-generated variants as evidence.

66. The Learner Can Use AI to Create Their Own Transfer Test

For an older learner, a useful task is:

Ask for three new questions that preserve this underlying relationship but change the surface in three different ways. Then explain which feature remains invariant before solving them.

The important learning is not the volume of generated questions.

It is whether the learner can judge whether those questions really belong to the same conceptual family.

67. Parents Should Not Rescue the Changed Question Too Quickly

At home, a changed question may look like regression.

“You knew this yesterday. Why can’t you do it now?”

The more useful response is to preserve the evidence.

Let the learner attempt.

Record where they became stuck.

If support is given, make it visible.

The changed question is doing its job by revealing how far the learning currently travels.

68. Alicia: The Story Changes but the Ratio Survives

Alicia originally learned ratio using coloured counters.

She succeeds on the same item again.

The tutor changes the objects to ingredients in a recipe while keeping the ratio and arithmetic load comparable.

Alicia pauses, identifies the same multiplicative relation and builds the representation independently.

The story changed.

The structure travelled.

69. Beatrice: The Passage Changes and the Relationship Still Has to Be Named

Beatrice has learned that a “why” question requires a causal answer rather than a description.

On a new passage, the wording changes to:

What prompted the character to change her decision?

Beatrice initially searches for a repeated keyword.

The tutor asks only:

What relationship is the question asking for?

Beatrice says, “Cause.”

She then answers correctly.

The evidence is precise: the relationship is nearly available, but the classification trigger is not yet fully internal.

70. Ciara: New Numbers Reveal That the Tutor’s Layout Was Carrying Her

Ciara can solve algebra when the tutor writes each line in a familiar vertical arrangement.

The changed question uses new numbers and asks her to organise the working herself.

She knows the operations but loses signs because the external layout has disappeared.

The concept transferred further than the self-organisation routine.

The next repair is not “teach algebra again”.

71. Denise: The Topic Label Disappears

Denise is strong at differentiation practice.

The tutor gives a changed problem with no heading.

Denise solves it with algebra because that is the method family currently active in her mind.

Once the tutor says “rate of change”, Denise switches immediately and completes the calculus accurately.

Execution is not the weak link.

Recognition is.

72. Emily: The Surface Change Is Too Small to Be Informative

Emily solves the changed question instantly.

The tutor does not automatically celebrate mastery.

The numbers changed, but the layout, wording and method cue remained almost identical.

The test was too near to add much evidence for Emily.

The tutor changes representation next.

For one learner, a tiny change can be diagnostic.

For another, it is already beneath the frontier.

73. Faith: The Object Changes but the Mechanism Does Not

Faith learned a Science explanation using wet clothes.

The changed question uses water in a shallow tray.

Faith begins to search memory for the old sentence.

The tutor asks:

Forget the nouns for a moment. What condition changed, and what mechanism links that condition to the outcome?

Faith reconstructs the mechanism.

The tutor has not given the answer.

The tutor has redirected attention from surface to structure.

74. Failure Mode: Changing Everything at Once

New topic.

New notation.

New language.

New response format.

More difficult arithmetic.

Shorter time.

The learner fails.

The tutor learns almost nothing.

75. Failure Mode: Cosmetic Change

The tutor changes 12 to 13 and declares transfer.

For a fragile learner, that may be a useful first variation.

For a stable learner, it may add almost no information.

The change has to be meaningful relative to the learner’s current state.

76. Failure Mode: Hidden Method Cue

The new question looks different, but the tutor introduces it by saying:

Here’s another chain-rule question.

The surface changed.

The selection problem disappeared.

If selection is the target, do not announce the method.

77. Failure Mode: Unfair Language

The tutor tries to test Science transfer and accidentally introduces unfamiliar vocabulary that becomes the real barrier.

The result is not useless.

It is simply evidence about a different problem.

Label it correctly.

78. Failure Mode: Treating Failure as Proof of No Learning

A learner fails one changed question.

The tutor concludes that the original teaching “did not work”.

That conclusion is too broad.

The learner may have:

retained the concept but failed to recognise it;
recognised it but failed in execution;
been disrupted by new language;
needed a smaller transfer step;
encountered a genuinely new demand.

Changed-question failure is evidence.

It is not a verdict.

79. Failure Mode: Treating Success as Proof of Unlimited Transfer

The learner succeeds on one new question.

Good.

The evidence has strengthened.

It does not mean the capability will automatically survive every context, representation, subject, delay and performance condition.

Keep the claim proportional to the test.

80. Failure Mode: Creating Changed Questions With No Stop Rule

The tutor keeps generating variants indefinitely.

The student becomes exhausted.

The evidence stopped improving twenty questions ago.

Know what success would justify next.

For example:

Two independently solved changed questions across different surfaces are enough to move this skill into a mixed set.

The exact rule depends on the capability.

The principle is that evidence should change the route.

81. A Changed-Question Design Board

Field	Tutor question
Target capability	What exactly should transfer?
Invariant	What must remain structurally the same?
Surface change	What will I deliberately alter?
Difficulty control	Am I also changing load unintentionally?
Support condition	Independent, prompted, scaffolded?
Delay	Immediate or after time has passed?
Success claim	What would success actually justify?
Failure claim	What would failure make more plausible?
Next move	Another change, mixed set, delay, reteach or retire?

