Maya’s Mathematics mark rises from 61 to 78.

Everyone is pleased.

Her mother says:

“The tuition is working.”

Perhaps it is.

But between the two tests, several other things happened.

• school finished the topic and revised it;
• Maya completed more homework than usual;
• the second paper contained fewer long algebraic questions;
• she slept better the night before;
• the tutor repaired one recurring sign-control weakness;
• Maya had also simply spent another six weeks becoming more familiar with the syllabus.

The mark rose.

That is an observation.

“Tuition caused the rise” is a causal claim.

Those are not the same sentence.

To make a causal claim, we would ideally know what Maya’s result would have been over the same period without the tuition intervention while everything else remained appropriately comparable.

That alternate outcome cannot be observed directly for the same Maya at the same moment.

It is the counterfactual.

What would probably have happened if this intervention had not occurred?

Large research studies use designs such as randomised controlled trials to estimate that missing comparison by creating treatment and comparison groups. Individual tuition cannot normally reproduce that standard for one child.

But it can still think more carefully.

It can ask whether the timing fits.

Whether the repaired mechanism is the same mechanism that improved.

Whether the improvement survives changed questions.

Whether another plausible explanation fits better.

Whether the improvement repeats.

Whether the gain remains after tutor support is reduced.

This is the Counterfactual Check.

Quick Read: The Counterfactual Check in One Sentence

When performance changes after tuition, ask what else changed, what would plausibly have happened without the intervention, and whether the specific mechanism tuition targeted is the mechanism that improved.

This is not a demand that parents or tutors become causal-inference researchers.

It is a demand for intellectual modesty.

Improvement deserves celebration.

Credit deserves evidence.

1. Before-and-After Is Not Automatically Cause-and-Effect

A student starts tuition in January.

Marks rise by March.

The sequence is compatible with tuition helping.

It does not isolate tuition as the only cause.

During those same months:

• school teaching continued;
• the student practised more;
• the syllabus became more familiar;
• parents may have changed routines;
• the next assessment may have been easier or better matched to the learner;
• ordinary maturation and adaptation occurred.

“After” is necessary for a causal story.

It is not sufficient.

2. Why Researchers Need a Counterfactual

Educational impact research faces a basic impossibility.

For one learner at one moment, we cannot observe both:

• the outcome after receiving the intervention; and
• the outcome the same learner would have had at that same time without receiving it.

Randomised controlled trials address this by assigning sufficiently large groups to intervention and control conditions. If randomisation works, the comparison group helps estimate what would have happened without the intervention.

EEF explains the logic directly in its current evaluation pages: random assignment creates a comparison that allows differences in outcomes to be attributed more confidently to the programme rather than to pre-existing group differences.

A tuition centre working with Maya cannot clone Maya and send one version to tuition and one version home.

So individual attribution must remain weaker than a well-run trial.

That does not make evaluation useless.

It changes the language we should use.

3. Prefer “Contributed To” When That Is What the Evidence Supports

There is a useful difference between:

“Tuition caused Maya’s seventeen-mark improvement.”

and:

“The timing and mechanism evidence suggest the targeted tuition repair contributed to Maya’s stronger algebra performance.”

The second claim can be both meaningful and honest.

Education is usually multi-causal.

School teaches.

The student practises.

Parents stabilise routines.

Tutors diagnose and repair.

Sleep, motivation, confidence and examination conditions matter.

Good tuition does not need to claim every gain as its own in order to be valuable.

4. The Counterfactual Check Is Different From the Evidence Threshold

The Evidence Threshold asks whether enough evidence exists to change the learning plan.

The Counterfactual Check asks something after change is observed:

How much of this improvement can we reasonably attribute to the intervention rather than to other changes occurring at the same time?

Evidence Threshold:

“Do we know enough to act?”

Counterfactual Check:

“After acting, what probably produced the change?”

5. The Counterfactual Check Is Different From the Flight Recorder

The Flight Recorder preserves history.

The Counterfactual Check interrogates that history.

Before intervention:

sign error appears under negative-bracket rearrangement.

Intervention:

explicit sign-state check trained.

After intervention:

sign error disappears across fresh mixed questions and later school work.

That sequence is stronger attribution evidence than a total score rising while the targeted mechanism remains unexamined.

