The Tutor Handbook Vol No.0036 | The Differential — How a Tutor Separates Old Cause, New Cause and Mixed Cause Before Choosing the Next Intervention

The Tutor Handbook · Volume 0036 · Series ID THB-0036

Sometimes the hardest tutoring decision begins after the tutor has already done something right.

The tutor notices a familiar problem. The learner misses a sign again. A comprehension answer expands beyond the evidence. A student who had learned to begin independently waits for a prompt. A Science explanation states what happened but does not connect the observation to the mechanism. The pattern looks old.

But looking old and being caused by the old mechanism are not the same thing.

The previous chapter, The Tutor Handbook Vol No.0035 | The Fresh Look, established the first discipline: history should generate hypotheses, not verdicts. This chapter takes the next step.

If several causes could produce the same visible difficulty, what evidence would let the tutor tell them apart?

That is the job of The Differential.

In this handbook, a differential is not a medical diagnosis and it is not a label attached to a child. It is a disciplined educational comparison between plausible explanations for the same observed learning or performance problem. The tutor asks whether the present signal is best explained by an old cause, a new cause, or a mixed cause—and then chooses the smallest useful test that can separate those possibilities before a major intervention begins.

This chapter sits inside the larger eduKate Sengkang architecture. How Learning Works maps durable change in capability. How Studying Works owns deliberate learner activity between teaching moments. eduKate Sengkang Education Runtime connects learner state, diagnosis, repair, practice, transfer and independence. How to Improve Anything owns the wider improvement loop. How Education Works holds the wider education-system frame. The Tutor Classification Model by eduKateSG remains the tutor ladder from Class 0 Homework Helper to Class 6 Learning Architect.

Quick Read

A visible problem is not yet a cause.
The old cause should remain one hypothesis, not the default conclusion.
A new cause can produce an old-looking symptom.
A mixed cause exists when the old mechanism still matters but interacts with a new condition or new weakness.
The tutor should compare explanations by prediction, not by familiarity.
The best discriminating test changes one meaningful condition at a time where practical.
Current capability, support demand, recovery speed, task load and transfer boundary are all useful evidence.
A helpful intervention does not automatically prove the diagnosis that motivated it.
Class 3 tutors separate mechanisms; Class 4 tutors convert the result into route design; Class 5 tutors test performance conditions; Class 6 tutors integrate the learner model across learning, studying, training and education.
When evidence is weak, uncertainty should remain visible.
The goal is not to produce a perfect explanation. The goal is to choose a better next educational move than guessing.

1. The Same Symptom Can Be Produced by Different Systems

A wrong answer is an output. It does not tell us by itself which internal process failed.

A Mathematics answer may be wrong because the concept is missing, the representation is weak, the method was selected incorrectly, the method was selected correctly but executed badly, the learner copied a value incorrectly, time pressure compressed the working, or the learner simply made a one-off slip. The page can show the same final cross in every case.

A comprehension answer may be too broad because the learner misunderstood the passage, misunderstood the question, located weak evidence, over-inferred from strong evidence, copied too much, or rushed because the clock was running down. A tutor who sees only “answer too broad” sees the end of the chain, not the branch point that produced it.

2. The Differential Begins With Three Cause Families

Old cause: the previously identified mechanism has genuinely become active again.
New cause: a different mechanism is creating a similar visible problem.
Mixed cause: the old mechanism still contributes, but the present failure cannot be explained without a new interacting condition or weakness.

These three families are broad on purpose. They give the tutor a clean first comparison before finer detail is added.

3. Old Cause Means More Than “I Have Seen This Before”

An old cause has to reproduce more than appearance. The mechanism, conditions and pattern should meaningfully resemble the earlier case.

If the learner previously lost algebraic signs specifically when several transformations were compressed into one mental step, an old-cause recurrence becomes plausible when the same sign errors reappear in dense transformations, intermediate working again disappears, and restoring one visible line improves control. That is stronger than simply noticing another negative-sign error.

4. New Cause Means the Surface Repeats but the Mechanism Changes

A new cause is especially common after transitions. The student moves into harder curriculum, longer papers, new question forms, different school expectations or greater independence. The visible error may resemble an old weakness even though the original mechanism remains repaired.

A learner who once forgot formulas because retrieval was weak may now remember the formula perfectly but apply it to the wrong relation because method selection has become the new bottleneck. The answer looks like “forgot the formula”. The actual route has changed.

