H1 General Paper Paper 2 Feedback | Turn Marked Comprehension Papers Into Better Unseen Performance

41. Paper 2 Feedback Casebook | Sixty Failure Patterns, Diagnoses and Transfer Tests

The cases below are not a catalogue of “common mistakes” for memorisation. Each case shows a different feedback problem: the visible loss, the likely first weak decision, the smallest useful repair and the transfer condition that proves whether the learner has actually changed.

Case 1 — Correct passage, wrong operation

The learner finds the right lines but answers literally when the question asks for inference.

Diagnosis: operation recognition. Repair: label the task before using the evidence. Transfer: mix literal and inference questions on a new passage without labels.

Case 2 — Right operation, wrong passage location

The learner understands that an inference is required but selects a related line rather than the proposition the question targets.

Diagnosis: conceptual locating. Repair: paraphrase the question target before searching. Transfer: use a passage where the same keyword appears in several paragraphs.

Case 3 — Keyword match selects a distractor

The learner hunts for the question’s vocabulary rather than its meaning.

Repair: write a short conceptual synonym for the target before re-entering the passage. Transfer: use questions whose wording deliberately differs from the passage.

Case 4 — Literal answer copies too much

The correct phrase appears, but extra material introduces a second idea or changes scope.

Diagnosis: minimum sufficient completeness. Repair: underline only the words required to satisfy the command. Transfer: compare insufficient, sufficient and overextended answers.

Case 5 — Literal answer copies too little

The learner omits a condition, contrast or second required element.

Repair: mark the quantity and relationship in the question before writing. Transfer: use two-part items with different grammatical forms.

Case 6 — Inference repeats the evidence

The learner quotes or paraphrases the clue without stating what it implies.

Repair: complete the sentence “This suggests that…” and then remove the frame if unnecessary. Transfer: use new clues with no topic overlap.

Case 7 — Inference becomes a story

The learner invents motives, emotions or events not required by the evidence.

Diagnosis: boundary control. Repair: write clue → relationship → inference → what is not proven. Transfer: compare strong inference with speculation across several passages.

Case 8 — Inference imports background knowledge

The candidate knows the issue well and answers from prior knowledge.

Repair: require one textual clue before any inference. Transfer: use an unfamiliar topic where outside knowledge cannot rescue the task.

Case 9 — Inference is too strong

The passage supports concern; the answer says outrage. The passage supports reluctance; the answer says refusal.

Repair: calibrate evaluative force. Transfer: confidence-band drills on new wording.

Case 10 — Inference is too weak

Strong evidence is reduced to “maybe”.

Repair: distinguish direct evidence, strong inference, tentative inference and speculation. Transfer: mixed-strength evidence set.

Case 11 — Wrong pronoun referent

A whole answer follows the wrong antecedent.

Repair: replace the pronoun with each plausible referent and test coherence. Transfer: passages with concept-level rather than noun-level reference.

Case 12 — “This” refers to a whole proposition, not one noun

The learner hunts for the nearest noun.

Repair: paraphrase the full preceding idea. Transfer: track reference across sentence and paragraph boundaries.

Case 13 — Some becomes all

A small quantifier disappears in paraphrase.

Diagnosis: semantic-strength loss. Repair: verify population after rewriting. Transfer: mixed quantifier set across Passage 1 and Summary.

Case 14 — May becomes will

A possibility becomes certainty.

Repair: preserve modal force. Transfer: different topic and different sentence structure.

Case 15 — Contributes to becomes causes

The relationship is strengthened beyond the passage.

Repair: label the original relation before paraphrasing. Transfer: association/contribution/cause contrast set.

Case 16 — More effective than becomes effective

The comparison target disappears.

Repair: preserve direction and comparison object. Transfer: comparative claims in Summary and Connections.

Case 17 — Intended effect becomes actual effect

The passage says a policy aims to reduce harm; the answer says it reduces harm.

Repair: distinguish purpose from outcome. Transfer: proposal, pilot and outcome statements.

Case 18 — Survey result becomes population truth

Respondents become everyone.

Repair: preserve sample boundary. Transfer: evidence-evaluation questions using different samples.

Case 19 — Tone guessed from topic

A serious issue is labelled angry or pessimistic with no language evidence.

Repair: require diction, syntax or rhetorical movement as proof. Transfer: same topic written in different tones.

Case 20 — Tone label stronger than the wording

Sceptical becomes contemptuous.

Repair: build a force ladder and justify the chosen level. Transfer: changed author stance.

Case 21 — Technique label replaces analysis

“Metaphor”, “rhetorical question” or “short sentence” is written as if it were an effect.

Repair: wording → contextual meaning → precise effect or function. Transfer: unfamiliar prose using different devices.

Case 22 — Generic “creates impact” answer

The learner cannot name the actual impression or argumentative function.

Repair: ask what specific understanding changes for the reader. Transfer: effect vocabulary generated from context, not memorised lists.

Case 23 — Short sentence automatically called dramatic

The answer relies on a stock rule.

Repair: identify whether the shortness creates finality, interruption, emphasis, contrast or pace. Transfer: several short sentences with different functions.

Case 24 — Anecdote described, purpose missed

The learner retells the story.

Repair: state what the anecdote does for the argument: humanises, exemplifies, establishes stakes, creates contrast or limits abstraction. Transfer: new anecdotes serving different roles.

Case 25 — Statistic described, function missed

The answer repeats the number.

Repair: identify whether it establishes scale, trend, prevalence, comparison or credibility. Transfer: statistics with different inferential jobs.

Case 26 — Claim confused with reason

The learner cannot tell what the writer wants accepted versus why.

Repair: map conclusion and support separately. Transfer: short arguments with reordered sentences.

Case 27 — Evidence confused with assumption

The unstated bridge is treated as if it were observed fact.

Repair: claim → reason → evidence → assumption. Transfer: different arguments with hidden assumptions.

Case 28 — Assumption treated as automatically bad

The learner believes every assumption invalidates an argument.

Repair: ask whether the assumption is necessary and reasonable. Transfer: compare plausible and weak assumptions.

Case 29 — Credible source treated as conclusive

A reputable institution’s data are assumed to prove the writer’s interpretation.

Repair: separate data reliability from inference strength. Transfer: strong source, weak causal inference.

Case 30 — Biased source dismissed completely

The learner throws out evidence because the source has an interest.

Repair: identify what information may still be useful and where framing or selection needs caution. Transfer: advocacy and institutional sources.

Case 31 — Recent evidence called strong because it is recent

Recency substitutes for relevance or representativeness.

Repair: evaluate measure, sample, context and inference. Transfer: compare recent weak evidence with older strong evidence.

Case 32 — Old evidence dismissed automatically

Age is treated as fatal even where the mechanism is stable.

Repair: ask whether the time difference affects applicability. Transfer: historical case with enduring mechanism.

Case 33 — Summary target is misread

The learner selects related ideas but from the wrong category.

Repair: restate the target in plain language before point collection. Transfer: same passage with two different summary targets.

Case 34 — Summary misses a point because it looks like an example

The learner removes a specific detail that actually carries a distinct mechanism.

Repair: distinguish illustrative detail from independently markable idea. Transfer: examples where some details are points and some are support.

Case 35 — Summary keeps examples instead of the underlying point

The learner fills the word budget with illustrations.

Repair: abstract the general reason or consequence. Transfer: paragraph with several examples of one mechanism.

Case 36 — Summary merges distinct ideas

Cost, access and trust become “problems”.

Repair: preserve distinct mechanisms even when compressing. Transfer: clustering set with near-overlap.

Case 37 — Summary counts one idea twice

Two source sentences express the same underlying point.

Repair: cluster before drafting. Transfer: deliberate repetition across paragraphs.

Case 38 — Summary paraphrase is too close

Word substitution leaves source syntax almost intact.

Repair: close source and reconstruct proposition. Transfer: changed sentence structures.

Case 39 — Summary paraphrase is independent but inaccurate

Original wording is changed successfully but meaning drifts.

Repair: run scope, strength, relation and attribution checksum. Transfer: changed topic with the same semantic risks.

Case 40 — Summary is accurate but too slow

The learner repeatedly rewrites acceptable sentences.

Diagnosis: decision closure, not comprehension. Repair: timed compression with reason-before-change rule. Transfer: new summary under moderate clock.

Case 41 — Summary is fast but incomplete

Point selection is rushed.

Repair: time selection separately from writing. Transfer: skeleton-only drills.

Case 42 — Connection answer compares topics

“Both discuss education.”

Repair: write each proposition first and name the comparison dimension. Transfer: same topic, opposite propositions.

Case 43 — Connection answer overstates disagreement

One writer qualifies; the learner writes “completely disagrees”.

Repair: classify contradiction, qualification, narrowing or extension. Transfer: relation contrast set.

Case 44 — Connection answer misses hidden agreement

Examples and vocabulary differ but underlying proposition is shared.

Repair: abstract both claims before comparing. Transfer: different-word same-concept pairs.

Case 45 — Connection answer finds lines but no relation

The learner writes Passage A, then Passage B.

Repair: relation first, evidence second. Transfer: sentence frame removed after several guided examples.

Case 46 — Same word, different concept

Both passages use “freedom”, but one means freedom from state interference and another practical capacity to choose.

Repair: define the concept in each passage. Transfer: repeated vocabulary with shifted meanings.

Case 47 — Application begins with local knowledge instead of the writer

The paragraph becomes a Singapore essay.

Repair: passage proposition first. Transfer: new theme with the same local example bank.

Case 48 — Application represents the writer inaccurately

The local evidence is good but tests a stronger claim than the passage makes.

Repair: preserve scope and strength before applying. Transfer: changed writer position on the same theme.

Case 49 — Application names a local example but never explains fit

Singapore appears only as a proper noun.

Repair: identify the local mechanism and compare it with the writer’s mechanism. Transfer: different local context.

Case 50 — Application forces agreement

The learner assumes a passage idea must apply locally.

Repair: allow partial or weak applicability where conditions differ. Transfer: passage claim intentionally poorly fitted to the local context.

Case 51 — Application rejects by saying “Singapore is different”

Difference is asserted, not analysed.

Repair: identify the condition that breaks transfer. Transfer: another cross-context comparison.

Case 52 — Application repeats one familiar local example

The example is forced into multiple unrelated claims.

Repair: organise context by mechanism rather than memorised paragraph. Transfer: retrieve different examples for trust, regulation, inequality and access.

Case 53 — Application contains evidence but no judgement

The paragraph lists cases.

Repair: conclude how strongly the writer’s idea applies and under what condition. Transfer: new passage idea with two local examples of different strength.

Case 54 — Application is balanced but indecisive

The learner describes both fit and mismatch without hierarchy.

Repair: identify the decisive local condition. Transfer: contexts where evidence pulls in both directions.

Case 55 — Summary language becomes telegraphic

Compression produces fragments and unclear reference.

Repair: concise complete sentences. Transfer: short summary under word pressure.

Case 56 — Application language becomes inflated

Formal register produces long, unclear sentences.

Repair: simplify syntax while preserving evaluation. Transfer: timed paragraph with one-main-clause discipline.

Case 57 — Passage 1 is excellent; Application begins too late

Early perfection creates late time debt.

Repair: section closure and bounded review. Transfer: full paper with timestamps.

Case 58 — Summary consumes excessive time

The learner searches for perfect synonyms.

Repair: meaning-first paraphrase and reason-before-rewrite rule. Transfer: timed summary where selection and writing are logged separately.

Case 59 — Connections consume excessive time

The learner rereads entire passages for each comparison.

Repair: preserve compact proposition maps and relation-first search. Transfer: passage pair with several connection items.

Case 60 — One difficult early item damages the rest of the paper

The candidate keeps returning mentally to the missed question.

Repair: bounded attempt, visible return cue, reset routine. Transfer: simulation with one deliberately difficult early item.