6. The Counterfactual Check Is Different From the Transfer Gate

The Transfer Gate asks whether a capability survives changed conditions.

Passing transfer strengthens the causal story because it makes an explanation such as “the student merely copied the tuition worksheet pattern” less plausible.

But transfer does not prove attribution alone.

School may have taught the same strategy.

The learner may have practised independently.

Transfer tells us the capability travels.

Counterfactual reasoning asks what most plausibly helped create it.

7. The Counterfactual Check Is Different From the Diagnostic Probe

The Diagnostic Probe separates competing explanations of current performance.

The Counterfactual Check separates competing explanations of change.

Why did Maya fail?

Diagnostic question.

Why did Maya improve?

Attribution question.

The logic is similar:

do not settle on the first plausible story while alternatives still fit.

8. Start With a Timeline

Attribution becomes clearer when events are placed in order.

Week 1: Maya repeatedly loses signs during rearrangement.

Week 2: tutor isolates the sign transition and trains one verification cue.

Week 3: fresh tuition questions show fewer sign losses.

Week 4: mixed questions remain stable without prompting.

Week 6: school paper shows no recurrence in comparable algebra.

Now compare with another story.

If the improvement began three weeks before the tuition intervention, the causal story weakens.

If improvement appears immediately after a mechanism-specific intervention and repeats in the exact kind of task that previously failed, the story strengthens.

Time order does not prove causality.

But causes should precede their effects.

9. Ask What Else Changed

Before crediting the intervention, list concurrent changes.

• school completed or revised the topic;
• home practice increased;
• a new teacher or school resource appeared;
• the student began sleeping earlier;
• the next test was easier or differently weighted;
• the student moved from an unfamiliar topic into a stronger one;
• the examination format changed;
• motivation rose for reasons unrelated to tuition;
• another tuition or enrichment source was added.

This is not an attempt to deny tuition credit.

It is how credit becomes believable.

10. Ask Whether the Improvement Matches the Targeted Mechanism

This is one of the strongest practical checks available to an individual tutor.

The intervention targets:

pronoun reference.

The improvement appears:

in vocabulary spelling.

Those events may both be good.

The causal connection is weak.

The intervention targets:

pronoun reference.

The improvement appears:

in pronoun-reference questions across new passages and school work.

That causal story is more coherent.

Mechanism match matters.

11. Ask Whether the Dose Matches the Claim

A student attends one ninety-minute lesson.

Two months later, overall English rises by fifteen marks.

Could that lesson have helped?

Possibly.

Should the centre claim the whole fifteen marks?

No reasonable evidence supports that scale of ownership.

Causal claims should be proportionate to intervention exposure and mechanism.

A narrow intervention may have a narrow but valuable effect.

That is enough.

12. Ask Whether the Comparison Changed

Marks are not always comparable across papers.

61 on one assessment and 78 on another may reflect:

• different difficulty;
• different topic weighting;
• different question formats;
• different marking strictness;
• different preparation time;
• different performance conditions.

The mark difference matters.

But mechanism-level comparison can be more informative.

Instead of:

61 → 78.

also ask:

sign-state errors: 5 → 1.

unattempted questions: 4 → 1.

unsupported inference errors: 6 → 2.

These measures still need interpretation, but they align more closely with what tuition actually trained.

13. Beware Regression Toward the Middle

Students often seek extra help after an unusually bad result.

That creates a classic attribution trap.

An unusually bad performance can be partly produced by temporary factors:

• illness;
• sleep;
• an unusually difficult paper;
• panic;
• random mistakes;
• one poor topic match.

The next result may improve even if nothing important changes, simply because the extreme conditions do not repeat.

This phenomenon is often described as regression toward the mean.

The practical lesson is simple:

if tuition begins immediately after an unusually poor result, do not automatically claim the next less-poor result as proof of treatment effect.

Look for mechanism change and repeated evidence.

14. Beware Maturation

Children change while we teach them.

They become older.

More experienced.

More familiar with school.

More familiar with examination formats.

A Primary 3 child may become more fluent over six months partly because six months of reading, school and ordinary development happened.

Tuition may have accelerated that growth.

But “the child improved while enrolled” should not be confused with “the child would otherwise have stayed unchanged.”

The counterfactual is almost never zero growth.