5. Mixed Cause Is Often the Most Educationally Interesting State

Real learners do not always present one neat cause at a time.

A once-repaired retrieval weakness may remain mild, while a new workload increase makes it operationally important again. A learner may still have a slight tendency to rush, but that tendency becomes costly only when a longer paper reduces checking time. A student may understand evidence scope but lose it when a new text type raises reading load and leaves less working-memory capacity for answer control.

Mixed cause does not mean “everything is wrong”. It means the tutor should identify the interaction precisely enough to know which part deserves intervention.

6. Do Not Use “Mixed” as a Refuge From Thinking

Mixed cause can become lazy language if it means “probably a bit of everything”. A useful mixed explanation names the components and the interaction.

For example: “Retrieval remains slightly fragile after long gaps, but the present failures occur mainly because the new mixed-question format removes the chapter cue that used to support method selection.” That statement can guide a route. “Memory plus exam technique” is too broad.

7. Write the Symptom Before Writing the Explanation

Begin with a neutral description.

“The student changed +3x to −3x between line two and line three.”

Not: “The old careless-sign problem is back.”

“The student selected two sentences of evidence, then wrote a claim broader than both.”

Not: “Inference is weak again.”

Neutral description delays commitment long enough for the evidence to do some work.

8. Separate Symptom, Mechanism, Trigger and Consequence

Symptom: What could be directly observed?
Mechanism: What process may have produced it?
Trigger: What condition activated or amplified the mechanism?
Consequence: What capability actually failed because of it?

This separation matters because the same mechanism may produce different symptoms, and the same symptom may arise from different mechanisms.

9. Add a Fifth Field: Support Demand

How much help does the learner now need?

If an old weakness once required full explanation but now disappears after one brief cue, the learner state has changed materially even if the same surface error appears. Support demand acts like a depth gauge. It helps the tutor distinguish a genuine return to an earlier state from a lighter instability near the new boundary.

10. Add a Sixth Field: Recovery Speed

How quickly does the learner regain control after the failure is noticed?

Fast self-correction suggests a different state from prolonged dependence. A learner who once needed ten minutes of explanation but now needs five seconds to notice the contradiction has not returned to the original starting point.

11. Add a Seventh Field: Boundary

Where does the capability stop being reliable?

Routine questions? Changed wording? Mixed method choice? Long multi-step tasks? Timed work? Full-paper fatigue? Unfamiliar context?

Boundary movement is one of the clearest signs that learning changed. If failure once occurred on routine work and now appears only under full-paper pressure, an old-looking symptom is happening in a stronger system.

12. The First Differential Table

Question	Old Cause	New Cause	Mixed Cause
Does the old mechanism reproduce?	Usually yes	No or weakly	Partly
Are the old conditions present?	Often	Not necessary	Some old, some new
Has the learner boundary moved?	May regress toward old boundary	Often a new boundary	Old fragility appears at a new boundary
Does old support restore control?	Often strongly	May help little or incidentally	Helps partly
Is a new condition required to explain failure?	No	Yes	Yes, together with old mechanism
Route implication	Repair or reactivate old control	Build a new route	Repair interaction, not every component equally

13. A Differential Is a Competition Between Predictions

The tutor should not ask which explanation sounds most persuasive. Ask what each explanation predicts.

If the old cause is retrieval decay, then a small retrieval cue should have a strong effect, conceptual explanation may still be intact, and performance should worsen after delay. If the new cause is method selection, then naming the method should produce a sharp improvement in execution even without extra content teaching. If the cause is mixed, both a retrieval cue and a selection contrast may matter.

Predictions convert a story into something that can be checked.

14. Prefer the Test That Makes the Explanations Disagree

A weak test is one where all explanations predict the same result. A stronger test creates different expectations.

If both retrieval weakness and selection weakness predict that extra practice will help, “give more practice” is a poor discriminating test. Instead, hold the question constant and provide the formula without naming the method. Or name the method without giving the formula. The explanations now predict different patterns.

15. Change One Meaningful Condition at a Time Where Practical

Real classrooms are not laboratories, but the logic of controlled comparison is useful.

same concept, changed representation;
same representation, timed versus untimed;
same method, familiar versus unfamiliar surface;
same question, with and without a retrieval cue;
same task, with and without a planning scaffold;
same item family, isolated versus mixed;
same skill, early versus late in a long paper.