42. Reduce sixty cases to nine repair families

1. Operation errors: wrong intellectual task is performed.
2. Location errors: wrong proposition or source section is selected.
3. Semantic errors: quantity, certainty, causality, comparison or attribution changes.
4. Inference/analysis errors: reasoning bridge is absent or overextended.
5. Selection/compression errors: summary target, point set or clustering fails.
6. Relation errors: cross-passage agreement, qualification, extension or contradiction is misclassified.
7. Application errors: passage idea, contextual mechanism or judgement fails.
8. Language-construction errors: meaning is understood but not expressed clearly or faithfully.
9. Execution errors: time, closure, recovery or checking causes otherwise-known skills to collapse.

43. The symptom-to-cause matrix

44. The repair-to-proof matrix

45. Mark patterns are clues, not diagnoses

A learner can lose the same number of marks through completely different mechanisms. One may miss content because they misunderstand the passage; another because they know the answer but write too vaguely; another because late-paper time runs out. The mark points toward investigation, not directly toward treatment.

46. Error recurrence matters

One accidental pronoun mistake is local. The same source-attribution error across Passage 1, Connections and Application is system-level. Track recurrence across tasks and papers.

47. Error severity matters

A rare error can be high priority if it changes the writer’s central claim or causes a 12-mark Application task to begin too late. Frequency is not the only signal.

48. Transfer value matters

Prioritise repairs that improve several tasks: claim-strength control, operation recognition, answer scope, source identity, evidence relation and time closure have broad Paper 2 value.

49. Repair cost matters

Late in preparation, choose changes that can become stable before the examination. A complex new annotation system may have lower expected value than improving closure on existing answer routines.

50. The casebook checkpoint

A marked Paper 2 is ready to drive deliberate practice when the learner can identify one primary failure family, state the repair as an observable decision and name the unseen condition that will be used to test transfer.

51. Forty deliberate-practice laboratories for Paper 2 feedback

These laboratories are designed to repair decisions, not to generate busywork. Use the smallest task that isolates the active failure, then vary text, topic and timing until the skill becomes portable.

Lab 1 — Operation sort

Mix literal, inference, language, purpose, evaluation, synthesis and application questions. The learner labels the operation before solving. Remove labels after several sets.

Lab 2 — Operation without keywords

Rewrite question wording so familiar verbs disappear. The learner must recognise the intellectual job from the demand, not a memorised trigger word.

Lab 3 — Proposition location

Give one question and three candidate passage lines. The learner chooses which proposition truly answers the target and explains why the others are related but insufficient.

Lab 4 — Keyword trap

Use a passage where the question keyword appears several times. Require conceptual paraphrase of the target before searching.

Lab 5 — Minimum complete literal answer

Compare three responses: incomplete, sufficient and overextended. The learner identifies the exact point at which completeness is reached.

Lab 6 — Quantity control

Mix questions requiring one reason, two reasons, a before/after contrast or both groups. The learner marks required quantity before answering.

Lab 7 — Clue-to-inference bridge

Provide a clue. The learner writes the relationship and then the inference. No answer may skip the reasoning step during training.

Lab 8 — Inference boundary

For each inference, complete “The evidence does not prove…” This develops control over speculation.

Lab 9 — Confidence band

Classify answers as direct evidence, strong inference, tentative inference or speculation. Rewrite language to match the band.

Lab 10 — Outside-knowledge quarantine

Use a familiar theme. The learner must cite passage evidence before writing any inference. Afterwards, discuss where outside knowledge would become legitimate in Application instead.

Lab 11 — Quantifier preservation

Paraphrase statements containing some, many, most, rarely, often and generally. Compare source and answer population.

Lab 12 — Modal preservation

Paraphrase may, might, can, should, must and will without changing commitment.

Lab 13 — Causation ladder

Differentiate follows, coincides with, is associated with, contributes to, enables, causes and guarantees.

Lab 14 — Comparison preservation

Paraphrase more effective than, less suitable than, equally important and better suited to without losing the comparison target.

Lab 15 — Attribution preservation

Mix author claims, critic claims, survey findings and participant reports. The learner must preserve who owns each proposition.

Lab 16 — Reference reconstruction

Resolve this, they, it, such measures and these effects by replacing them with the full referent.

Lab 17 — Tone evidence

Give three short passages on the same issue with different tones. The learner must support the label with diction and structure rather than topic.

Lab 18 — Wording → meaning → effect

Train language-for-effect without technique labels first. Add terminology only after contextual meaning is clear.

Lab 19 — Purpose function

For anecdotes, statistics, questions, concessions and examples, ask what role the choice performs in the argument.

Lab 20 — Claim–reason–evidence–assumption map

Use short arguments. The learner marks each function and identifies the bridge that could be challenged.

Lab 21 — Evidence-bearing capacity

For a case study, survey and statistic, write what each can establish and what would require stronger evidence.

Lab 22 — Source versus inference strength

Use a reputable source making an overextended conclusion and a less prestigious source providing useful direct evidence. Compare data reliability with argumentative force.

Lab 23 — Summary target switch

Use the same passage with two different targets, such as reasons versus consequences. Compare selected point sets.

Lab 24 — Point or example?

Classify source material as main point, explanation, example, qualification or repetition. Select only what the target requires.

Lab 25 — Summary clustering

Give six candidate points with partial overlap. The learner groups duplicates without merging genuinely distinct mechanisms.

Lab 26 — Summary skeleton

Build point notes only. No prose. This isolates selection from writing.

Lab 27 — Closed-text paraphrase

Read selected summary points, hide the passage and reconstruct the meaning. Reopen and check scope, strength, cause and comparison.

Lab 28 — Compression ladder

Reduce a passage through several lengths while naming what is removed at each stage. Stop when meaning damage begins.

Lab 29 — Connection proposition alignment

Write A’s proposition, B’s proposition and the exact comparison dimension before using a relation label.

Lab 30 — Relation classification

Sort pairs into agreement, partial agreement, contradiction, qualification, extension, narrowing, different level, different time or different criterion.

Lab 31 — Relation-first answer

Write the connection sentence before locating quotations or line references. Then add evidence.

Lab 32 — Same-word trap

Use passages where the same word carries different meanings. The learner defines the concept in each text before comparing.

Lab 33 — Different-word same-concept

Use different vocabulary for compatible propositions. The learner identifies hidden agreement.

Lab 34 — Application anchor

Before local evidence is allowed, write one accurate sentence stating the passage idea being tested.

Lab 35 — Mechanism match

Given three local examples, choose which one operates through the same mechanism as the writer’s claim.

Lab 36 — Application mismatch

Use a local case where the passage claim applies poorly. The learner must explain which condition breaks transfer rather than forcing agreement.

Lab 37 — Application judgement

After local evidence, require a final sentence stating how strongly the passage idea applies and why.

Lab 38 — Timed section reconstruction

Take the active repair into a timed Passage 1, Summary, Connections or Application segment before a full paper.

Lab 39 — No-cue unseen transfer

Use a new theme, remove the feedback label and test whether the learner independently makes the repaired decision.

Lab 40 — Full-paper retirement test

After repeated section success, run a representative Paper 2 with no visible cue. If the skill survives, move it to maintenance.

52. How to choose the laboratory

If the problem is early comprehension, use Labs 1–10. If semantic distortion recurs, use Labs 11–16. If language and argument analysis are weak, use Labs 17–22. If Summary is the bottleneck, use Labs 23–28. If Connections are weak, use Labs 29–33. If Application is weak, use Labs 34–37. If transfer is the issue, use Labs 38–40.

53. The Paper 2 repair ladder

explain → guided reconstruction → independent reconstruction → changed line → changed passage → changed task → delay → timed section → full-paper integration.

Do not jump from explanation to another full mock and assume failure means the learner did not understand. The full paper adds many competing demands.

54. Immediate reconstruction tests understanding

If the learner cannot repair the original answer with the source present and the feedback understood, explanation is incomplete. Clarify the mechanism before transfer.

55. Changed-line transfer reduces copying

One sentence with the same semantic structure can reveal whether the learner has extracted the rule rather than remembered wording.

56. Changed-passage transfer tests portability

Move from technology to culture, science, economics or society. The reading decision should survive topic change.

57. Changed-task transfer tests system learning

Use a repair across task boundaries. Claim-strength control learned in Passage 1 should help Summary and Application. Source identity learned in Connections should help Application.

58. Delayed transfer tests durability

Fresh correction can temporarily inflate performance. A later unseen retest gives stronger evidence that the better decision is becoming retrievable.

59. Timed transfer tests execution

A skill can be conceptually understood but too slow to deploy under exam load. Add timing only after untimed or moderate-load accuracy is reasonably stable.

60. Full-paper transfer tests coordination

The final question is not whether the skill exists in isolation. It is whether it remains available while the learner reads three passages, switches task types and manages the whole clock.

61. Cue design for Paper 2

Useful temporary cues are short and decision-specific:

• “Operation before evidence.”
• “Meaning, not keyword.”
• “What does the clue justify?”
• “Keep the qualifier.”
• “Target before points.”
• “Relation before lines.”
• “Writer before Singapore.”
• “Close and move.”

62. Bad cues are vague

“Read carefully.”

“Be more analytical.”

“Use own words.”

These may be true but do not identify a controllable decision.

63. Cue fading

Move from full instruction to short cue to self-generated question to no visible prompt. Record where the repaired behaviour disappears.

64. Support fading

Remove support progressively:

1. teacher identifies the error;
2. teacher identifies the family;
3. learner identifies the family;
4. learner predicts the risk before attempting;
5. learner detects after attempting;
6. learner prevents the error during independent work.

65. The learner should predict feedback

Before seeing the marked paper, ask:

• Which answer felt least secure?
• Where did you overread or underread?
• Which section took longer than planned?
• Where might claim strength have shifted?
• Which Application example had the weakest fit?

Prediction develops self-monitoring.

66. The learner should compare prediction with actual feedback

Accurate self-prediction is itself a useful signal. If the learner repeatedly cannot see the issue that teachers identify, metacognitive monitoring becomes a training target.

67. Feedback should preserve successful behaviour

When repairing one weakness, name a strength that must remain. A learner fixing slow Summary writing should not become careless with selection. A learner fixing overcautious inference should not start overclaiming.

68. Do not practise only failures

Strong answers can reveal useful procedures: why the learner located evidence efficiently, why one paraphrase preserved the writer exactly, or why one Application paragraph used local evidence proportionately. Extract strengths into reusable decision rules.

69. The strength-to-procedure conversion

After a strong answer, complete:

This worked because I ______. I should repeat that decision when ______.

This makes success transferable too.

70. Passage 1 repair cycle

1. Identify operation.
2. Locate evidence.
3. State source meaning.
4. Identify the lost relationship.
5. Repair answer.
6. Use new item of the same operation.
7. Use another operation sharing the same semantic control.
8. Retest under section time.

71. Summary repair cycle

1. Restate target.
2. Rebuild point set.
3. Classify each missed or duplicated point.
4. Cluster overlap.
5. Paraphrase selected points.
6. Run semantic checksum.
7. Change passage and target.
8. Add time only after selection is stable.

72. Connections repair cycle

1. Represent A accurately.
2. Represent B accurately.
3. Name comparison dimension.
4. Classify relation.
5. Choose evidence.
6. Write synthesis.
7. Change passage pair.
8. Remove relation menu.

73. Application repair cycle

1. Represent writer idea.
2. Identify mechanism.
3. Retrieve local evidence.
4. Check mechanism fit.
5. Explain similarity or mismatch.
6. State contextual judgement.
7. Change passage idea.
8. Change local example.

74. Language repair cycle

1. Identify meaning-changing language error.
2. Explain what meaning changed.
3. Repair source fidelity or argument clarity.
4. Use a new sentence with the same risk.
5. Test across Summary and Application task voices.
6. Add moderate timing.

75. Timing repair cycle

1. Instrument the paper with timestamps.
2. Locate the first significant time debt.
3. Identify the decision creating the debt.
4. Practise that decision separately.
5. Retest the section.
6. Retest the full paper.

76. Overannotation repair

Give the learner a passage and allow only five marks. They must choose central claim, major shift and crucial qualification. Then answer questions. Compare re-entry efficiency with their previous dense annotation.

77. Underannotation repair

If the learner repeatedly rereads whole passages, require one-line passage identities and paragraph-function notes during practice. Fade physical notes later.

78. Slow rereading repair

After first reading, close the passage and state the structure from memory. Reopen only to verify. This trains a global map rather than constant local re-entry.