15. Beware School Teaching as a Concurrent Intervention

This sounds obvious, but it is frequently forgotten.

The school teacher is teaching too.

If Maya improves in simultaneous equations after both school and tuition teach simultaneous equations, attribution belongs carefully to the combined learning environment unless stronger evidence separates the contributions.

This is not a problem.

Education is not a competition to own the child’s mark.

The goal is that Maya learns.

16. Beware Increased Practice

Tuition may cause a student to practise more.

Or parents may separately impose more practice.

Or the examination season may increase practice naturally.

If performance improves, practice volume can be part of the causal chain.

The useful question becomes:

Did tuition improve performance through better instruction, through increased practice, through better practice selection—or through some combination?

That is a more informative question than simply claiming the final mark.

17. Beware Motivation Effects

A student likes the new tutor.

Starts homework earlier.

Practises voluntarily.

Marks rise.

Was the improvement caused by explanation quality or by engagement?

Possibly both.

Motivation is not a fake pathway.

If the educational relationship makes useful practice more likely, that can be a real mechanism of benefit.

The Counterfactual Check simply asks us to name the pathway accurately.

18. Look for Mechanism-Specific Leading Indicators

Total marks are late and noisy.

Mechanism indicators can appear earlier.

Tuition trains:

identify the reference whole before percentage calculation.

Leading indicator:

Maya correctly marks the reference quantity on fresh questions before calculating.

Tuition trains:

support inference with passage evidence.

Leading indicator:

Hana rejects plausible unsupported answers more often.

Tuition trains:

condition → process → outcome Science explanation.

Leading indicator:

Jia Jun writes the causal middle step without a template.

These changes fit the intervention mechanism more closely than a broad result alone.

19. Look for a Dose–Response Pattern Carefully

If a capability improves as the learner receives appropriately targeted exposure, and stabilises after repeated practice, that pattern can strengthen the contribution story.

But do not treat “more tuition, higher marks” as automatic proof.

Students receiving more tuition may have greater need.

Families buying more tuition may also provide more home support.

More hours can create fatigue.

Dose needs interpretation.

The useful individual question is narrower:

Did the targeted mechanism improve as the targeted work was introduced and practised?

20. Look for Replication

One improvement can be chance or context.

A repeated pattern is more persuasive.

Maya’s sign control improves.

Then remains strong on a different algebra topic.

Then survives delay.

Then appears at school.

Replication does not turn a single-student observation into a randomised trial.

It does make the causal story less fragile.

21. Look for Selective Improvement

Suppose tuition focuses on inference but not vocabulary.

Afterward:

• inference improves;
• vocabulary remains approximately unchanged.

That selectivity fits the intervention story better than a uniform improvement across unrelated skills.

If everything improves at once, tuition may still have helped.

But broader explanations—better sleep, maturation, easier paper, stronger motivation—become more plausible too.

Specific interventions should often leave specific fingerprints.

22. Look for Failure Where the Intervention Should Not Help

Negative evidence can strengthen attribution.

Maya’s sign-state control improves.

Her geometry theorem recall does not.

That is not disappointing if geometry recall was never targeted.

It may actually make the mechanism story more coherent.

A treatment that supposedly improves everything is harder to interpret than one whose effects appear where its mechanism predicts.

23. Use Withdrawal Carefully as an Independence Check, Not as Proof

Sometimes support can be reduced after a skill stabilises.

If performance remains strong, that suggests the learner now carries more of the capability internally.

If performance collapses immediately, tutor support may still be part of the active system.

This is useful for the Exit Ramp.

But withdrawal should not be used recklessly to manufacture causal evidence.

Do not deliberately destabilise a child near a high-stakes examination merely to see whether marks fall.

Educational decisions come before experimental purity.

24. Use Natural Contrasts

Sometimes ordinary learning creates useful contrasts without withholding support.

The tutor targets one skill but not another similar one.

One topic receives explicit repair before a school test; another stable topic receives normal practice.

A new checking routine is introduced for Mathematics while English routines remain unchanged.

These contrasts cannot establish laboratory-level causality.

They can help answer practical questions about whether change appears where the intervention predicts.

25. Do Not Change Five Things at Once If You Want to Learn What Worked

A student receives:

• new tutor;
• new textbook;
• new study timetable;
• two extra weekly lessons;
• new parent reward system.

