PSLE English Comprehension Open-Ended Performance | Turn 10 Questions into One Controlled Evidence System

The ten-question section has a state, not just a score

At any moment, the student carries more than an answer. The student carries a passage model, a sense of time, unresolved uncertainty, fatigue, confidence and memory of previous questions. These variables form a performance state. The state can improve, remain stable or deteriorate across the ten-item sequence.

A student may begin with a clean model and high confidence, encounter a difficult inference at Question 4, overspend time, carry doubt into Question 5, begin shortening answers at Question 7 and stop verifying at Question 9. The visible errors may be different, but the state transition connects them.

World-class open-ended training therefore studies sequence, not only outcomes.

The state vector

For diagnostic purposes, the student’s state can be described through six variables:

• Model clarity: how stable the passage structure remains.
• Task clarity: whether the current obligation is understood.
• Evidence certainty: how clearly the student knows what supports the answer.
• Time position: whether the student is ahead, on pace or behind.
• Residual uncertainty: how much unresolved thinking from earlier questions remains active.
• Execution quality: whether writing and verification remain controlled.

These are training lenses, not exam-day labels. They help the tutor explain why the same student can be excellent on Question 2 and careless-looking on Question 9.

State drift

State drift occurs when performance gradually degrades without one dramatic failure. Rereads become longer. Answers become shorter. Pronouns become less explicit. Verification disappears. The student may not notice because each change is small.

Tracking early, middle and late questions exposes drift. The repair may be endurance, automaticity or pacing rather than more content instruction.

State shock

State shock is different. One item abruptly changes the student’s behaviour. A difficult inference, unfamiliar word or apparent contradiction triggers overchecking or panic, and the next questions suffer.

Leonie’s recovery training focuses on state shock: contain one item, restore the next-question routine, and keep the rest of the section normal.

The recovery-latency measure

Recovery latency is the number of questions or minutes required before the student returns to normal process after disruption. A strong system shortens this latency.

If Question 5 is difficult but Questions 6–10 remain normal, recovery is good. If Questions 6 and 7 show rushed omissions and Question 8 still contains overchecking, the shock propagated.

Propagation cost

The cost of a difficult question is not only the mark on that question. It can include time, attention and downstream errors. This is propagation cost.

A one-mark uncertainty that causes two later omissions can cost more than the original item. Training should therefore evaluate containment, not merely whether the hard item was eventually solved.

The question-sequence map

After practice, mark each question as one of four states:

• Green: correct process, normal cost.
• Amber: correct or nearly correct but expensive or uncertain.
• Red: process failure or mark loss.
• Grey: answer outcome unclear but process evidence insufficient.

The pattern matters. Green–green–amber–red–red may indicate propagation. Red–green–red–green may indicate independent micro-skill gaps. A cluster of late ambers may indicate fatigue.

The evidence-ownership matrix

For advanced training, place passage regions on one axis and questions on the other. Mark which regions support each question.

The matrix reveals several important structures:

• one evidence zone supporting several questions;
• one question requiring distributed evidence from several zones;
• questions that should remain local;
• questions whose evidence overlaps but whose relations differ.

The matrix teaches evidence reuse without answer reuse.

Evidence reuse versus evidence recycling

Evidence reuse means returning to a passage detail because it legitimately supports another question. Evidence recycling means inserting the same detail repeatedly because it is memorable, even when its function no longer matches the obligation.

Good students reuse. Weak systems recycle.

The evidence monopoly problem

Sometimes one vivid sentence becomes a monopoly: the student uses it for four answers. This can happen after a dramatic turning point or memorable quotation.

Ask whether each question could be answered more directly from another evidence zone. If yes, the monopoly is a warning that passage mapping or evidence discrimination has narrowed too much.

The evidence orphan problem

The opposite occurs when an important passage region is never used. The student may have missed its function entirely. If a paragraph containing the key qualification never appears in any evidence set, later evaluation answers may become too certain.

Unused high-value zones deserve review.

The answer-ownership matrix

Evidence ownership answers “where does support come from?” Answer ownership answers “which semantic job belongs to this question?”

A student may have the right evidence and still assign the wrong relation. The matrix can label each item as literal, cause, change, evidence, comparison, inference, evaluation, reference or multipart. Its purpose is not to memorise question types but to check whether relation variety is producing answer-form drift.

Relation drift

Relation drift occurs when the student changes the semantic relationship during writing. “After” becomes “because”. “Suggested” becomes “proved”. “Different from” becomes two disconnected descriptions. “One reason” becomes a general summary.

This can happen even when every content word is accurate. The relation is part of the mark-bearing meaning.

Relation words deserve verification

In open-ended answers, small relation words carry large meaning:

• because;
• after;
• before;
• although;
• therefore;
• while;
• unlike;
• so that;
• only when;
• despite.

A final check should sometimes target the relation word rather than the nouns.

The proposition stack

Some answers contain more than one proposition. A two-reason answer may contain P1 and P2. A change answer may contain before-state and after-state. A cause–effect answer may contain mechanism and outcome.