79. Unknown-word repair

Classify unfamiliar words as central, inferable or currently ignorable. Do not stop the entire reading process for every unknown item.

80. Paraphrase-perfectionism repair

Once the new sentence is faithful, grammatical and sufficiently independent, move. Rewriting repeatedly can increase semantic drift.

81. Connection-perfectionism repair

Do not search for the most sophisticated relation label. A precise plain-language comparison is enough if it captures the actual relationship.

82. Application-knowledge-dump repair

Restrict each paragraph to one passage idea, one or two local examples and one judgement. Extra local material must earn its place through a distinct analytical function.

83. Application-example panic repair

If the ideal example cannot be recalled, use a second known case, narrow the claim or reason from a verified institutional pattern. Do not invent a statistic.

84. The no-fabrication rule

Paper 2 Application rewards contextual evaluation, not fictional precision. Uncertain dates or numbers should be omitted or stated more generally unless known confidently.

85. The reason-before-change rule during checking

Do not rewrite an answer merely because it “sounds odd”. Change only when you can identify a source mismatch, logical problem, grammar issue or clarity failure.

86. Checking order for Passage 1

1. Operation.
2. Evidence/source.
3. Inference strength.
4. Quantity.
5. Language clarity.

87. Checking order for Summary

1. Target coverage.
2. Distinct points.
3. Semantic preservation.
4. Compression.
5. Grammar and clarity.

88. Checking order for Connections

1. Both propositions accurate.
2. Relation exact.
3. Evidence from correct passages.
4. Synthesis within scope.

89. Checking order for Application

1. Writer idea accurate.
2. Local evidence relevant.
3. Mechanism comparison explicit.
4. Judgement present.
5. Language controlled.

90. Checking order for the whole paper

Protect high-value semantic risks before cosmetic proofreading. A wrong source attribution or modal shift usually matters more than replacing a repeated adjective.

91. Build a Paper 2 transfer ledger

For each active target, record:

• date;
• original paper;
• error family;
• repair instruction;
• changed task;
• support level;
• timed or untimed;
• success or failure;
• next condition to test.

92. The transfer ledger should stay small

Do not create administrative burden. The ledger exists to answer one question: has this repair survived enough variation to move out of active training?

93. Close transfer

Same operation, similar text type, changed content.

94. Far transfer

Same decision, different theme, different task or different surface form.

95. Cross-task transfer

The same semantic discipline appears in several Paper 2 components.

96. Timed transfer

The decision remains available as the clock creates competition.

97. Full-paper transfer

The decision survives the whole 90-minute architecture without causing regression elsewhere.

98. Retirement standard

Move an active target to maintenance when it repeatedly survives new passages, delay, reduced support and representative time.

99. Maintenance standard

Sample the skill occasionally. Do not dedicate full training blocks unless failure returns.

100. Reopening standard

If a retired skill reappears only in one condition, reopen at that boundary. Do not erase evidence of previous learning.

101. The deliberate-practice checkpoint

A Paper 2 feedback target is ready for full-programme integration when the learner can explain the original error, independently repair it, perform the better decision on changed passages and retain it with reduced prompts. The remaining question is whether it stays stable over weeks and under full-paper load.

102. A twelve-week Paper 2 feedback-to-transfer programme

This programme is a teaching architecture rather than an official examination schedule. Its purpose is to turn marked-script evidence into stable reading and response decisions while preserving the integrated Paper 2 system.

Weeks 1–2 — Build the error map

Collect several recent Paper 2 scripts where possible: one full timed paper, one less pressured task and one paper that felt unusually strong or weak. For each, record section timing, support conditions and the exact questions that produced errors.

Do not begin by averaging total marks. Identify repeated error families and whether they are local or cross-task. Select one active target, one maintenance target and one preserved strength.

Weeks 3–4 — Repair below full-paper load

Use targeted tasks. Operation errors receive mixed-question classification. Semantic errors receive short paraphrase and inference work. Summary errors receive point-selection drills. Connection errors receive proposition alignment. Application errors receive passage-to-context mapping.

Do not require another 90-minute paper merely to practise one decision.

Weeks 5–6 — Vary passage and theme

Change topic, prose style, writer stance and task wording. Remove model answers. The learner should begin to recognise the underlying decision without being told that today is “a modal-preservation exercise” or “a qualification exercise”.

Weeks 7–8 — Add delay and cross-task transfer

Retest after several days. Use the same semantic control across different Paper 2 components. A learner repairing writer representation in Passage 1 should later preserve it in Connections and Application.

Weeks 9–10 — Add timing and section competition

Use timed component sets, then linked sections. Measure whether a repair remains accurate when reading, writing and time are competing. If the skill fails only under pressure, the target becomes execution rather than conceptual understanding.

Weeks 11–12 — Full-paper stabilisation

Run representative 90-minute Paper 2 practice. Do not announce the target before the paper. Afterwards, check whether the repaired decision survived, whether another skill regressed and whether timing remained balanced.

Move stable targets to maintenance and select the next bottleneck only when justified by evidence.

103. A six-week accelerated programme

Week 1 — Diagnose

Find the earliest repeated decision causing the largest cross-task loss.

Week 2 — Isolate

Practise the decision in reduced tasks.

Week 3 — Vary

Change passages and themes.

Week 4 — Delay

Retest after forgetting and remove visible cues.

Week 5 — Time

Place the decision under section pressure.

Week 6 — Integrate

Run a full Paper 2 and decide whether the target is active, maintenance or retired.

104. A four-week programme for semantic distortion

Week 1

Collect every quantifier, modal, causal, comparison and attribution shift from marked scripts.

Week 2

Use contrast sets and closed-text paraphrase.

Week 3

Transfer the same controls across Passage 1, Summary and Application.

Week 4

Test under timed Paper 2 conditions.

105. A four-week programme for Summary selection

Week 1

Target interpretation and idea-unit identification.

Week 2

Main point versus explanation and example.

Week 3

Overlap clustering and changed-target practice.

Week 4

Timed unseen summary with no visible checklist.

106. A four-week programme for Connections

Week 1

Faithful proposition statements.

Week 2

Agreement, contradiction, qualification, extension and context differences.

Week 3

Synthesis consequences and evidence selection.

Week 4

Unseen passage pairs under timed section conditions.

107. A four-week programme for Application

Week 1

Passage proposition fidelity.

Week 2

Local mechanism and evidence fit.

Week 3

Qualification and judgement across varied local contexts.

Week 4

Timed Application after earlier Paper 2 work so late-paper execution is tested.

108. A four-week programme for timing collapse

Week 1

Instrument a full paper with timestamps.

Week 2

Repair the first major time leak separately.

Week 3

Run linked-section practice with closure rules.

Week 4

Retest the whole paper and compare quality distribution across sections.

109. The 24-hour feedback cycle

Within one day of receiving feedback, the learner should be able to state:

1. the first weak decision;
2. why it mattered;
3. what the repair is;
4. one immediate reconstructed example;
5. what the next unseen retest will change.

110. The 72-hour cycle

Use changed content and reduced prompting. The learner should reconstruct the decision from understanding rather than memory of teacher wording.

111. The one-week cycle

Use a different unseen passage or task. Remove the old correction. Test whether the decision appears naturally.

112. The one-month cycle

Sample retired and maintenance targets. Durable skills should survive without constant foreground attention.

113. The final-month rule: reduce novelty

Near the examination, feedback should concentrate on recurrent risks and stable repair methods. Do not introduce a new annotation architecture or complicated answer system unless the existing method has a demonstrated failure and enough time remains to practise the replacement.

114. Final month: shrink active targets

A learner should not enter the final weeks carrying fifteen correction codes. Consolidate to a small operating card that addresses the current limiting risks.

115. Fourteen days before the examination

Prioritise:

• unseen passage mapping;
• semantic preservation;
• Summary selection;
• Connections relation accuracy;
• Application passage anchoring;
• full-paper timing;
• one or two active feedback cues;
• sleep and representative practice conditions.

116. Seven days before the examination

Use shorter maintenance tasks and only purposeful full simulations. Review the operating card. Avoid mass correction of old scripts if the relevant behaviour has already been repaired and tested.

117. The day before

Feedback should already be internal. The learner should not be trying to memorise a thick error log. Review stable cues lightly and protect cognitive condition.

118. Tutor Protocol 1 — Keep question, passage and answer together

Never diagnose a Paper 2 response from the learner’s answer alone. Meaning depends on the exact source and command.

119. Tutor Protocol 2 — Ask the learner what they thought the operation was

This reveals whether the failure began before evidence search.

120. Tutor Protocol 3 — Separate local error from system error

A Summary clustering error needs local repair. Repeated modal distortion across three sections needs broader semantic training.

121. Tutor Protocol 4 — Use the earliest-cause rule

If the wrong source line was selected, polishing the answer wording is downstream. Repair locating first.

122. Tutor Protocol 5 — Convert comments into observable decisions

“Be precise” should become something like “preserve the writer’s quantity and certainty when paraphrasing”.

123. Tutor Protocol 6 — Require learner reconstruction

Model if needed, then close the model and return production to the learner.

124. Tutor Protocol 7 — Change the passage

Do not accept a repaired old answer as proof of learning. Use new material quickly enough that the decision must be reconstructed.

125. Tutor Protocol 8 — Track support conditions

Was the passage discussed before the task? Were summary points identified by the teacher? Were local examples suggested? Supported success is valuable but should not be logged as independent performance.

126. Tutor Protocol 9 — Fade cues intentionally

Plan when the feedback label disappears. A cue that never fades can become a hidden dependency.

127. Tutor Protocol 10 — Preserve one strength

When repairing speed, protect accuracy. When repairing confidence, protect calibration. When repairing concision, protect completeness.

128. Tutor Protocol 11 — Retest under the condition that matters

If the error appears only late in the paper, an untimed morning drill is not the final verification. Retest where the failure actually occurs.

129. Tutor Protocol 12 — Retire solved targets

Stop correcting a behaviour as if it were still active when recent evidence shows it has stabilised.

130. Student Protocol 1 — Reconstruct the task before reading the mark

State what the question asked and what passage evidence mattered. This turns feedback into an analysis of decisions rather than a reaction to score.

131. Student Protocol 2 — Identify the first answer you would change

Do this before reading the full teacher explanation. Compare self-diagnosis with external feedback.

132. Student Protocol 3 — Explain what changed in meaning

For paraphrase or inference errors, do not merely say “wrong”. State whether scope, certainty, causation, comparison, attribution or relation changed.

133. Student Protocol 4 — Repair something immediately

Produce one corrected answer, point set, relation statement or Application paragraph.

134. Student Protocol 5 — Schedule the unseen retest

The repair is not complete until a different passage tests it.

135. Student Protocol 6 — Predict feedback before marking

Estimate which section or answer is most vulnerable and why. This strengthens monitoring.

136. Student Protocol 7 — Keep only one primary active cue

Paper 2 already contains many task switches. Do not overload working memory with a long correction list.

137. Student Protocol 8 — Remove the cue when stable

A final examination will not provide personalised reminders. Independence requires fading.

138. Student Protocol 9 — Track when the skill fails

Does it fail on unfamiliar themes? Under time? In Application only? Late in the paper? The condition tells you what to train next.

139. Student Protocol 10 — Stop revisiting solved errors

Maintenance is lighter than active repair. Move attention to the present bottleneck.

140. Parent Protocol 1 — Ask what will change on the next unseen paper

Move the conversation from “What mark did you get?” to “Which decision are you repairing, and how will you know it transferred?”

141. Parent Protocol 2 — Do not demand another full paper automatically

The correct repair may be a fifteen-minute Summary selection set or a short Connections drill.

142. Parent Protocol 3 — Ask the learner to explain one correction

If the student can explain why the old answer distorted meaning and how the repair transfers, understanding becomes visible.

143. Parent Protocol 4 — Respect learning versus testing

Supported reconstruction is learning. Unseen timed work is testing. Both matter; they answer different questions.

144. Parent Protocol 5 — Watch for feedback debt

If the learner keeps completing new papers while the same comments recur, more simulation may not be the answer.

145. Readiness Test 1 — Operation recognition

Can the learner identify what a new question requires without relying on memorised command words?