Marks rise.

Wonderful.

Which element helped most?

We do not know.

If urgent circumstances demand several simultaneous changes, make them.

But accept that attribution becomes harder.

If the situation is stable enough, introduce important changes sequentially so the learning system itself becomes more interpretable.

26. The Counterfactual Check Should Not Block Action

There is an obvious danger here.

Once tutors start questioning causality, they can become afraid to claim anything or change anything.

That is not the goal.

The Evidence Threshold already gives the answer.

Use enough evidence for the decision in front of you.

Then act.

Counterfactual thinking improves interpretation after the action.

It should not become an excuse for paralysis.

27. Primary 1: Be Especially Careful With Natural Growth

Young children can change rapidly.

Reading fluency can improve because school instruction, home reading, maturation and tuition all interact.

If Maya reads more smoothly after three months, celebrate.

Then look for what the tuition specifically targeted.

If tuition worked on full-pattern decoding rather than first-letter guessing, does that behaviour change on unfamiliar words?

Mechanism evidence makes the contribution claim more credible.

28. Primary 2: Separate Routine Effects From Subject Effects

Ethan’s marks rise after the family introduces an earlier homework time and tuition starts targeted reading work.

Perhaps both helped.

If improvement appears across several subjects, the routine change may be important.

If the largest change appears in the specific reading behaviour tuition trained, tuition’s contribution becomes more plausible.

Family and tuition do not need to fight for ownership.

The child benefited from a system.

29. Primary 3 and Primary 4: Science Makes Mechanism Matching Visible

Jia Jun’s Science answers improve after tuition trains causal explanation.

Check the middle of the answer.

Is the causal process now being written?

Or did marks rise because the new paper contained more factual questions?

Subject totals can hide mechanism changes.

Inspect the actual task family tuition claimed to repair.

30. Primary 5: Use the Long Runway to Replicate

Primary 5 offers time.

If an intervention appears to work, do not immediately broaden the claim.

Retest after delay.

Change the surface.

Wait for school evidence.

If the change repeats, the contribution claim becomes stronger before Primary 6 pressure arrives.

31. Primary 6: Causal Purity Is Less Important Than Useful Action

Near PSLE, the priority is the child’s performance and wellbeing.

If two reasonable interventions are helping together, do not dismantle one merely to discover which deserves more credit.

Keep the system that is working.

Use the Counterfactual Check mainly to avoid exaggerated claims and to understand which mechanisms should be retained.

Education comes before attribution.

32. Secondary 1: Transition Is a Confounder

Secondary 1 students often improve quickly after the first difficult months because they adapt to the new environment.

A tuition programme introduced during that transition may genuinely help.

But ordinary adaptation is another plausible explanation for some improvement.

Again, inspect targeted mechanisms.

If tuition trained algebraic notation and the notation errors disappear, that contribution is more defensible than claiming the entire Secondary 1 adjustment.

33. Secondary 2: Workload Changes Can Mimic Subject Improvement

A student reorganises the week.

Sleep improves.

Homework begins earlier.

Marks rise in three subjects.

A Mathematics tutor should be cautious about claiming a system-wide effect from Mathematics tuition alone.

The broader routine change may explain part of the pattern.

This is where eduKatePunggol’s Family Life Education lens matters.

Sometimes the best educational intervention is not more subject teaching.

34. Secondary 3: New A-Math Creates Multiple Causal Paths

Maya improves in Additional Mathematics after tutoring.

Was it because the tutor taught the new concept better?

Because algebraic foundations were repaired?

Because school practice increased?

Because the second test covered a stronger topic?

Use the mechanism record.

If the intervention targeted algebra inside calculus and those algebraic breakdowns disappear across several A-Math topics, that is useful contribution evidence.

35. Secondary 4: Prelim to O-Level Improvement Needs Careful Attribution

Preliminary examinations are often deliberately demanding.

A later result can improve because:

• the student genuinely improved;
• the paper changed in difficulty;
• revision intensified everywhere;
• school and tuition both targeted the same gaps;
• the learner became more familiar with examination conditions.

Do not claim the difference between prelim and final result as a clean tuition effect.

Instead ask which repaired mechanisms survived the final environment.

That is a more meaningful educational story.

36. JC: Strong Students Can Improve for Subtle Reasons

Ethan moves from 74 to 82.