The contrast does not need to be perfect. It needs to be informative enough to improve the next decision.

16. Do Not Change Five Things and Call the Result a Diagnosis

If the tutor reteaches the concept, slows the pace, gives a hint, shortens the task and removes the timer—and the learner improves—we still do not know which change mattered.

Sometimes broad rescue is necessary, especially when a learner is distressed or a deadline is immediate. But rescue and diagnosis are different jobs. Once the learner is stable, return and separate the important variables.

17. The Retrieval Test

Ask whether the knowledge can be accessed without full reteaching.

Can the learner recall after a small cue?
Can they recognise the relation when shown a partial representation?
Can they explain the concept even when the exact term is unavailable?
Does retrieval deteriorate after delay?

If a tiny cue unlocks a coherent network, access may be the problem. If the learner cannot explain the relation even after the term is supplied, the weakness may sit deeper.

18. The Meaning Test

Fluent recall can hide weak understanding. Ask the learner to explain why the method works, generate an example and non-example, or predict what changes when one condition changes.

If the formula is remembered but meaning is fragile, more retrieval practice alone will not solve the problem.

19. The Representation Test

Some learners know the idea but cannot move between words, symbols, diagrams, tables or graphs. A new representation can therefore create an old-looking failure.

Keep the relation constant while changing representation. If performance changes sharply, representation may be the current bottleneck.

20. The Selection Test

Give several plausible methods and remove the chapter label. Ask the learner to choose and justify the choice before execution begins.

If the learner executes strongly once the correct method is named, the bottleneck sits before execution. This is especially important because topical practice can hide selection weakness by announcing the method in advance.

21. The Execution Test

Ask the learner to state the correct route first, then perform it. If the route is sound but the work breaks later, inspect execution: notation, arithmetic, sentence construction, unit control, copying, checking and step visibility.

22. The Time Test

Compare untimed and timed performance.

If the learner is accurate when untimed but deteriorates under a realistic clock, the present cause may sit in performance control rather than foundational understanding. If the same mechanism appears in both conditions, time may be an amplifier rather than the primary cause.

23. The Position Test

Where in a long task does failure appear?

Errors clustered in the final third suggest a different differential from errors distributed evenly across the whole paper. Position can reveal fatigue, pacing, reduced checking or attentional drift.

24. The Changed-Surface Test

Preserve the underlying structure but change wording, context, numbers or visual form. If success survives, the capability is more transferable than the symptom suggested. If performance collapses immediately, recognition may have been carrying more of the work than durable learning.

25. The Delayed Test

Immediate correction can flatter the apparent strength of learning. Return after a meaningful delay. If the learner can reconstruct the control without the original explanation in working memory, the old learning is more likely to remain healthy.

26. The Support-Removal Test

Success with help is useful evidence, but the differential becomes clearer when the support is removed again.

Hint → success → fresh item without hint.

If control remains, the hint may have reactivated an available capability. If performance collapses again immediately, the learner may still depend on external structure.

27. The Environment Test

Compare school, tuition, home, quiet practice and authentic examination conditions where evidence is available.

A pattern that appears only in one environment should not automatically become a global learner label. Different environments alter cueing, noise, pace, expectations, social pressure, tool access and task sequence.

28. The Workload Test

A study routine that worked when the learner managed four major demands may fail when eight demands compete. That does not prove the old initiation weakness returned unchanged.

Ask whether the planning horizon, number of competing priorities, switching cost and recovery time have changed. A stronger learner can still need a more advanced system when the environment becomes more complex.

29. The Tool Test

Calculators, formula sheets, model answers, vocabulary banks, worked examples and digital tools alter what the learner must carry internally. Compare conditions before concluding that capability changed.

A student may look stronger because a tool carries one component, or weaker because a familiar support has been removed. Neither observation should be interpreted without the tool condition.

30. The Error-Family Test

Do several wrong answers share the same mechanism?

Three wrong questions do not automatically form one pattern. A skipped condition, a copied digit and a method-selection failure belong to different families even if all three lose one mark. Group by process, not merely by outcome.