146. Readiness Test 2 — Semantic fidelity

Can the learner preserve quantity, certainty, cause, comparison and attribution across new passages?

147. Readiness Test 3 — Inference boundary

Can the learner distinguish strong inference from speculation and state the answer at the appropriate force?

148. Readiness Test 4 — Summary selection

Can the learner build the correct point set for an unfamiliar target before drafting?

149. Readiness Test 5 — Summary compression

Can points be clustered and paraphrased without flattening distinctions?

150. Readiness Test 6 — Connections

Can the learner represent both writers accurately and state the exact relation on a new passage pair?

151. Readiness Test 7 — Application anchoring

Can the learner begin from the writer’s actual proposition rather than local knowledge?

152. Readiness Test 8 — Application context fit

Can the learner select local evidence because the mechanism fits rather than because the example is familiar?

153. Readiness Test 9 — Application judgement

Can the learner state how strongly the writer’s idea applies and which condition most affects the judgement?

154. Readiness Test 10 — Language control

Can the learner keep Summary concise and Application analytical without semantic distortion under time?

155. Readiness Test 11 — Timing

Does the paper reach every major task with enough time for complete reasoning?

156. Readiness Test 12 — Recovery

Can the learner defer one stubborn item and return without allowing it to damage the rest of the paper?

157. Readiness Test 13 — Self-detection

Can the learner identify an overclaim, missed qualifier, duplicate summary point or false connection before seeing teacher feedback?

158. Readiness Test 14 — Delay

Does the repaired behaviour remain after the original correction is no longer fresh?

159. Readiness Test 15 — Full-paper transfer

Does the repair survive a new 90-minute paper with unfamiliar passages and no visible prompt?

160. Retirement rule

Retire a target from active training when it passes variation, delay, independence and representative-load tests to a reasonable degree.

161. Maintenance rule

Sample retired skills periodically without allowing them to dominate training.

162. Reopening rule

If the skill fails only under a specific condition, reopen at that condition. Previous learning still counts.

163. The end goal is self-regulation

A mature Paper 2 learner increasingly performs the feedback cycle internally: notices uncertainty, checks the source, calibrates the answer, protects the clock, and after the paper can identify the next repair before external marking arrives.

164. Programme checkpoint

When marked Paper 2 scripts consistently produce accurate self-diagnosis, efficient targeted repair and successful unseen retests, feedback has become a learning system rather than a correction event.

165. Frequently asked question: Should I redo the whole Paper 2 after feedback?

Not automatically. Redo the smallest task that tests the active weakness. If the weakness is Summary selection, a new summary target may provide better evidence than another 90-minute paper. Use a full paper when integration and timing are the actual questions.

166. Frequently asked question: Should I correct every wrong answer?

Understand every important error, but do not make every local mistake an active training target. Prioritise repeated, severe or high-transfer failures.

167. Frequently asked question: What is the best first question after receiving a marked paper?

Ask: “What is the first decision that went wrong, and where else could that same decision matter?”

168. Frequently asked question: How do I know whether a mistake is local or systemic?

Look across tasks. If the learner misclusters one Summary point but preserves meaning everywhere else, the error may be local. If qualifiers disappear in Passage 1, Summary, Connections and Application, the weakness is systemic.

169. Frequently asked question: Why do I understand the correction but repeat it later?

Understanding can exist without retrieval under pressure. Use changed-passage practice, delay, cue fading and timed transfer until the better decision becomes easier to access.

170. Frequently asked question: Is “read more carefully” useful feedback?

Only if it is translated into a specific decision. For example: “Mark quantifiers before paraphrasing” or “identify the operation before searching for evidence”.

171. Frequently asked question: Should I memorise model answers?

No. Extract the decision that made the model better: accurate proposition, correct relation, efficient clustering, explicit judgement. Then reproduce that decision on different content.

172. Frequently asked question: Should I compare my answer word-for-word with the model?

No. Compare meaning, selected points, relation, scope and completeness. Several strong answers can use different wording.

173. Frequently asked question: What if the model uses a word I would never think of?

Ask whether the word is necessary for precision. If your own wording is accurate and natural, it may already be sufficient. Vocabulary should serve meaning.

174. Frequently asked question: How many feedback targets should I carry into a mock?

Usually one primary active target and perhaps one maintenance cue. Paper 2 already demands several task switches; a long correction checklist can overload attention.

175. Frequently asked question: How do I stop overthinking Passage 1 answers?

Use a closure rule. Once the operation is satisfied, evidence is sufficient and meaning is clear, move. Rewriting a correct short answer repeatedly can create new errors.

176. Frequently asked question: How do I know if my inference is too strong?

Ask whether the passage proves the intensity, motive or certainty you wrote. If another plausible explanation remains, the wording may need calibration.

177. Frequently asked question: How do I know if my inference is too weak?

If the text strongly points to one implication and you hedge it into vague possibility, you may be underselling the evidence. Match confidence to support.

178. Frequently asked question: Why do I keep losing “own words” marks?

The problem may not be vocabulary. You may be preserving source syntax, forcing unsafe synonyms or changing meaning. Reconstruct the proposition first, then verify scope, strength and relation.

179. Frequently asked question: Should I replace every word from the passage?

No. Technical terms and necessary concepts can remain where replacement would reduce accuracy. Independence of expression is not a game of lexical avoidance.

180. Frequently asked question: Why does my Summary look good but score poorly?

The problem may be upstream selection. Elegant paraphrase cannot rescue the wrong point set. Review target, idea units and clustering before sentence style.

181. Frequently asked question: Why do I miss Summary points I understood?

You may have misclassified them as examples, merged them with another point or dropped them under word pressure. Diagnose the exact reason.

182. Frequently asked question: Why do I repeat points in Summary?

The source may restate one idea through several examples or explanations. Cluster overlap before drafting.

183. Frequently asked question: How do I improve Connections?

Represent both propositions faithfully, name the comparison dimension, classify the relation and only then choose evidence. Do not begin with repeated keywords.

184. Frequently asked question: Is disagreement always more important than agreement?

No. Writers may qualify, extend or narrow each other. Strong comparison identifies the exact relation rather than hunting for opposition.

185. Frequently asked question: How do I know if I am synthesising?

Your sentence should explain what becomes clearer when the two passages are considered together. “A says X; B says Y” is juxtaposition, not synthesis.

186. Frequently asked question: Why does my Application become a mini Paper 1 essay?

You may be starting from local knowledge rather than the passage. Begin with the writer’s idea, then test it against context.

187. Frequently asked question: Can I disagree with the passage in Application?

Yes, if local evidence and conditions justify the judgement. Application is not compulsory agreement.

188. Frequently asked question: How much local knowledge do I need?

Enough to test important passage ideas with specific, accurate examples across several mechanisms. Quality, fit and explanation matter more than a huge memorised catalogue.

189. Frequently asked question: What if I cannot remember the perfect local example?

Use another verified example, narrow the claim or reason from a well-understood institutional pattern. Do not fabricate detail.

190. Frequently asked question: Why does Application language get worse?

It may begin late, after cognitive fatigue and time debt. Compare early and late writing before assuming the problem is a general language weakness.

191. Frequently asked question: How do I improve Paper 2 timing?

Use timestamps to locate the first major leak. The issue may be rereading, overannotation, paraphrase perfectionism, slow Summary selection or Connections re-entry—not general reading speed.

192. Frequently asked question: Should I set rigid section times?

Use tested ranges and checkpoints rather than assuming one universal split fits every learner. The essential goal is that all major tasks receive sufficient time.

193. Frequently asked question: What if one question is taking too long?

Use bounded effort, record your best current thinking, mark it for return and continue. One low-value item should not consume the Application section.

194. Frequently asked question: Should I leave a blank answer?

Where possible, make the strongest evidence-based attempt before deferring. Avoid random guessing or speculative writing.

195. Frequently asked question: How should I check Paper 2?

Check content-semantic risks first: source, operation, quantity, claim strength, Summary coverage, Connections relation and Application anchoring. Then check language.

196. Frequently asked question: Should I change an answer if it sounds awkward?

Not without a reason. Change when you identify a source mismatch, logic problem, grammar issue or clarity failure. Unnecessary rewriting can introduce semantic drift.

197. Frequently asked question: How do I use teacher comments from several papers?

Group them by mechanism rather than by date. “Too strong”, “not in passage” and “overgeneralised” may belong to one semantic-control target.

198. Frequently asked question: What should go in a Paper 2 error log?

Keep it compact: task, first weak decision, repair rule, unseen retest, support condition and current status. Retire solved targets.

199. Frequently asked question: Should I keep every old error in the log?

No. An ever-growing list becomes noise. Active errors deserve attention; stable skills move to maintenance.

200. Frequently asked question: Can one feedback target improve several sections?

Yes. Claim-strength control, source identity, operation recognition and answer closure can improve multiple components.

201. Frequently asked question: Can a total mark rise while a weakness remains?

Yes. Familiar theme, easy passage fit or strong local knowledge can compensate temporarily. Track the active target directly.

202. Frequently asked question: Can a total mark stay flat while real improvement occurs?

Yes. One repaired skill may improve while a new difficult passage exposes another weakness. Marks are outcomes; mechanisms explain them.

203. Frequently asked question: Should I practise only the section where I lost the most marks?

Not necessarily. The biggest numerical loss may be downstream from an earlier system error such as timing or semantic distortion.

204. Frequently asked question: Should I do more full papers if my score is not moving?

Only if full-paper integration is the unresolved problem. Repeated simulations with the same uncorrected mechanism can stabilise weak habits.

205. Frequently asked question: How do I know when to return to full papers?

When the targeted repair works on changed unseen material with reduced support and reasonable section timing.

206. Frequently asked question: When can I retire a feedback target?

When it survives variation, delay, reduced prompting and representative Paper 2 load across several observations.

207. Frequently asked question: What if a retired problem comes back?

Identify the condition in which it returned and reopen training there. Do not assume all earlier learning disappeared.

208. Frequently asked question: How should JC2 use this page?

Use it as the Paper 2-specific repair engine after marked school papers, prelims and mocks. Year-level revision allocation still belongs to the JC2 marked-scripts owner.

209. Frequently asked question: How should JC1 use this page?

JC1 learners can use individual feedback cycles, but their main routes remain the JC1 foundations for reading, Summary/Paraphrasing, Connections/Synthesis and Application.

210. Can AI mark Paper 2?

AI can help identify possible semantic shifts, compare paraphrases, generate original practice passages and suggest alternative diagnoses. It should not be treated as an official SEAB examiner, and generated marks should not be presented as authoritative scores.

211. Use AI for explanation, not invisible replacement

A productive workflow is:

1. student attempts independently;
2. student predicts risks;
3. teacher or trusted marking provides primary feedback;
4. AI can generate changed-passage retests or contrast sets;
5. student evaluates and verifies suggestions.

212. AI can generate transfer tasks

For example, after a modal-strength error, AI can create original sentences with different certainty levels. After a connection error, it can generate paired passages with qualification and extension. Factual content should be checked when accuracy matters.

213. AI can create dependency

If the tool rewrites every answer before the learner diagnoses the weakness, the student may become better at accepting corrections rather than generating decisions.

214. Ask AI to identify meaning changes

Useful questions include: “Did this paraphrase change certainty?”, “Did I broaden the population?”, “Does this comparison overstate disagreement?” The learner should still verify against the exact source.

215. Do not use AI-generated local facts without verification

Application examples should remain accurate. For current policies, statistics and events, return to trustworthy sources.

216. Why this article is 20,000+ words

This page is long because the reader job is large. Paper 2 feedback must connect four task systems without merging their ownership: diagnose Passage 1 operations, trace Summary point loss, distinguish Connections relation errors, evaluate Application fit, identify cross-task semantic weaknesses, locate time debt, design reduced practice, vary passages, delay retests, fade support, verify under section time and finally prove transfer under the complete 90-minute paper.

A short “how to learn from comprehension mistakes” article would not provide enough structure to move from a marked response to an independently improved unseen response. The length exists for complete diagnosis → repair → transfer → verification coverage, not for padding.