The difference may come from only a few high-level decisions.

Better question selection.

Cleaner representation.

Less over-checking.

Faster recovery from one difficult item.

If tuition targeted one of these and the exact behaviour changed, mechanism attribution can be stronger even when total marks move only modestly.

Not every valuable intervention creates a dramatic score jump.

37. English: Did Vocabulary Improve—or Did the Passage Get Easier?

Hana scores much better in comprehension.

Before celebrating a vocabulary programme, inspect the new passage.

Were the unfamiliar words actually fewer?

Did Hana now infer meanings from morphology or context more accurately?

Did previously trained words appear?

Did vocabulary improve in writing too?

Different evidence supports different claims.

38. English: Did Writing Improve—or Did the Prompt Fit Better?

One composition score rises sharply.

Perhaps the writing intervention worked.

Perhaps the student loved the prompt.

Check the targeted mechanism.

Did Hana independently plan a viable turning point?

Did paragraphs now explain rather than merely list events?

Did the same improvement appear under another prompt?

Replication separates prompt luck from portable improvement.

39. Mathematics: Did Calculation Improve—or Did the Numbers Get Friendlier?

Maya becomes faster.

Compare like with like.

Same method family.

Comparable arithmetic complexity.

Comparable need for method selection.

If speed improves there too, the fluency story strengthens.

If only the easy-number set becomes fast, do not overclaim automaticity.

40. Science: Did Understanding Improve—or Did the Question Become More Familiar?

Jia Jun answers evaporation questions perfectly after practising evaporation questions.

Good.

Now use the Transfer Gate.

Fresh apparatus.

Different wording.

No topic label.

If the particle model still generates the right reasoning, understanding is a stronger explanation than memorisation of question form.

41. Additional Mathematics: Did the Concept Improve—or Did Algebra Repair Unlock It?

Maya’s differentiation marks rise.

The tutor had spent most of the month repairing algebraic simplification.

That can still be a tuition effect.

But the mechanism is not:

“We taught more differentiation.”

It may be:

“By stabilising algebraic manipulation, the learner could finally execute differentiation methods she already understood.”

Causal attribution improves when the mechanism is named correctly.

42. A Four-Question Counterfactual Check

For everyday tuition, start with four questions.

1. What exactly improved?
Prefer mechanism-level change over total-score storytelling.

2. What changed immediately before the improvement?
Map tuition, school, home practice, routine and assessment conditions.

3. What else could reasonably explain the improvement?
List at least one serious alternative.

4. What evidence would make the tuition-contribution story stronger or weaker?
Replication, transfer, school return, selective change, reduced prompting.

This is not formal causal estimation.

It is disciplined practical reasoning.

43. A Strong Contribution Story

Before:

Maya repeatedly chooses the new amount as the percentage-change base under dense wording.

Probe:

concept present; reference tracking fails under wording load.

Intervention:

train “measured from what?” trigger before calculation.

Immediate result:

fresh mixed items improve.

Delayed result:

the trigger remains after one week without cue.

School return:

same error no longer appears in comparable school work.

Concurrent changes:

school also revised percentages.

Conclusion:

“The evidence supports that targeted tuition likely contributed to improved reference-quantity control, alongside normal school revision.”

That is a strong, useful and proportionate claim.

44. A Weak Contribution Story

Before:

one unusually poor test.

Intervention:

four weeks of generic tuition across many topics.

After:

a different, easier paper produces a higher total mark.

No mechanism record.

No comparable tasks.

No account of school revision.

Claim:

“Tuition improved the student by twenty marks.”

The improvement may be genuine.

The attribution is weak.

45. Counterfactual Thinking Protects Parents From Marketing Noise

Parents often encounter claims such as:

“Improved from C to A after joining us.”

Perhaps true.

The useful question is:

What evidence shows that the programme caused that change rather than merely preceding it?

For one testimonial, that may be impossible to establish strongly.

Parents should therefore treat individual before-and-after stories as illustrations, not controlled evidence.

Programme-level claims deserve programme-level evaluation.

46. Counterfactual Thinking Protects Tutors From Ego

A tutor wants the child to improve.

When improvement arrives, it is emotionally satisfying to take ownership.

But educational integrity asks a better question.

What did the learner do?

What did the school do?

What did the family change?