31. Build the Differential Ledger

Observed signal: what happened?
Old mechanism: what previously produced a similar signal?
Old conditions: when did it used to happen?
Current conditions: what is different now?
New alternative: what different cause could produce the same signal?
Mixed alternative: how could old and new mechanisms interact?
Prediction A: what should happen if old cause is primary?
Prediction B: what should happen if new cause is primary?
Prediction C: what should happen if the causes interact?
Discriminating task: what small comparison separates those predictions?
Result: what happened?
Confidence: strong, moderate, weak or unresolved?
Next move: repair, train, redesign route, monitor or release?

The ledger should stay small. Its purpose is to improve decisions, not create administrative theatre.

32. Use Evidence Weights, Not Evidence Counts

Five weak observations do not automatically outweigh one strong discriminating result.

If an old retrieval hypothesis predicts that a tiny cue should restore knowledge and the cue has no effect across several comparable tasks, that result may matter more than several vague moments of hesitation that “felt like forgetting”.

33. Direct Evidence, Indirect Evidence and Context Evidence

Direct evidence: the learner performs a task that directly tests the suspected mechanism.
Indirect evidence: the pattern is inferred from related behaviour.
Context evidence: workload, environment, timing or support conditions make one explanation more or less plausible.

Good tutoring uses all three but does not confuse them.

34. Strong Evidence Changes the Probability of Competing Explanations

A useful test should make at least one explanation less plausible.

If every new result can be interpreted as proof of the tutor’s favourite explanation, the explanation is too elastic to guide teaching.

35. Keep Disconfirming Evidence Visible

The tutor should actively record evidence that does not fit the leading hypothesis.

“Retrieval looked weak, but the learner explained the relation accurately and generated a correct example without prompting.” That contradiction may be the most useful line in the entire note.

36. Do Not Diagnose Personality From Task Failure

“Lazy”, “careless”, “weak memory”, “not a language person”, “panics in exams” and similar labels compress many possible mechanisms into identity statements.

The Differential replaces identity with conditions and operations: what happened, under what load, with which support, at what boundary, and how the learner recovered.

37. Class 0 · Homework Helper

Class 0 sees visible task completion and routine. If homework starts failing again, the Homework Helper first separates initiation, task understanding, priority conflict and environmental interruption.

The old cause may have been dependence on a parent prompt. The new cause may be unclear task requirements. The mixed cause may be mild initiation fragility interacting with a much larger workload. The Class 0 tutor should restore only the minimum structure needed.

38. Class 1 · Explainer

Class 1 separates missing meaning from missing access. If the learner cannot answer, the Explainer asks whether the concept is absent, whether the learner cannot retrieve the language, or whether a new representation made the familiar idea hard to recognise.

The wrong move is to deliver the same full explanation every time uncertainty appears.

39. Class 2 · Drill Builder

Class 2 separates fluency weakness from selection weakness. If a learner is slow, is the procedure still effortful, or is time being spent deciding which procedure applies?

More repetition helps the first problem. Mixed discriminative practice helps the second. A mixed cause may require both, but not in equal amounts.

40. Class 3 · Diagnostic Tutor

Class 3 owns the core differential move. The Diagnostic Tutor states at least two plausible mechanisms, identifies a condition under which their predictions differ, and runs the smallest safe test that can separate them.

Class 3 must also be comfortable writing “unresolved”. A forced answer can be worse than a visible uncertainty because a false diagnosis can send many future lessons in the wrong direction.

41. Class 4 · Route Designer

Class 4 asks what the differential means for sequence.

If old cause is confirmed, repair the earlier dependency. If new cause is stronger, build the new capability without dragging the learner backward unnecessarily. If mixed cause is most plausible, identify the first component that unlocks the interaction.

The Route Designer does not repair every plausible weakness. The Route Designer repairs the earliest useful constraint.

42. Class 5 · Performance Coach

Class 5 asks whether the mechanism changes under authentic performance load. Timing, switching, uncertainty, paper length and recovery after difficult items can reorganise the failure chain.

The Performance Coach therefore compares clean-skill evidence with full-condition evidence before sending the learner back into foundational repair.

43. Class 6 · Learning Architect

Class 6 integrates learning, studying, training, education and performance.

A learner may know the content, study poorly, train under narrow conditions and enter an educational environment that suddenly demands greater independence. The Learning Architect does not compress that system into “weak topic”. The job is to identify which layer is limiting the whole route now.