217. Canonical ownership boundary

SETC-JC-GP-P2-050 owns the Paper 2 feedback-to-transfer layer: marked Paper 2 evidence, first-weak-decision diagnosis, local-versus-system error classification, repair priority, targeted reconstruction, changed-passage retest, cross-task transfer, delayed retest, timing transfer, support fading, maintenance and retirement.

The parent H1 General Paper Paper 2 Examination Performance page remains the integrated whole-paper owner.

218. Passage 1 ownership boundary

H1 General Paper Paper 2 Passage 1 Performance remains the canonical owner for comprehension, inference, analysis, evaluation, language effect and synthesis inside Passage 1.

219. Summary ownership boundary

H1 General Paper Paper 2 Summary Performance remains the canonical owner for selecting, clustering, compressing and preserving meaning under examination conditions.

220. Connections ownership boundary

H1 General Paper Paper 2 Connections Performance remains the canonical owner for cross-passage comparison and synthesis execution.

221. Application ownership boundary

H1 General Paper Paper 2 Application Performance remains the canonical owner for using passage ideas, contextual evidence and judgement.

222. Wider feedback ownership boundary

English Exam Feedback Transfer remains the exam-neutral feedback owner, while JC 2 General Paper | How to Use Feedback and Marked Scripts remains the year-level revision and marked-script planning owner.

223. Public integrity boundary

This page does not reproduce confidential examination material, claim access to private marking procedures or present SETC teaching labels as official SEAB terminology. Its examples, diagnostic families and programmes are original instructional tools. Official format facts come from published SEAB documents.

224. Official source: H1 General Paper 8881 for 2027

For the current 2027 syllabus, use the SEAB 2027 GCE A-Level syllabus index and the Singapore–Cambridge H1 General Paper 8881 syllabus for examination in 2027.

225. Research-informed teaching note

Broader writing and learning evidence supports explicit strategy instruction, modelling, guided practice, feedback, reflection and transfer. One useful public reference is the Institute of Education Sciences practice guide, Teaching Secondary Students to Write Effectively. That evidence base does not validate every SETC diagnostic label, interval or programme length here; those are teaching designs to be judged against learner performance.

226. Final learner checklist

• I can reconstruct the exact task before reacting to the mark.
• I can identify whether the failure began at operation, location, meaning, relation, answer construction or timing.
• I can distinguish local component errors from cross-task system errors.
• I can explain what meaning changed in an inaccurate paraphrase.
• I can rebuild a Summary point set before rewriting prose.
• I can state both passage propositions before comparing them.
• I can anchor Application in the writer before using local knowledge.
• I can identify where time debt first appears.
• I can repair the original answer without copying a model.
• I can perform the repair on a changed passage.
• I can retain the repair after delay.
• I can use it with reduced support.
• I can use it under section time.
• I can use it in a full Paper 2.
• I know when to move the target to maintenance.

227. Final tutor checklist

• Do I have the question, source and learner answer together?
• Have I identified the earliest high-leverage failure?
• Is the weakness local or cross-task?
• Have I converted the comment into an observable decision?
• Has the learner generated a repair?
• Have I changed the passage for transfer?
• Have I recorded support conditions?
• Have I retested after delay?
• Have I added time only when appropriate?
• Have I protected one existing strength?
• Have I faded cues?
• Have I verified the repair in a full paper before retiring it?

228. Final parent checklist

• Can the student explain what the question was really asking?
• Can they explain why the answer lost control?
• Can they state what will change on the next unseen passage?
• Are they doing targeted repair instead of endless full papers?
• Are repeated comments actually becoming less frequent?

229. Continue through the Paper 2 route

Return to H1 General Paper Paper 2 Examination Performance for integrated paper execution. Use the four specialist child owners for Passage 1, Summary, Connections and Application when the diagnosis points to a task-specific mechanism.

230. Closing principle: the old paper diagnoses; the new paper proves

A marked comprehension script can tell you where the system failed. It cannot prove that the system has changed. That proof comes later, when a different writer, a different passage and a different question trigger a better decision without the old correction being visible.

That is the purpose of Paper 2 feedback: not to make yesterday’s answers prettier, but to make tomorrow’s unseen reading more reliable.

231. Advanced Transfer Bank | Sixty Paper 2 Feedback Decisions

The cases below extend the feedback system into situations where the answer is not simply right or wrong. They test diagnosis, prioritisation, support fading and transfer. For each case, the learner or tutor should identify the first weak decision, the smallest useful repair and the next unseen condition that would prove improvement.

Transfer 1 — The answer earns the mark but the reasoning is fragile

A learner receives credit for an inference because the final wording happens to match the expected idea, but their explanation reveals that they guessed from topic knowledge rather than the passage. The mark is positive evidence about outcome, not necessarily about method. The repair is to reconstruct the clue-to-inference chain. The transfer test uses an unfamiliar topic where background knowledge cannot produce the same lucky match.

Transfer 2 — The answer loses the mark despite sound reasoning

The learner identifies the correct inference orally but writes a vague pronoun such as “this” without naming what is implied. The problem is answer construction rather than comprehension. The repair is to express the reasoning explicitly in one complete sentence. Transfer should use a new inference item where the conceptual reasoning is similar but the written form must be independently generated.

Transfer 3 — Two wrong answers share one source-reading error

A learner misreads one crucial qualifier in Passage 1. That same misunderstanding causes a direct comprehension loss and later a false Connection with Passage 3. Treating these as separate errors would duplicate practice. The repair is to update passage identity and qualifier tracking. Transfer should use a passage where stance changes mid-way so the learner must refresh the global model rather than rely on first impressions.

Transfer 4 — Passage map is accurate but local locating fails

The learner can summarise the writer’s overall position but repeatedly chooses the wrong sentence for short-answer evidence. This is not a global comprehension failure. It is a proposition-location problem. The repair is conceptual question paraphrase followed by targeted line search. Transfer should include repeated keywords in several paragraphs so lexical matching becomes unreliable.

Transfer 5 — Local locating is strong but passage identity is weak

The learner finds individual answers efficiently but later confuses which passage holds which overall stance. Connections and Application suffer. The repair is one-line passage identity after first reading. Transfer should use three passages on the same theme with overlapping vocabulary but different causal models.

Transfer 6 — The learner always chooses the strongest interpretation

Whenever evidence permits several levels of force, the learner prefers the most dramatic. Concern becomes outrage, possibility becomes certainty and partial disagreement becomes opposition. The repair is claim-strength calibration across tasks. Transfer should deliberately mix strong, medium and weak evidence so the correct answer is not always the cautious one.

Transfer 7 — The learner always chooses the safest interpretation

After being warned about overclaiming, the learner writes “may”, “perhaps” and “possibly” even when evidence is strong. The repair is not more caution but better calibration. Transfer requires distinguishing direct textual evidence from genuine uncertainty. The goal is accurate strength, not permanent hedging.

Transfer 8 — The learner knows a technique list but cannot analyse prose without one

When a phrase is clearly metaphorical, the student performs well. When the effect comes from clause order, contrast or repetition, the answer collapses. The repair is meaning-before-label analysis. Transfer should include passages where no famous technique term is available and the learner must explain how the actual wording shapes meaning.

Transfer 9 — The learner explains effect but forgets the writer’s purpose

The response correctly says a phrase makes a situation seem urgent but does not connect that urgency to why the writer uses it in the argument. The repair is to add function: what does creating urgency allow the writer to claim, prioritise or persuade? Transfer should use a different rhetorical purpose such as scepticism, concession or scale.

Transfer 10 — The learner identifies purpose but not language evidence

The student writes that a rhetorical question challenges the reader but does not cite or explain the wording that creates the challenge. The repair is wording → meaning → function. Transfer should use a different device so the learner cannot rely on one memorised sentence frame.

Transfer 11 — Summary point selection is strong, paraphrase is weak

The learner consistently chooses the right ideas but changes cause and certainty while rewriting. The active target is semantic reconstruction, not selection. The repair uses closed-text paraphrase and meaning checks. The transfer test keeps the selection simple while changing the linguistic relationships that must survive.

Transfer 12 — Summary paraphrase is strong, selection is weak

The learner writes elegant, independent sentences about the wrong ideas. The correction should not focus on language. The repair is target filtering and idea-unit selection. Transfer should use the same source passage with a different summary target so the learner sees that selection depends on the question.

Transfer 13 — Summary point set is correct but word budget collapses

The learner identifies all relevant points yet cannot fit them efficiently. The problem is compression and overlap management. The repair is to identify duplicated wording, combine compatible ideas and remove illustration. Transfer should use a new passage with repeated explanations of the same mechanism.

Transfer 14 — Summary is concise because distinct ideas were erased

The response sounds efficient but merges separate mechanisms such as cost, trust and access into “barriers”. The repair is to ask which distinctions the target needs preserved. Transfer should include several semantically related but separately important points so overcompression becomes visible.

Transfer 15 — Summary improves untimed but not in the full paper

Selection and paraphrase are stable in isolated exercises, but under full Paper 2 the learner reverts to copying and drops points. The skill exists but is not automated under competing demands. The repair is timed section work and then linked sections before another full paper.

Transfer 16 — Connections are accurate but too slow

The learner can compare passages precisely but rereads both entire texts for every item. The repair is compact passage identity and proposition mapping. Transfer should use multiple connection questions from the same passages so the learner can benefit from earlier mapping instead of starting from zero each time.

Transfer 17 — Connections are fast but superficial

The learner uses repeated words to write immediate comparisons. Speed hides relation errors. The repair is proposition alignment before relation labels. Transfer should include false lexical matches and hidden conceptual agreement.

Transfer 18 — Connection relation is correct but evidence mismatched

The learner says the writers qualify one another accurately but cites a line that belongs to a different sub-claim. The problem is evidence alignment, not relation understanding. The transfer test keeps the same relation type but requires evidence from separated parts of the passages.

Transfer 19 — Connection evidence is correct but synthesis is absent

The response presents both passages accurately yet never states what the relation implies. The repair is synthesis consequence: what becomes clearer when both propositions are considered together? Transfer should require a final integrated sentence after every connection.

Transfer 20 — Application knowledge is excellent but passage anchoring is weak

The learner knows many local examples and writes fluently, but the passage idea appears only in the opening line. The repair is to make the writer’s mechanism the organising lens throughout the paragraph. Transfer should use the same local example against a different passage claim to show how fit changes.

Transfer 21 — Passage anchoring is strong but local evidence is generic

The student represents the writer carefully but supports the judgement with broad claims such as “Singapore values education”. The repair is evidence specificity: policy, institution, observable pattern or demographic condition. Transfer should require one concrete example and one explanation of mechanism.

Transfer 22 — Local evidence is specific but outdated for the claim

The learner uses an old policy or social pattern to make a current factual conclusion. The issue is evidence freshness, not Application structure. The repair is to update the contextual bank and separate historical mechanism from current condition. Transfer should use a current claim requiring fresh evidence.

Transfer 23 — Old evidence remains useful but the learner discards it

A historical case still demonstrates a stable institutional mechanism. The learner assumes only recent examples count. The repair is to distinguish recency from relevance. Transfer should compare when age weakens evidence and when the mechanism remains informative.

Transfer 24 — One Application paragraph uses too many examples

The student lists three or four local cases and runs out of space for comparison and judgement. The repair is evidence economy: choose the case that best reveals the mechanism, then analyse it. Transfer should cap examples and require explicit evaluation.

Transfer 25 — Application example is vivid but atypical

A dramatic local incident is used to represent society broadly. The repair is to distinguish severity from prevalence and illustration from representativeness. Transfer should require either a narrowed claim or wider supporting evidence.

Transfer 26 — Application judgement is cautious because evidence conflicts

The learner sees both supporting and contradicting local evidence and concludes “it depends”. The repair is hierarchy: identify which condition is more decisive and for whom. Transfer should use another mixed-evidence context so the learner must make a measured but real judgement.

Transfer 27 — Application judgement is strong because evidence is one-sided

Here strong language may be justified. The feedback should not automatically force qualification. The learner must understand that calibration can produce decisive wording where support is consistent. Transfer should include a case where strong and weak evidence alternate.