What did tuition contribute?

The mark belongs first to the child’s performance.

The adults can share responsibility for building the conditions.

47. Counterfactual Thinking Protects Students From Dependency Narratives

“You improved because of tuition.”

A child may hear:

“Without tuition, I cannot do this.”

That may be the opposite of what good tuition wants to teach.

A more useful explanation is:

“Tuition helped us find and repair a specific problem. You practised it, carried it into school, and now you can do more of it yourself.”

Support contributed.

The capability now belongs increasingly to the learner.

48. Counterfactual Thinking Protects Against Over-Tuition

If every improvement is attributed to continued tuition, tuition becomes impossible to stop.

Marks good?

Tuition must be working. Continue.

Marks bad?

Student still needs tuition. Continue.

That logic has no exit condition.

The Counterfactual Check breaks the loop.

Which capability is tuition still uniquely improving?

Which improvement now appears to be maintained by the student, school and ordinary practice without premium tutor intervention?

This connects directly to The Exit Ramp.

49. Counterfactual Thinking Protects Against Under-Crediting Useful Tuition

Causal humility does not mean pretending tuition never matters.

Sometimes the contribution story is strong.

A specific recurring error is identified.

A targeted repair is introduced.

The exact error drops immediately.

Related untargeted skills remain unchanged.

The improvement survives fresh contexts and delay.

The same pattern appears at school.

That is meaningful evidence that the tuition intervention contributed.

Say so.

Accuracy is not modesty theatre.

It is accurate credit.

50. The Parent’s Counterfactual Check

• What exactly improved?
• What did tuition specifically work on?
• Did that exact behaviour improve?
• What else changed at school or home?
• Was the next assessment genuinely comparable?
• Did the improvement repeat?
• Did it survive without tutor prompts?
• Did it appear in school work?
• Could the child now maintain it with less tuition?

Parents do not need a laboratory.

They need enough clarity to distinguish useful evidence from attractive storytelling.

51. The Tutor’s Counterfactual Check

• What was the baseline pattern before intervention?
• Did the proposed cause precede the change?
• Which mechanism did the intervention target?
• Did that mechanism change specifically?
• What serious alternative explanations exist?
• Did school teach the same thing during the same period?
• Did practice volume or family routines change?
• Is the later assessment comparable?
• Did improvement replicate across fresh tasks?
• Did it survive delay and transfer?
• Are there untargeted skills that did not change?
• Am I claiming more than the evidence permits?

This is enough to make routine attribution more disciplined.

52. The Student’s Counterfactual Check

An older student can ask:

• What did I change?
• What changed in school?
• What practice did I add?
• Which specific skill improved?
• Can I still do it without the tutor?
• What would probably have happened if I had kept studying exactly as before?

The final question cannot be answered perfectly.

But asking it changes the learner’s thinking.

Success becomes something to analyse rather than merely celebrate.

53. Back to Maya’s 78

The tutor places the two Mathematics papers side by side.

He does not start with the seventeen marks.

He starts with the mechanism they repaired.

On the first paper, five sign-state errors appear under negative brackets and rearrangement.

On the second, one appears.

In tuition, fresh unlabelled algebra has shown the same improvement for three weeks.

Maya now performs the sign check without prompting.

School revised algebra during the same period.

So the tutor does not say:

“We gave you seventeen marks.”

He says:

“The sign-control repair is showing up both here and at school. School revision probably helped too. The important thing is that you now seem to own the check yourself.”

Maya nods.

The mark matters.

But the better story is not that tuition produced a number.

The better story is that a specific learning mechanism changed, travelled, and became increasingly hers.

That is the kind of causal claim education can use responsibly.

---

The Counterfactual Check in One Page

1. Improvement is an observation; attribution is a causal claim.
Do not confuse sequence with proof.

2. Ask the counterfactual.
What would probably have happened without this intervention?

3. Accept that one learner cannot supply a perfect control condition.
Individual tuition attribution is necessarily weaker than a well-run impact trial.

4. Build a timeline.
The proposed cause should precede the effect.

5. List concurrent changes.
School, practice, routines, maturation and paper difficulty may all matter.

6. Match mechanism to outcome.
A targeted intervention should leave its strongest fingerprint on the process it targeted.