44. Learning Differential

Learning asks whether durable capability changed. Old cause may mean a previously weak concept or retrieval pathway genuinely destabilised. New cause may mean a different knowledge dependency has become limiting. Mixed cause may mean old fragility survives only at a harder transfer boundary.

Test meaning, retrieval, transfer, durability and independence.

45. Studying Differential

Studying concerns what the learner deliberately does to prepare, practise, retrieve, organise and revisit learning.

An old studying cause might be passive rereading. A new cause might be poor prioritisation across a larger workload. A mixed cause might be passive review returning because time pressure pushes the learner toward the fastest-feeling activity.

Do not reteach content when the failure sits in the learner’s study decisions.

46. Training Differential

Training asks what repeated exposure is adapting. A learner may understand a skill but lack speed, endurance, stability or recovery under a new condition.

Old cause may be lost fluency after a long gap. New cause may be a more demanding performance condition. Mixed cause may be mild fluency decay interacting with increased task length.

47. Education Differential

Education includes the system surrounding the learner: curriculum, teacher expectations, assessment design, school transitions, timetable, class environment and available support.

If a learner suddenly appears weaker after moving into a new stage, the differential must include environmental change. A stable capability can become insufficient when the system asks for more speed, abstraction, independence or integration.

48. Improvement Differential

The improvement system is iterative because every successful repair changes what becomes limiting next. A plateau after progress is therefore not automatically proof that the old bottleneck returned.

Ask whether the previous target remains limiting, whether a new bottleneck has become visible, or whether the learner is now operating at a boundary that did not exist before the improvement.

49. Alicia: Old Retrieval or New Representation?

Alicia once struggled to retrieve ideas after long gaps. Spaced retrieval stabilised the problem. Months later she cannot answer a question immediately.

The old-cause hypothesis says delayed retrieval weakened again. The new-cause hypothesis says the same concept is now represented in unfamiliar notation. The mixed hypothesis says retrieval remains slightly fragile and the unfamiliar notation removes cues that used to support access.

The tutor gives the same underlying relation in the old notation and the new notation. Alicia answers the old form immediately but hesitates on the new one. A small representational bridge restores performance and remains after the bridge is removed.

The differential moves away from “memory problem returned” and toward “representation transition”.

50. Beatrice: Old Method Selection or New Concept?

Beatrice previously executed methods well but selected poorly when similar methods competed. Later mixed practice became stable. A new chapter produces wrong choices.

The tutor asks Beatrice to explain the new mathematical relation before naming any method. Her explanation is incomplete. Once the relation is clarified, she selects correctly across a mixed set.

The visible symptom resembled the old selection problem. The stronger explanation is new conceptual structure.

51. Ciara: Old Answer Scope or New Reading Load?

Ciara once wrote comprehension claims broader than the evidence. She later became precise. A more difficult text brings broad answers back.

Untimed, Ciara still limits claims accurately once the relevant evidence is located. Timed, she spends longer finding the evidence and writes too quickly afterward.

The old scope control remains. The new bottleneck is reading-load allocation under time. The route becomes performance training rather than a full return to answer-scope instruction.

52. Denise: Old Carelessness or Final-Third Fatigue?

Denise begins making copying and sign errors again. The phrase “careless mistakes” would hide too much.

The tutor maps error position and discovers that the first two thirds are clean while the final third deteriorates. When the same items are attempted fresh, accuracy returns.

The differential points toward pacing, fatigue and visible-working loss late in the paper—not a broad return of weak accuracy.

53. Emily: Old Initiation Fragility or New Planning Complexity?

Emily once struggled to start work. A one-task restart rule eventually became independent. During a crowded school period, she stalls again.

If the tutor automatically restores the old start cue, the deeper problem may be missed. Emily can start any clearly chosen task. She is getting stuck deciding which of six competing deadlines should come first.

The old symptom is present. The cause has moved from initiation to prioritisation.

54. Faith: Old Science Knowledge Gap or Explanation Architecture?

Faith’s Science marks fall. Earlier in her learning history, content gaps mattered. Later, explanation structure became the stronger bottleneck.

The tutor checks concept retrieval first. Faith recalls the content and identifies the observation correctly but fails to connect observation to mechanism in writing.

The differential protects Faith from unnecessary content reteaching. The route targets evidence-to-mechanism explanation.

55. Primary English Example

A Primary writer once produced flat event lists because the plan captured sequence but not consequence. Causal planning fixed the issue. Months later, one composition becomes flat again.