Transfer 28 — Early Paper 2 accuracy produces overconfidence

The learner finds Passage 1 easy and accelerates through Summary and Connections, making avoidable errors. The repair is task-specific reset: every new operation begins from its own command and source. Transfer should use an easy opening section followed by subtle later tasks.

Transfer 29 — Early Paper 2 difficulty produces defeat

One hard inference causes the learner to doubt the entire paper and rush later tasks. The repair is reset and bounded effort. Transfer should deliberately place a difficult low-value item early so resilience can be practised without surprise.

Transfer 30 — The learner finishes too early with thin Application

Completion time looks impressive, but later paragraphs lack contextual development. The repair is not to slow every section. Reallocate saved time to high-value evaluation and final checking. Transfer should compare quality distribution rather than raw completion time.

Transfer 31 — The learner finishes late because of repeated rereading

Every question triggers a full passage reread. The repair is passage structure memory and conceptual locating. Transfer should track how often entire rereads are replaced by targeted re-entry.

Transfer 32 — The learner finishes late because of paraphrase editing

Answers are repeatedly rewritten for elegance. The repair is decision closure and reason-before-change. Transfer should use a timed set where the learner can revise only when a specific semantic or grammatical reason is named.

Transfer 33 — The learner finishes late because Summary point selection is slow

Expression is not the bottleneck. The learner hesitates over relevance. The repair is target filtering and point classification without prose. Transfer should time skeleton construction separately.

Transfer 34 — The learner finishes late because Connections re-entry is inefficient

Each comparison restarts passage reading. The repair is proposition map reuse. Transfer should include several connection questions from the same three texts.

Transfer 35 — The learner finishes late because Application planning becomes a second essay plan

Too many local examples and subpoints are generated before writing. The repair is one writer idea, one mechanism, one or two examples, one judgement per paragraph. Transfer should limit planning space.

Transfer 36 — The learner checks only grammar

Source attribution and claim strength errors remain. The repair is a content-semantic checking order before surface language. Transfer should include planted semantic errors in otherwise grammatical answers.

Transfer 37 — The learner checks only content

Summary and Application meaning are accurate, but grammar and reference reduce clarity. The repair is a second-pass language check targeted to known patterns. Transfer should distinguish meaning-first and surface-accuracy passes.

Transfer 38 — The learner changes correct answers during review

Anxiety, not evidence, drives revision. The repair is a reason-before-change rule. Transfer should record whether each change had a source, logic or grammar justification and whether it improved the answer.

Transfer 39 — The learner refuses to change any answer

Overconfidence prevents correction even when a source mismatch is found. The repair is evidence-based review: change when a specific problem is demonstrated. Transfer should include one answer that genuinely needs revision and one that does not.

Transfer 40 — The learner depends on teacher relation labels

Connections are strong only when the tutor says “this is qualification” or “look for extension”. The repair is cue fading. Transfer should remove labels and ask the learner to explain the relationship in plain language before naming it.

Transfer 41 — The learner depends on teacher summary-point selection

Writing looks strong because the hardest decision was supplied externally. The repair is to move support earlier: the learner selects points, then teacher verifies. Transfer ends when verification is no longer needed.

Transfer 42 — The learner depends on suggested local examples

Application appears strong in tuition but weakens independently. The repair is mechanism-based retrieval. Transfer should remove example suggestions while keeping the passage claim clear.

Transfer 43 — The learner depends on model paraphrases

They recognise good wording but cannot generate it. The repair is closed-text reconstruction before model comparison. Transfer should use a new sentence not represented in the model bank.

Transfer 44 — The learner can correct others but not self

They identify semantic drift in sample answers but miss it in their own writing. The repair is delayed self-review before teacher marking. Transfer should measure whether self-detection rises across several scripts.

Transfer 45 — The learner can detect their own error after writing but not prevent it

The repair has reached monitoring but not generation. Move the cue earlier: before paraphrase, before Summary drafting, or before Application evidence. Transfer should track first-draft error frequency rather than only final corrected quality.

Transfer 46 — The learner prevents the error untimed but not timed

The conceptual repair is stable, but automaticity is incomplete. Use progressively timed sections while holding accuracy constant. Transfer ends with the full-paper condition in which the weakness previously returned.

Transfer 47 — The learner’s active target is already solved

Tutor and learner keep talking about the same old problem because the label has become habitual. Recent scripts show stable control. The repair is actually to retire the target and reallocate attention. Transfer evidence should update the learner model.

Transfer 48 — A new weakness appears after an old one is solved

Once semantic control improves, timing or answer economy becomes the new bottleneck. This is not regression. It is the next limiting factor becoming visible. The training plan should advance rather than reopen solved issues unnecessarily.

Transfer 49 — One low mark triggers panic revision

The learner changes annotation, Summary method and Application structure after one difficult paper. The repair is diagnostic restraint. Compare task difficulty, theme familiarity, timing and recurrent patterns before redesigning the system.

Transfer 50 — One high mark triggers premature confidence

A familiar theme and favourable passages produce a strong result. The repair is verification on unfamiliar material. Stability requires several observations across variation.

Transfer 51 — Two teachers give different feedback

One emphasises concision; another wants fuller explanation. Return to the exact question, passage and response. The issue may be local task demand rather than contradiction. Distinguish official assessment needs from stylistic preferences.

Transfer 52 — Feedback comment is vague but the script is informative

The teacher writes only “weak” or “develop”. Reconstruct the missing job: source evidence, inferential bridge, relation, local mechanism or judgement. Script evidence can often produce a more operational repair than the brief annotation.

Transfer 53 — Feedback comment is detailed but too local

The teacher explains exactly why one answer is wrong but does not state the general principle. The learner can correct that line but not transfer. Extract the portable decision and test it on changed material.

Transfer 54 — Feedback volume creates overload

Twenty annotations appear on one paper. The learner tries to remember all twenty during the next mock and performance worsens. Group local notes into families, select one primary active target and keep the rest as reference or maintenance.

Transfer 55 — Feedback focuses only on losses

The learner stops trusting stable strengths. Add one preserved behaviour: accurate passage mapping, efficient Summary skeleton, strong local example selection or reliable time closure. Repair should not dismantle functioning parts of the system.

Transfer 56 — Feedback focuses only on encouragement

The learner leaves knowing the paper was “better” but not what decision improved or what still limits performance. Convert praise into procedure and identify the next bottleneck. Positive feedback should be operational too.

Transfer 57 — AI produces a polished corrected answer

The student submits it as evidence of improvement. The artefact improved, but learner control is unknown. Ask the student to explain the change, close the AI output and reconstruct on a new passage. Transfer, not polish, is the proof.

Transfer 58 — AI flags a problem the teacher did not mention

Treat it as a hypothesis. Compare with the passage, question and marking logic. If valid and recurrent, it can become a secondary target. If not, discard it. AI feedback should be evidence-checked rather than automatically accepted.

Transfer 59 — AI and teacher disagree about a paraphrase

Return to meaning. Does quantity, certainty, causation, comparison and attribution remain intact? The deciding standard is source fidelity and task requirement, not authority of the commenter.

Transfer 60 — The learner reaches self-regulation

Before marking, the learner identifies a possible Summary duplication, explains a Connection relation they are unsure about, notices that Application began late and proposes the next repair. External feedback now confirms and refines an internal diagnostic system. That is the endpoint this owner is designed to create.

232. The advanced transfer standard

A mature feedback system can distinguish luck from control, local mistakes from shared mechanisms, supported success from independent success and conceptual understanding from timed execution. It can also retire old targets, accept genuine new weaknesses and use disagreement among markers as evidence to inspect rather than as a reason to panic.

233. Feedback should update the learner model

Every marked Paper 2 changes what we know about the learner. A good training system is therefore versioned. Stable strengths become assumptions that require only sampling. Active weaknesses receive concentrated repair. New conditions reopen old targets only when the evidence warrants it.

234. The learner model should remain falsifiable

“Weak at Summary” is too broad. A useful model might say: “Summary target recognition is stable; overlap clustering fails under time; paraphrase preserves meaning.” That model predicts what should happen in the next task and can be confirmed or disproved.

235. The final purpose of feedback

The mature learner should need less external explanation for routine errors because they can increasingly identify operation, evidence, semantic relation, answer scope and timing state independently. Feedback then becomes calibration for harder problems rather than basic error detection.

236. Marked Paper 2 Workshop | Thirty-Two Full Diagnostic Profiles

The profiles below show why a total Paper 2 mark should never be treated as a diagnosis. Each fictional profile combines section evidence, timing and answer behaviour. The purpose is to decide what should be repaired first and what should be preserved.

Profile 1 — Strong Passage 1, weak Summary, stable timing

The learner answers literal, inference and language questions accurately and finishes Passage 1 within a workable range. Summary content is weak because several examples are retained while underlying reasons are missed. Connections and Application are reasonable because enough time remains.

Diagnosis: local Summary selection/abstraction weakness, not whole-paper comprehension. Active repair: target → idea unit → point/example classification. Preserve: Passage 1 operation recognition and overall pacing. Transfer proof: two unseen summary targets with accurate point sets before prose, followed by one timed Summary inside a linked Paper 2 segment.

Profile 2 — Strong Summary, weak Passage 1 inference, strong Application

The learner selects and compresses Summary points accurately and writes a well-grounded Application. Several Passage 1 inference answers either repeat the clue or jump to a stronger emotional interpretation.

Diagnosis: inference calibration is local to Passage 1 rather than a general paraphrase problem. Active repair: clue → relationship → inference → boundary. Preserve: Summary semantic control and Application anchoring. Transfer proof: unfamiliar inference set with mixed evidence strength, then a timed Passage 1 section. Do not redirect the whole training plan toward Summary or local knowledge simply because the overall score is lower.

Profile 3 — Strong first half, weak Application, late start

Passage 1 and Summary are accurate, but the learner spends excessive time rereading connection evidence. Application begins with little time left and receives thin development and language errors.

Diagnosis: execution and connection re-entry, not Application knowledge. Active repair: compact proposition map and section closure. Preserve: early accuracy. Transfer proof: multiple connection items using one passage set with timestamped re-entry, then a full paper verifying that Application begins earlier without reducing connection accuracy.

Profile 4 — Weak across all sections through claim-strength distortion

Passage 1 turns “may” into “will”, Summary broadens “some” to “people”, Connections treat partial disagreement as contradiction and Application overstates a writer’s limited claim.

Diagnosis: cross-task semantic-strength system error. Active repair: modal, quantifier, scope and relation calibration across tasks. Preserve: whatever passage locating and local knowledge already function. Transfer proof: mixed-task semantic contrast sets followed by unseen passages. This learner does not need four separate correction programmes; one shared meaning-control repair has higher leverage.

Profile 5 — Weak total score but excellent self-diagnosis

The learner loses marks across Summary and Application but accurately predicts which answers are weak and can explain the causes after the paper. Timing is reasonable.

Diagnosis: monitoring is ahead of production. Active repair: move known corrections earlier into generation through short cues and repeated production. Preserve: self-awareness. Transfer proof: compare first-draft error frequency across several tasks. The goal is prevention, not more post-hoc explanation.

Profile 6 — Higher score, poor self-monitoring

The learner performs well on a familiar theme but cannot explain why the answers worked and misses several risky overclaims that happened not to be penalised heavily.

Diagnosis: outcome stronger than control. Active repair: self-explanation and far-transfer verification. Preserve: fluent execution. Transfer proof: unfamiliar theme with the same semantic demands. Do not retire weaknesses simply because the total mark is high.

Profile 7 — Strong content, low language marks

Summary points and Application evidence are generally correct, but Summary becomes telegraphic and Application sentences are long, loosely referenced and grammatically unstable.

Diagnosis: task-specific language construction. Active repair: concise complete Summary sentences and one-main-clause Application control. Preserve: content selection. Transfer proof: rewrite new content without changing selected ideas, then retest under time. Avoid turning this into a vocabulary escalation programme.

Profile 8 — Strong language, weak content

The paper reads fluently, but several answers address the wrong operation, Summary selects outside-scope points and Application applies a distorted passage claim.

Diagnosis: comprehension and task interpretation, not expression. Active repair: operation recognition and source fidelity. Preserve: clear prose. Transfer proof: mixed operations and target-switch Summary work before another full paper.