7. Compare like with like.
Total marks from different papers may not be directly comparable.

8. Beware unusually low baselines.
The next result can improve partly because temporary bad conditions do not repeat.

9. Look for replication.
Fresh tasks, delay and school return strengthen the contribution story.

10. Look for selective improvement.
Specific change where the intervention predicts is useful evidence.

11. Use natural contrasts where ethical and practical.
Do not withhold useful support merely to create experimental purity.

12. Prefer proportionate language.
“Likely contributed to” may be more accurate than “caused.”

13. Do not under-credit strong mechanism evidence.
Causal humility is not a refusal to recognise useful teaching.

14. Give the learner ownership.
The capability should increasingly belong to the student, not to the story of the tutor.

15. Use attribution to improve the next decision.
Keep what plausibly worked; reconsider what did not.

Where This Fits in eduKatePunggol

• How Tuition Works at eduKatePunggol — the broad tuition relationship.
• How Tuition Works | The Weak Link — identify the first limiting mechanism.
• How Tuition Works | The Learning Simulator — create controlled rehearsal conditions.
• How Tuition Works | The Sparring Partner — apply intelligent resistance.
• How Tuition Works | The Flight Recorder — preserve the before-and-after history needed for attribution.
• How Tuition Works | The Learning Dispatcher — route learning jobs to the right environment.
• How Tuition Works | The Exit Ramp — reduce support as capability becomes independent.
• How Tuition Works | The Priority Queue — allocate scarce expert attention.
• How Tuition Works | The Intervention Threshold — decide when to enter a learner’s attempt.
• How Tuition Works | The Translation Layer — turn school and home signals into precise learning jobs.
• How Tuition Works | The Transfer Gate — test whether improvement travels beyond original conditions.
• How Tuition Works | The Diagnostic Probe — separate competing explanations of performance.
• How Tuition Works | The Evidence Threshold — decide when evidence is sufficient to change the plan.
• Returned Paper Review — use authentic school evidence to inspect what actually changed.

This completes Batch 03:

• The Transfer Gate — does the capability survive changed conditions?
• The Diagnostic Probe — which cause best explains the observed performance?
• The Evidence Threshold — do we know enough to change the plan?
• The Counterfactual Check — after performance changes, what most plausibly caused the change?

The evidence loop now becomes:

Observe → translate → probe → reach an evidence threshold → intervene → test transfer → compare the new state with the counterfactual story → keep, change, reduce or retire the intervention.

---

Research Foundations

The Counterfactual Check is an explanatory model for practical tuition evaluation. It is not a method for producing formal individual causal estimates from one learner’s before-and-after scores.

• Education Endowment Foundation — Tutor Trust Pilot (2026). EEF explains why randomised controlled trials use intervention and control groups so differences can be attributed more confidently to the programme. The page also demonstrates an important evaluation principle: implementation evidence can show that a programme is not yet ready for an impact trial.
• Education Endowment Foundation — National Tutoring Programme: Nimble RCTs (completed September 2026). These trials use randomised comparisons to estimate the effects of strategies intended to improve tutoring engagement and attendance, illustrating the role of comparison conditions in causal evaluation.
• Education Endowment Foundation — Action Tutoring Trial (current evaluation protocol and statistical analysis plan updated in 2026). The programme is being evaluated through a randomised controlled trial alongside implementation and process evaluation, separating questions about whether an intervention has an effect from questions about how and why implementation succeeds or fails.
• Institute of Education Sciences — The Estimation of Average Treatment Effects for Clustered RCTs of Education Interventions. The report is grounded in the potential-outcomes framework used for causal inference in experiments and discusses estimating treatment effects from randomised education evaluations.
• Institute of Education Sciences — Randomized Controlled Trials in Education Research. The report explains the counterfactual problem directly: the untreated outcome for the same participating learner cannot be observed, so a properly randomised control group is used to estimate what would otherwise have happened.
• MDRC / IES — Measuring Treatment Contrast in Randomized Controlled Trials. The paper shows that the meaning of an intervention effect depends partly on what the comparison group receives; when ordinary practice changes, the counterfactual changes too.

The practical conclusion is simple.

When a child improves, celebrate first.

Then ask what changed.

Give school its share.

Give practice its share.

Give family routines their share.

Give tuition the share the evidence supports.

And give the learner the growing capability that now belongs to them.