Old cause: planning again lacks consequence. New cause: the new prompt requires emotional development the learner does not yet know how to construct. Mixed cause: planning is slightly compressed because the unfamiliar topic consumes idea-generation capacity.

Ask the child to tell the story orally, then build a causal plan. If oral development is rich but the written plan is thin, the old planning route deserves attention. If oral development is also thin, the new issue may be idea generation or domain knowledge.

56. Primary Mathematics Example

A learner once selected the wrong percentage base. Later work stabilised. A new percentage question is wrong.

Ask first: “What is this percentage a percentage of?” If the learner identifies the base correctly, do not reopen the old base-selection repair. Follow the chain into arithmetic, representation or interpretation.

57. PSLE Science Example

A learner once confused observation and explanation. Later they separated them well. A graph question now produces an unsupported causal statement.

Old cause: observation-inference distinction weakened. New cause: graph reading is weak. Mixed cause: graph uncertainty creates pressure to fill missing information with explanation.

Test graph reading separately, then ask for observation-only statements before mechanism is added. The route becomes clear quickly.

58. Secondary English Example

A Secondary learner who previously over-inferred begins losing comprehension marks again. The tutor compares evidence location, answer scope and timing.

If evidence is located accurately but answers expand under the clock, scope control is now condition-sensitive. If the wrong evidence is selected before writing begins, the new problem sits earlier in reading and evidence selection.

59. Secondary Mathematics Example

A learner who previously struggled with mixed method selection starts failing algebra questions. Before rebuilding selection, ask the learner to identify the relation and intended method without calculating.

Correct classification with poor execution points toward algebraic control. Incorrect classification with clean execution once cued points back toward selection. Both failing together may indicate mixed cause.

60. Additional Mathematics Example

An Additional Mathematics learner once lost control when symbolic transformations were compressed. Later the habit of visible intermediate working became stable.

Calculus introduces new algebraic slips. One extra written line removes some errors but not all. The remaining errors cluster around a new derivative relation.

This is a classic mixed cause: the old symbolic-compression fragility still contributes, but a new conceptual demand is also active. The route should not treat either component as the whole story.

61. JC Example

A JC learner becomes reactive about revision after previously developing good self-direction.

Old cause: tutor dependence returned. New cause: the decision space is now larger, with more papers, more topics and less time. Mixed cause: self-direction is mostly intact but breaks when prioritisation uncertainty rises beyond the learner’s current planning system.

The differential prevents the tutor from taking over again when the learner really needs a stronger decision framework.

62. Vocabulary Example

A learner once recognised advanced vocabulary but failed to produce it. Retrieval and deliberate use improved output. Later writing becomes plain again.

Old cause: production retrieval weakened. New cause: the topic requires domain vocabulary the learner never acquired. Mixed cause: known vocabulary remains available but time pressure pushes the learner toward safe, high-frequency words.

63. Homework Example

A child who learned to start homework independently misses several assignments.

Before rebuilding external reminders, ask whether the assignments were understood, deadlines recorded, materials available and priorities clear. Homework completion is an output produced by several systems.

64. The Smallest Useful Differential

Not every problem needs a large investigation. Often the entire differential can fit inside five questions:

What exactly happened?
What old cause does it resemble?
What new cause could produce the same thing?
What one task would make those explanations predict different results?
What should we do differently depending on the result?

65. The One-Lesson Differential

0–10 minutes: normal learner-state and retrieval check.
10–25 minutes: continue current target work.
25–35 minutes: when a meaningful repeated signal appears, collect one clean example.
35–45 minutes: state old, new and mixed possibilities.
45–55 minutes: run one discriminating contrast.
55–70 minutes: teach according to the strongest current evidence.
70–82 minutes: remove support and test independent application.
82–88 minutes: changed-surface, timed or delayed proxy check where suitable.
88–90 minutes: record only what changes the next lesson.

The differential should make teaching more efficient, not consume the whole class.

66. The Two-Lesson Differential

Some explanations cannot be separated immediately because durability or delay is part of the mechanism. In that case, Lesson 1 collects and repairs; Lesson 2 returns after the evidence has had time to mature.

For retrieval, transfer and maintenance questions, delayed evidence is often more valuable than further same-day repetition.