Profile 9 — Excellent untimed, weak timed

When working without a clock, the learner produces accurate inferences, complete Summary point sets and nuanced Application. Under 90 minutes, qualifiers disappear, Summary is shortened excessively and Application becomes generic.

Diagnosis: execution automaticity. Active repair: timed sections with protected accuracy and gradual load increase. Preserve: conceptual method. Transfer proof: same decisions under moderate, then full timing. Do not reteach the concepts from the beginning unless they also fail untimed.

Profile 10 — Weak untimed and timed

The learner misidentifies operations, guesses inference and cannot distinguish Summary points from examples even without pressure.

Diagnosis: foundational reading system, not timing. Active repair: reduced operation, evidence and idea-unit practice. Preserve: any local strengths such as vocabulary or passage interest. Transfer proof: stable untimed accuracy on varied short tasks before adding section timing.

Profile 11 — Good Summary score from copied wording

The learner captures many correct points but remains very close to the passage. One paper scores reasonably because wording happens to remain acceptable.

Diagnosis: fragile paraphrase independence. Active repair: closed-text conceptual reconstruction. Preserve: point selection. Transfer proof: new passage with semantic-strength verification. This is a preventive repair: the current mark may not reveal the future risk.

Profile 12 — Independent Summary wording but semantic drift

The learner clearly uses their own words but changes degree, causation and attribution.

Diagnosis: originality without fidelity. Active repair: meaning checksum after reconstruction. Preserve: willingness to restructure language. Transfer proof: changed sentences where accuracy, not lexical distance, is the scoring target.

Profile 13 — Connections weak because passage identities are weak

The learner can compare individual lines but often attributes Passage 1’s stance to Passage 3 or forgets a later qualification.

Diagnosis: global source model. Active repair: compact passage identity and stance-update tracking. Preserve: relation language. Transfer proof: three passages with similar topics but evolving positions.

Profile 14 — Passage identities strong, Connections still vague

The learner knows each writer’s stance but writes “both agree” or “they differ” without specifying the dimension.

Diagnosis: relation precision and synthesis architecture. Active repair: proposition alignment plus comparison dimension. Preserve: source identity. Transfer proof: unseen pairs requiring qualification, extension and different-level relations.

Profile 15 — Application knowledge bank is narrow

The learner anchors passages well but repeatedly uses the same two local examples even when the mechanism differs.

Diagnosis: context retrieval breadth. Active repair: mechanism-based local knowledge bank. Preserve: passage anchoring. Transfer proof: new passage themes requiring examples from different mechanisms such as trust, inequality, regulation and access.

Profile 16 — Knowledge bank is broad but examples are forced

The learner knows many policies and cases but chooses them by topic similarity rather than mechanism fit.

Diagnosis: evidence selection, not knowledge quantity. Active repair: state writer mechanism before example choice. Preserve: retrieval range. Transfer proof: three candidate examples for one passage idea with justification of best fit.

Profile 17 — Application is strong when agreeing, weak when disagreeing

The learner can show local support for a writer but becomes generic when the claim applies poorly.

Diagnosis: mismatch reasoning. Active repair: identify which local condition breaks transfer. Preserve: supporting evidence use. Transfer proof: passage claims deliberately chosen to fit weakly or partially.

Profile 18 — Application is strong when disagreeing, weak when agreeing

The learner enjoys finding exceptions and limitations but underdevelops straightforward support.

Diagnosis: evaluation bias toward critique. Active repair: explain why a mechanism genuinely transfers before searching for caveats. Transfer proof: strong-fit cases where decisive agreement is justified.

Profile 19 — Many local examples, no societal hierarchy

The learner lists different groups and policies but never decides which condition matters most.

Diagnosis: contextual judgement. Active repair: rank evidence by scale, representativeness or causal relevance. Preserve: breadth. Transfer proof: mixed local evidence requiring one final hierarchy.

Profile 20 — Passage 1 language-effect answers are formulaic

Every response says a device “emphasises the point” or “creates impact”.

Diagnosis: effect specificity. Active repair: wording → contextual meaning → precise impression/function. Preserve: ability to notice features. Transfer proof: prose where effect comes from diction, syntax and structure rather than obvious figurative language.

Profile 21 — Argument evaluation attacks weak versions

The learner notices limitations but exaggerates the writer’s claim before criticising it.

Diagnosis: representation before evaluation. Active repair: restate exact claim, then test evidence-bearing capacity. Preserve: critical instinct. Transfer proof: nuanced arguments where the writer has already qualified the conclusion.

Profile 22 — Evaluation is accurate but too slow

The learner maps claim, evidence and assumption carefully but spends disproportionate time on low-mark items.

Diagnosis: decision closure. Active repair: one main inferential weakness per answer. Preserve: accuracy. Transfer proof: timed evaluation items with a bounded response length.

Profile 23 — Passage reading is slow because every unknown word is decoded

Vocabulary interruptions destroy the passage map.

Diagnosis: unknown-word tolerance and contextual inference. Active repair: central/inferable/ignorable classification. Preserve: careful reading. Transfer proof: unfamiliar prose containing non-essential difficult vocabulary.

Profile 24 — Passage reading is fast but qualifiers are missed

The learner knows the main topic quickly but drops important limiting language.

Diagnosis: precision sacrificed for speed. Active repair: qualifier detection and stance-update pauses. Preserve: global reading speed. Transfer proof: passages whose final judgement differs materially from the opening claim.

Profile 25 — Overannotation causes time loss

Nearly every line is underlined and multiple symbols have no later use.

Diagnosis: annotation selectivity. Active repair: mark only claim, structural shift and high-value qualifier during training. Preserve: active reading. Transfer proof: compare question re-entry time under old and reduced annotation systems.

Profile 26 — No annotation causes repeated rereading

The learner refuses all marking but repeatedly loses passage position.

Diagnosis: insufficient external memory support. Active repair: one-line passage identities and minimal structural marks. Preserve: uncluttered reading. Transfer proof: fewer full rereads on a new paper.

Profile 27 — Full-paper score fluctuates with theme familiarity

The learner performs strongly on technology and education, weakly on arts or environment.

Diagnosis: possible dependence on background knowledge rather than pure reading control. Active repair: unfamiliar-theme Passage 1 and Connections work plus targeted concept vocabulary where necessary. Transfer proof: reduced performance gap across domains.

Profile 28 — Full-paper score fluctuates with prose density

Topic does not matter, but abstract argumentative prose causes errors.

Diagnosis: sentence parsing and argument mapping. Active repair: clause-level unpacking and claim–reason–evidence maps. Preserve: task strategy. Transfer proof: new dense passages across different themes.

Profile 29 — Full-paper score fluctuates with time of day

Morning practice is stable; late-evening practice shows slower reading and weaker Application.

Diagnosis: condition sensitivity rather than a new conceptual gap. Active repair: representative scheduling and stamina, not wholesale method changes. Transfer proof: stable execution under examination-relevant conditions.

Profile 30 — Full-paper score drops after a difficult school period

Sleep, workload and stress are worse; several previously stable skills regress temporarily.

Diagnosis: interpret cautiously. Active repair: protect core routines, reduce unnecessary novelty and retest after conditions stabilise. Preserve: evidence of prior mastery. One bad period should not erase the learner model automatically.

Profile 31 — School teacher and tutor disagree on one connection answer

The difference concerns whether Passage B “qualifies” or “partly disagrees” with Passage A. Both acknowledge the same underlying nuance.

Diagnosis: label disagreement may be superficial. Active repair: explain the relation in plain language and show evidence. Preserve: conceptual understanding. Transfer proof: unseen pair where the learner can justify the relation without depending on terminology.

Profile 32 — Learner reaches self-regulated Paper 2 review

Before receiving marks, the learner identifies one overstrong inference, notices a duplicated Summary point, explains that one Connection may be qualification rather than contradiction and records that Application began six minutes later than intended. The tutor’s role shifts from basic diagnosis to calibration and prioritisation.

Training implication: external feedback can now become more selective. The system has achieved its deeper goal: the learner is increasingly able to inspect their own meaning control, execution and transfer state.

237. Workshop synthesis: same total, different next lesson

Two learners can both score 32/50 and need completely different training. One may require Summary selection; another may require whole-paper timing; another may have high content but low language control; another may have strong language but weak source fidelity. A score is useful only when connected back to the decisions that produced it.

238. Workshop synthesis: different total, same repair

A 26/50 learner and a 37/50 learner can share the same overclaiming habit. The stronger learner may compensate through vocabulary, local knowledge or easier theme fit. If the mechanism is recurrent, both need semantic calibration, though the intensity and surrounding support can differ.

239. Workshop synthesis: a repair can improve the mark indirectly

Reducing overannotation may not teach any new comprehension concept, yet it can protect Summary and Application time. Improving passage identity can raise Connections and Application without separate content lessons. Paper 2 feedback should therefore value system-level leverage rather than only direct correction of the section where marks were lost.

240. Workshop synthesis: every repair should name the proof condition

Before training begins, write what evidence would show the repair worked. “Better at Summary” is vague. “Correctly selects distinct target-relevant points on two unseen passages and retains them under timed section conditions” is testable. Training improves when success conditions are explicit.

241. Full Repair-Cycle Laboratory | Twenty Complete Marked-Paper Journeys

The earlier casebooks isolate errors and diagnostic profiles. This final laboratory follows twenty complete repair journeys from the marked paper to the next unseen verification. Each cycle answers five questions: what did the old paper reveal, where did the first important decision fail, what is the smallest useful practice, what changes in the transfer task, and what evidence is required before the repair can be retired?

Cycle 1 — From literal inference answers to evidence-bound interpretation

Marked-paper evidence. Three inference questions receive weak credit. In each, Iona locates the correct passage area but either copies the clue or adds a motive that the writer never establishes. Her Summary and Application are otherwise sound, so this is not a general comprehension collapse.

First weak decision. She does not distinguish evidence from implication. The clue and the inference are mentally merged.

Reduced repair. Five short items are rewritten using clue → relationship → inference → boundary. For each, Iona must explain what the evidence supports and what it does not prove. The tutor models only the first example.

Changed transfer. Two days later she receives an unseen passage on a different topic. The questions use different wording and one inference is deliberately strong while another is tentative.

Verification. The repair counts only if she calibrates both directions correctly without the old four-step frame visible. A later timed Passage 1 checks whether the same control survives section pressure.

Cycle 2 — From word matching to proposition locating

Marked-paper evidence. Maren repeatedly chooses lines that contain the same keyword as the question but do not answer the requested idea. Her explanations are articulate once the correct line is shown.

First weak decision. Evidence search begins lexically rather than conceptually.

Reduced repair. Before opening the passage, Maren paraphrases each question target in plain language. She then chooses among three candidate lines and must explain why two are distractors despite vocabulary overlap.

Changed transfer. The next passage deliberately repeats key words such as “trust”, “risk” and “choice” across different paragraphs. The correct answers use paraphrased rather than repeated wording.

Verification. Success means proposition accuracy improves without large increases in search time. The final test is a full Passage 1 in which she uses conceptual locating naturally, without writing the target paraphrase explicitly.

Cycle 3 — From repeated Summary comments to one selection repair

Marked-paper evidence. Across three papers, Leonie receives “missing points”, “too many examples” and “outside scope”. Her paraphrase is reasonably accurate.

First weak decision. The common mechanism is target filtering. She reads the source well but does not classify idea units according to the exact Summary category before drafting.

Reduced repair. She completes target-switch exercises on the same passage: reasons, consequences and proposed solutions. She builds point skeletons only, with no prose. Each candidate idea is tagged include, exclude, example or overlap.

Changed transfer. An unseen passage uses a less obvious target that cuts across paragraph boundaries.

Verification. The repair succeeds when Leonie selects the correct set before writing, then retains those points during a timed Summary. Better sentence style alone does not count as proof.

Cycle 4 — From good Summary content to unstable semantic paraphrase

Marked-paper evidence. Iona usually selects the correct Summary points but repeatedly changes “may”, “some”, and “is associated with” into stronger claims while writing her own words.

First weak decision. Reconstruction prioritises lexical distance over truth conditions.