82. A Five-Minute Changed-Question Routine

Name the invariant privately. The tutor decides what must remain the same.
Change one useful surface feature. Numbers, nouns, wording, representation or cue.
Let the learner attempt before announcing the method.
Ask what stayed the same and what changed.
Decide the next evidence level. Another changed item, delayed return, mixed set or performance condition.

83. A One-Minute Changed-Question Routine

This looks different. What is actually the same? Try it without me naming the method.

Even this short routine teaches the learner to search for structure rather than visual repetition.

84. Twelve Questions Every Tutor Should Ask Before Using a Changed Question

What capability am I trying to test?
What feature is structurally invariant?
What surface feature am I changing?
Am I changing difficulty at the same time?
Am I introducing new vocabulary or notation?
Does the learner need to recognise the method independently?
What support will remain available?
How much time has passed since teaching?
What would success actually prove?
What would failure make more plausible?
What is the smallest next prompt if the learner stalls?
What evidence level comes after this one?

85. Twelve Questions Learners Can Ask Themselves

What looks different?
What might still be the same underneath?
What is the question really asking me to do?
Which earlier idea could be relevant?
Am I choosing by surface similarity or by relationship?
What condition makes this method appropriate?
What would make this method inappropriate?
Can I represent the problem another way?
Do I need the old example visible?
Can I explain why the same idea applies?
Where did I first become uncertain?
What did this changed question reveal about what I actually know?

86. Twelve Questions Parents Can Ask Without Turning Home Into a Test Centre

Was the question genuinely new or only cosmetically changed?
Did my child attempt before being told the method?
What stayed structurally the same?
What surface feature changed?
Did new vocabulary create unnecessary difficulty?
Did the tutor preserve comparable difficulty?
Did my child need the same old prompt?
Could my child explain why the method still applied?
Did failure occur at recognition, selection or execution?
Did one successful changed question lead to a proportionate claim?
Is the tutor gradually reducing cues?
Is the learner becoming better at finding structure independently?

87. What the Research Supports

Transfer has been studied for more than a century, and the research does not support a simplistic claim that learning automatically generalises whenever teaching is good.

Barnett and Ceci’s taxonomy is useful because it reminds educators to specify what kind of context change is involved rather than treating all “far transfer” as one phenomenon.

Pan and Rickard’s 2018 meta-analysis found that retrieval practice can produce transferable learning, but transfer varied substantially across types of task and depended on factors such as elaboration and the nature of the test.

Gentner, Loewenstein and Thompson’s work on analogical encoding showed that comparing cases can support abstraction of shared relational structure and later transfer.

The Education Endowment Foundation’s August 2026 material on metacognition and self-regulation emphasises helping learners plan, monitor and evaluate strategies so these processes can become increasingly internal over time.

These evidence streams support the operating principles used here:

transfer needs to be tested rather than assumed;
the dimensions of change should be specified;
comparison can help learners abstract shared structure;
retrieval can support transfer but does not guarantee every kind of transfer;
metacognitive reflection can help learners notice which strategies and structures travel.

88. What the Research Does Not Give the Tutor

Research does not provide one universal transfer ladder for every subject, age and learner.

It does not tell the tutor that changing three nouns is always enough.

It does not tell the tutor how many changed questions every child must solve.

It does not justify calling one successful unfamiliar item proof of broad mastery.

The tutor still needs subject knowledge, curriculum knowledge, diagnostic judgement and proportional claims.

89. Evidence and Further Reading

90. The Changed-Question Operating Cycle

Confirm the repair on the old item → name the target capability → identify the invariant → choose one meaningful surface change → keep irrelevant load controlled → remove unnecessary method cues → let the learner attempt → record the smallest support required → compare old and new reasoning → ask what stayed the same and what changed → decide whether the evidence supports transfer, cue dependence, selection weakness or another explanation → move to another change, delay, mixed selection or performance conditions.

This is how a changed question becomes more than variety.

It becomes evidence.

91. Final Compression

The old question is familiar.

The learner succeeds.

Now change something.

Change the numbers.

Change the nouns.

Change the wording.

Change the representation.

Remove the topic label.

Change the context.

But know what you are preserving.

Know what you are changing.

Know what success would mean.

Know what failure would mean.

Do not call every harder question transfer.

Do not call every unfamiliar question fair.

Do not call one successful variant unlimited mastery.

And do not rescue so quickly that the method-selection evidence disappears.

The changed question has one disciplined job:

Move the surface far enough that the learner has to recognise the underlying structure again, while preserving enough of the structure that the tutor can understand what the new performance means.

When the learner can do that, learning has begun to move.

When the learner can explain what stayed the same, learning has begun to become abstract.

When the learner can choose the capability among competing alternatives without a topic label, learning is ready for a stronger test.

That is the discipline of the changed question.

That is Tutor Handbook Volume 0019.

Next in the series: The Tutor Handbook Vol No.0020 | The Mixed Set — How a Tutor Tests Method Selection When No Topic Label Announces the Answer.

Small Group Tutorials

The Tutor Handbook Vol No.0019 | The Changed Question — How a Tutor Tests Whether Learning Travels When the Surface Changes