67. The Weekly Differential Review

At the end of a week, ask:

Did the signal repeat?
Did it repeat under comparable conditions?
Did support demand change?
Did the learner recover faster?
Did one explanation predict the week better than the others?
Did the boundary move?

If the signal does not repeat, do not manufacture a long-term problem from a one-off event.

68. The Monthly Differential Review

Monthly review is where the tutor can distinguish noise from a persistent mechanism. Compare error families, task difficulty, school evidence, tuition evidence, support demand and transfer range.

A strong learner model becomes simpler over time because weak explanations are retired rather than endlessly accumulated.

69. Parent Communication: Explain the Branch, Not the Drama

Parents often see the same surface pattern and understandably worry that an old problem has returned. A useful explanation is:

This looks similar to the earlier issue, so the old history is useful. We are checking whether the same mechanism is active, whether a new condition is producing a similar result, or whether both are interacting. The next task is designed to separate those possibilities before we change the programme.

This communicates seriousness without pretending certainty.

70. Student Communication: Keep Identity Out of It

Say: “This error looks similar to one we have seen before. I want to check whether it is happening for the same reason.”

Avoid: “You are back to your old problem.”

The first statement invites investigation. The second announces regression before evidence exists.

71. Failure Mode: The Old Diagnosis Wins by Seniority

Once a tutor has invested time in an explanation, that explanation can gain psychological weight. New evidence is then interpreted through the old model.

The Differential prevents this by forcing at least one credible alternative into the comparison.

72. Failure Mode: Every New Error Becomes a New Cause

The opposite error is fragmentation. Every wrong answer receives a new explanation, and the learner model becomes a catalogue of unrelated defects.

Look for mechanism families. Several symptoms may arise from one upstream process.

73. Failure Mode: “Mixed Cause” Means We Stop Separating

If mixed cause is selected too early, it can absorb every contradiction. Require the tutor to name the old component, the new component and the condition under which they interact.

74. Failure Mode: The Intervention Is Mistaken for the Test

A learner improves after slower pacing. That does not prove the cause was rushing. Slower pacing may also improve retrieval, reading and working-memory control.

Whenever several mechanisms could respond to the same support, treat the response as suggestive rather than conclusive.

75. Failure Mode: The Tutor Tests Too Much

Diagnostic curiosity can become expensive. If one small contrast is enough to choose between two sensible routes, do not spend the next forty minutes proving a theory more precisely than teaching requires.

The tutoring standard is decision usefulness, not perfect causal certainty.

76. Failure Mode: The Tutor Tests Too Little

The opposite mistake is to call the first plausible story a diagnosis. One extra paired item, one timed-versus-untimed contrast or one support-removal check can sometimes prevent weeks of misdirected practice.

77. Failure Mode: Score Changes Are Treated as Mechanism Changes

A score is the result of content, question mix, time, difficulty, marking and execution. A lower mark is useful evidence but does not identify the cause by itself.

78. Failure Mode: Difficulty Change Is Ignored

A learner can improve while the score stays flat if the task becomes harder. Conversely, a score can rise because the task becomes easier.

The differential therefore compares capability relative to demand, not score in isolation.

79. Failure Mode: Support Conditions Are Not Recorded

“Got it correct” means little if one attempt was independent and the other followed five tutor prompts. Support is part of the condition and must remain visible when comparing states.

80. Failure Mode: Historical Notes Become Permanent Truth

Learning history should age. Old explanations can remain useful, become conditional, be superseded or lose relevance entirely.

A healthy learner model has version history, not permanent labels.

81. What Research Supports

There is no universal educational research protocol called “The Differential” in the exact form used here. This handbook chapter combines several established ideas: the distinction between learning and immediate performance, the value of testing competing hypotheses, the risk of confirmation bias, the importance of retrieval and transfer, and the use of changed conditions to reveal what a learner can actually carry independently.

These sources support the general logic of cautious inference, discriminating evidence and transfer checks. They do not prescribe the exact Class 0–6 differential workflow used by eduKate.

82. What Research Does Not Justify

Research does not justify diagnosing a child from one worksheet, assigning personality explanations to bounded errors, assuming one helpful strategy proves one mechanism, or treating educational support as a substitute for professional medical or psychological assessment when those domains are genuinely involved.

The Tutor Handbook remains an educational operating guide: observe learning and performance, compare plausible educational mechanisms, choose proportionate instructional tests and adjust teaching.