Reduced repair. She hides the source after understanding each point, writes the meaning from memory, then reopens the passage and checks population, certainty, causality and attribution. Synonym hunting is prohibited.

Changed transfer. New material includes several technically simple sentences whose difficulty lies in qualification rather than vocabulary.

Verification. The repair is considered stable only when semantic force remains accurate across both Summary and Passage 1 paraphrase under time. This is a cross-task skill, so one good Summary is not enough.

Cycle 5 — From vague Connections to exact relation control

Marked-paper evidence. Maren finds relevant lines in both passages but writes “they agree” or “they disagree” without specifying the proposition or dimension.

First weak decision. She treats comparison as a label rather than a relation between aligned claims.

Reduced repair. For each pair she writes A’s proposition, B’s proposition, the comparison dimension and the relationship in plain language before using any label. Relation types include partial agreement, qualification, extension, narrowing, different level and different time horizon.

Changed transfer. The next pair shares topic vocabulary but differs mainly in criterion rather than conclusion.

Verification. Maren must identify the exact relation and explain what the connection changes about the combined understanding. A later timed Connections set removes the relation menu.

Cycle 6 — From accurate Connections to expensive rereading

Marked-paper evidence. Iona’s Connections answers are conceptually strong, yet they consume so much time that Application begins late.

First weak decision. She rebuilds passage understanding from zero for every comparison.

Reduced repair. Passage identity lines and proposition maps are built once, then reused across several connection questions. Re-entry is limited to verifying exact evidence rather than rereading whole passages.

Changed transfer. A new three-passage set contains four connection operations, forcing repeated use of the same map.

Verification. The repair succeeds only if connection accuracy stays stable while total re-entry time falls and the later Application section gains protected time. Speed without preserved precision is not success.

Cycle 7 — From Singapore knowledge dump to passage-led Application

Marked-paper evidence. Leonie writes detailed local examples but the writer’s proposition disappears after the first sentence. The paragraphs could be moved into a Paper 1 essay with little change.

First weak decision. Local knowledge is selected before the passage mechanism is fixed.

Reduced repair. Every practice paragraph begins with a one-line passage claim and a mechanism label. Local evidence is not allowed until the learner states what condition would make the claim stronger or weaker in the chosen society.

Changed transfer. A new passage addresses a familiar theme but through a different causal mechanism, so the old memorised examples no longer fit automatically.

Verification. The learner must choose new evidence because it tests the writer’s mechanism, then state how strongly the idea applies. Local detail without passage fidelity does not count.

Cycle 8 — From weak local evidence to mechanism-based context retrieval

Marked-paper evidence. Maren represents passage ideas accurately but uses vague context claims such as “people in Singapore value education” or “the government is strict”.

First weak decision. Context knowledge is stored as broad topics rather than usable mechanisms and evidence.

Reduced repair. She reorganises local knowledge under mechanisms: incentives, trust, regulation, access, inequality, demographics, state capacity and media structure. Each card contains one verified example and one limitation.

Changed transfer. A new Application passage is selected specifically because its surface topic differs from the stored example but its mechanism matches.

Verification. Maren must retrieve by mechanism and explain fit. If she simply recognises a topic and inserts a familiar example, the repair has not transferred.

Cycle 9 — From late Application collapse to early time-debt repair

Marked-paper evidence. Application is underdeveloped in four papers. Initial feedback focuses on AQ practice, but timestamps reveal that the learner reaches Application with too little time because Passage 1 and Summary are repeatedly overchecked.

First weak decision. The system fails earlier through decision closure, not through lack of Application knowledge.

Reduced repair. Leonie practises bounded review on short-answer items and a reason-before-change rule for paraphrases. Summary sentences are closed once faithful and grammatical.

Changed transfer. A linked Passage 1 + Summary + Application exercise measures whether earlier sections finish inside a tested range.

Verification. Application quality must improve because more time is available, while early-section accuracy remains stable. If early marks fall, the repair has merely shifted the problem.

Cycle 10 — From fast completion to underdeveloped reasoning

Marked-paper evidence. The learner finishes Paper 2 ten minutes early but receives thin Application marks and several one-line evaluation answers.

First weak decision. Speed is being treated as the goal. Answers close before the reasoning job is complete.

Reduced repair. Each task receives a minimum completeness criterion. Passage 1 evaluation must include the limitation and consequence; Application paragraphs must include passage idea, context evidence, mechanism fit and judgement.

Changed transfer. The learner completes a timed section without any instruction to “slow down”. The only requirement is full task completion before closure.

Verification. Total time may rise modestly, but content development should improve while the whole paper still finishes. The target is efficient completeness, not maximum speed.

Cycle 11 — From source-attribution errors to writer identity control

Marked-paper evidence. A claim from Passage 3 is attributed to Passage 1 in Connections, and an objection quoted by a writer is treated as that writer’s own view in Passage 1.

First weak decision. Claim ownership is not encoded clearly during reading.

Reduced repair. Iona labels proposition owner: author, critic, study, participant or institution. Passage identity lines distinguish central stance from reported alternatives.

Changed transfer. A passage set contains multiple quoted voices and one writer who concedes an opposing view before rebutting it.

Verification. Iona must preserve ownership across short-answer, Connections and Application. Because attribution is cross-task, retirement requires more than one task type.

Cycle 12 — From formulaic language effect to contextual analysis

Marked-paper evidence. Every language question receives phrases such as “creates impact”, “engages the reader” or “emphasises the point”.

First weak decision. Technique terminology is retrieved before contextual meaning.

Reduced repair. Technique names are temporarily banned. The learner must state what the wording means here, what impression or relationship it creates and why that matters in the writer’s argument.

Changed transfer. New prose uses syntax, clause order and contrast rather than obvious metaphor.

Verification. Strong answers must remain specific even when no familiar device label is available. Later, terminology can return where it genuinely clarifies the explanation.

Cycle 13 — From criticism of every assumption to proportionate evaluation

Marked-paper evidence. The learner identifies assumptions accurately but treats the existence of any assumption as proof that the argument is weak.

First weak decision. Evaluation is binary rather than conditional.

Reduced repair. For each assumption, ask whether it is necessary, whether it is plausible, what evidence supports it and how much the conclusion depends on it.

Changed transfer. New arguments include both reasonable assumptions and fragile ones.

Verification. The learner should criticise only where the assumption materially weakens the reasoning and should explain the degree of impact. “There is an assumption” is no longer accepted as evaluation.

Cycle 14 — From reliable untimed performance to cue dependence

Marked-paper evidence. After tutoring, the learner can execute Summary and Connections correctly while a checklist sits beside the paper. Without it, the old errors return.

First weak decision. The strategy is externally cued rather than internally selected.

Reduced repair. The checklist is faded in stages: full instructions → short cue → self-generated question → no prompt. The learner predicts which strategy is needed before attempting each task.

Changed transfer. An unseen mixed task does not announce whether the main risk is selection, relation or semantic strength.

Verification. Independence is demonstrated when the learner selects the right process without the external label. A correct answer produced under a visible checklist is still supported performance.

Cycle 15 — From model recognition to independent generation

Marked-paper evidence. The learner can explain why a model Summary or Application paragraph is strong but cannot produce similar decisions independently.

First weak decision. Recognition is being mistaken for retrieval and generation.

Reduced repair. Model analysis is followed by closed-model reconstruction on different content. The learner states the principle, then writes without visible wording.

Changed transfer. A new passage shares the same reasoning demand but uses different topic, vocabulary and structure.

Verification. The learner must generate a faithful answer without borrowing model phrases. Later comparison with the model checks decisions rather than sentence similarity.

Cycle 16 — From teacher diagnosis to student diagnosis

Marked-paper evidence. Corrections work only after the tutor identifies the error family. The learner rarely notices issues before marking.

First weak decision. Monitoring is externally located.

Reduced repair. Before teacher feedback, the student completes a brief self-audit: uncertain answer, likely semantic risk, slowest section, weakest evidence fit and one preserved strength. The tutor then compares diagnoses rather than immediately supplying them.

Changed transfer. Several papers are reviewed in this order so self-prediction can be measured over time.

Verification. Progress is visible when student and tutor diagnoses converge and the learner begins to repair high-value errors before they are pointed out.

Cycle 17 — From good self-detection to prevention

Marked-paper evidence. The learner can find their own modal shifts and duplicate Summary points during review but still produces them in every first draft.

First weak decision. Monitoring occurs too late.

Reduced repair. The cue is moved into generation: before paraphrasing, ask “what must remain?”; before Summary drafting, check distinct point set; before Connections, state relation.

Changed transfer. First-draft error frequency is tracked on new tasks.

Verification. Prevention, not post-hoc correction, is the success criterion. The target is ready for retirement only when the old error occurs less often before checking begins.

Cycle 18 — From full-paper volatility to condition-specific diagnosis

Marked-paper evidence. Scores vary widely, but section skills look stable when isolated. Variation correlates with fatigue, unfamiliar prose and late testing time.

First weak decision. There may be no new semantic weakness; execution is condition-sensitive.

Reduced repair. Keep the established methods stable. Practise representative full-paper conditions and one pressure variable at a time. Avoid introducing new strategies simply because one low score occurred.

Changed transfer. Run papers across several realistic conditions while recording quality distribution.

Verification. Stability improves when performance variance narrows without changing the underlying method. The learner model should distinguish capability from condition-dependent access.

Cycle 19 — From broad feedback log to a small operating system

Marked-paper evidence. The learner maintains twenty active correction notes: tone, pronouns, own words, Summary overlap, timing, AQ examples, connectors and more. During mocks the list is unusable.

First weak decision. Feedback has not been prioritised or retired.

Reduced repair. Group notes into error families. Choose one primary active target, one maintenance target and one preserved strength. Move solved items out of the live card.

Changed transfer. The learner completes a new Paper 2 using only the compact operating card.

Verification. Performance should remain stable or improve while cognitive overhead falls. If a retired error returns repeatedly, reopen it at the condition where it failed.

Cycle 20 — From external correction to self-regulated Paper 2

Marked-paper evidence. Before receiving teacher comments, the learner identifies that one inference overreaches, one Summary point duplicates another, one Connection confuses qualification with disagreement, and Application started later than intended. They also identify a strong passage map that should be preserved.

First weak decision. At this stage there may be no basic feedback dependency. The task is calibration and prioritisation.

Reduced repair. The tutor verifies the learner’s diagnosis, challenges one questionable judgement and helps select the highest-value next test rather than rewriting answers.

Changed transfer. The learner designs part of the next retest: new theme, no visible cue, full timing.

Verification. The system has reached mature self-regulation when the learner can use the old paper as evidence, choose a plausible repair, test it on unseen material and update their own learner model with increasing accuracy.

242. Cross-task interaction: why repairs can have multiplicative value

Paper 2 is integrated. A single improvement can propagate. Better claim-strength control improves inference, Summary paraphrase, Connections representation and Application anchoring. Better passage identity improves locating, Connections and Application. Better decision closure protects later tasks. Feedback therefore should search for shared mechanisms before allocating four separate practice programmes.

243. Cross-task interaction: why repairs can also create side effects

Every repair can overshoot. A learner corrected for overclaiming may become excessively cautious. A learner trained for concision may omit necessary reasoning. A learner pushed for speed may lose semantic precision. A learner told to use more local evidence may turn Application into a knowledge dump. That is why every cycle preserves one functioning strength and checks for regression elsewhere.

244. The full-paper verification standard

A repair should be considered verified only when four conditions are reasonably satisfied: the better decision appears on unseen material; it survives after delay; it occurs with reduced support; and it remains available under representative Paper 2 load. Not every target requires perfection before moving to maintenance, but the evidence should be stronger than a corrected original answer.

245. The final feedback architecture

The mature Paper 2 cycle is compact:

1. Read the marked paper as evidence.
2. Find the first high-leverage weak decision.
3. Decide whether it is local or systemic.
4. Repair below full-paper load.
5. Change passage, wording or task.
6. Retest after delay.
7. Add section timing.
8. Verify in a full paper.
9. Retire, maintain or reopen according to evidence.

This architecture is intentionally reusable. The passages change; the diagnostic logic remains.