83. The Differential Confidence Scale

Weak: one signal, several plausible causes, little discriminating evidence.
Moderate: pattern repeats and one explanation predicts more of the evidence.
Strong: the mechanism reproduces under comparable conditions and alternative explanations fail key tests.
Operationally sufficient: evidence is strong enough to choose a low-risk next intervention even if perfect certainty is impossible.

“Operationally sufficient” is often the most useful tutoring threshold. Teaching must continue while evidence develops.

84. The Differential State Model

Signal: a meaningful problem appears.
Old-Cause Candidate: history provides a plausible mechanism.
New-Cause Candidate: a different mechanism can explain the same output.
Mixed-Cause Candidate: old and new mechanisms may interact.
Discriminating Test: one condition is changed to separate predictions.
Provisional Lead: one explanation now fits best.
Route Trial: smallest useful intervention is applied.
Independent Check: support is reduced and performance is retested.
Delayed/Transfer Check: the result is tested beyond the immediate repair.
Model Update: old cause confirmed, new cause established, mixed cause specified or uncertainty preserved.

85. The Differential Operating Cycle

Observe the signal → describe it neutrally → retrieve relevant history → state the old mechanism → name a credible new mechanism → specify a possible interaction → compare conditions → compare support demand → compare boundary → write competing predictions → choose one discriminating contrast → run the smallest useful test → keep disconfirming evidence visible → select the leading explanation provisionally → apply the smallest route change → remove support → retest after delay or variation → update the learner model only as far as the evidence earns.

86. When Old Cause Wins

Old cause earns priority when the earlier mechanism reproduces under comparable conditions, support demand rises toward the old level, the old boundary reappears, and the old repair predicts the pattern better than credible alternatives.

Then repair the mechanism cleanly. Do not punish the learner for recurrence. Systems drift. Skills can weaken. Conditions can reactivate old fragilities. The value of history is that the route back may now be shorter.

87. When New Cause Wins

New cause earns priority when the old mechanism fails to reproduce, old support has little specific effect, the learner remains strong at the old boundary, and a new condition consistently predicts the failure.

Then build forward. Do not send a stronger learner backward because the symptom happens to look familiar.

88. When Mixed Cause Wins

Mixed cause earns priority when neither the old nor new mechanism alone explains the pattern, but together they predict when failure appears and when it disappears.

The tutor then asks which component is the first useful lever. Sometimes a small restoration of the old control is enough for the learner to meet the new demand. Sometimes the new condition must be trained directly while the old fragility is merely monitored.

89. When No Cause Wins Yet

Sometimes the correct conclusion is uncertainty.

If the signal is low-cost, continue normal teaching while collecting better evidence. If the consequence is high, choose a safe intervention that helps across several plausible mechanisms without locking the learner into a permanent diagnosis.

Uncertainty is not indecision when it is managed deliberately.

90. The Final Tutor Standard

A weak tutor asks, “What worked last time?”

A better tutor asks, “Is this the same problem?”

A stronger tutor asks, “What evidence would separate the old cause from the strongest new alternative, and what would each result make me do next?”

The strongest tutor also asks whether both explanations may be partly right, whether the learner’s boundary has moved, whether task demand changed, and whether the simplest useful route can preserve independence instead of rebuilding unnecessary support.

Evidence and Connected Reading

Final Compression

A familiar symptom gives you history. It does not give you the answer.

Write what happened. Separate symptom from mechanism. State the old cause precisely. Name a credible new cause. Ask whether the two could interact. Compare the learner’s current support demand, recovery speed, task conditions and capability boundary.

Then make the explanations disagree.

Change one meaningful condition. Use a paired item. Add a retrieval cue without naming the method. Name the method without reteaching the content. Compare timed with untimed. Compare familiar with changed surface. Compare early with late in the paper. Remove the support again.

If the old mechanism reproduces, repair it. If a new mechanism explains the evidence better, build forward. If both are active, identify the interaction and repair the first useful lever. If the evidence remains weak, preserve uncertainty and keep teaching.

The tutor does not need to know everything before acting. The tutor needs enough evidence to make the next move more intelligent than habit.

That is The Differential.

That is Tutor Handbook Volume 0036.

Next in the series: The Tutor Handbook Vol No.0037 | The Branch Point — How a Tutor Chooses the Smallest Intervention After the Cause Has Been Separated.

Small Group Tutorials