Writing should preserve proposition boundaries clearly. Long sentences that merge several propositions can create pronoun ambiguity or relation drift. The best form is whichever makes the required propositions easiest to read faithfully.

The answer atom

An answer atom is the smallest meaning unit that could independently earn part of the answer’s credit. This is a training concept, not an official marking formula.

Identifying atoms helps students see why one long sentence can still be incomplete: it may contain only one distinct answer atom repeated in different words.

Redundancy between answer atoms

“He was tired because he was exhausted” looks longer but contains one idea. “She changed her plan because the venue moved and because the new time conflicted with her appointment” contains two distinct reasons.

Semantic distinction matters more than sentence count.

The question-contract ledger

For one training passage, write the ten obligations in a compact ledger before answering. Then compare written answers with the ledger afterwards. Any mismatch reveals contract drift.

Over time the ledger should disappear. Its purpose is to make an invisible process visible long enough to repair it.

Question-contract failure 1: quantity loss

The question asks for two reasons. The student identifies both while planning but writes one. This is not evidence failure; it is contract preservation failure during construction.

Question-contract failure 2: time boundary loss

The question asks why something happened after a specific event. The answer gives the general reason from earlier in the passage. Broad truth replaces local timing.

Question-contract failure 3: actor loss

The question asks why Lina was surprised. The student explains why Omar was surprised because both actors are central. Entity map is stable, but the local contract switched actor.

Question-contract failure 4: relation loss

The question asks how two approaches differed. The student gives two facts without establishing the common comparison dimension.

Question-contract failure 5: certainty loss

The question asks what the evidence suggests. The answer states the conclusion as definite fact. The content may be plausible, but the certainty relation changed.

Question-contract failure 6: source loss

The question asks what the survey revealed. The student gives a later outcome from the final paragraph. Both are true, but the source condition is wrong.

Question-contract failure 7: perspective loss

The question asks what Amir believed. The student answers with what the reader knows. The global passage model is richer than the character’s local knowledge and overwhelms the perspective boundary.

Question-contract failure 8: evidence-level loss

The question asks for an example from the passage. The student gives a general inference instead of the textual example.

Question-contract failure 9: response-form loss

The student knows the answer but leaves a fragment or ambiguous reference that does not stand clearly as a response. The reasoning survived; the executable answer did not.

Question-contract failure 10: stop-rule loss

The answer is complete, but the student keeps adding related material until an unsupported claim appears. The contract was fulfilled; stopping failed.

The ten-question state audit

After a full set, ask the student to reconstruct the performance sequence without seeing the marks. Which question felt expensive? Where did the passage model become less clear? Which answer was changed? Where did the student stop checking? This metacognitive report can be compared with actual errors.

Students often discover that the question they remember as “hardest” was not the one that cost the most marks. Cost and feeling are not identical.

The hidden correct-but-expensive answer

A correct answer can still be diagnostically weak if it required four rereads, two rewrites and three minutes. In a ten-question sequence, expensive correctness can create later failure.

Track correct-but-expensive items because they consume the system’s resources.

The hidden wrong-but-promising answer

A wrong answer can contain strong components: correct obligation, correct evidence, one certainty word too strong. This is very different from a completely wrong passage model. Repair only the broken layer.

The correction depth rule

Match correction depth to failure depth.

• Wrong punctuation in otherwise clear answer → surface correction.
• Ambiguous pronoun → construction correction.
• Correct evidence, wrong relation → proposition correction.
• Wrong evidence zone → routing correction.
• Wrong actor model affecting several questions → passage-model correction.

Do not reteach the whole comprehension system for a local defect.

The minimal correction principle

Whenever possible, make the smallest change that turns the student’s process into a reliable one. Preserve good reading, good evidence and good language. This reduces unnecessary cognitive reorganisation.

Original ten-question case 1: the forgotten qualifier

Question 1 asks why a character changed plans only after receiving a second message. Maren answers with the general concern from the first message. Her passage model is correct, but the temporal qualifier disappears. Later questions remain accurate. Diagnosis: local obligation failure, not global reading weakness.

Original ten-question case 2: the actor switch

Question 2 asks why Lina was surprised. Two same-gender characters appear in the paragraph. The student explains why Maya was surprised. Later inference questions also drift. Diagnosis: entity/reference instability with downstream propagation.

Original ten-question case 3: the expensive correct inference

Iona spends four minutes on Question 3, eventually writes a strong supported inference, then rushes Questions 4 and 5. The hard question itself is correct. The performance error is resource allocation and propagation.

Original ten-question case 4: evidence recycling

A vivid turning-point sentence answers Question 2 well. The student reuses it for Questions 3 and 4 even though one asks for the earlier cause and one for the later outcome. The evidence monopoly hides role differences.

Original ten-question case 5: one answer contaminates another

Question 4 asks what the character’s behaviour suggests. The student writes “She feared embarrassment.” Question 5 asks why she left early. The student uses “because she feared embarrassment” even though the passage explicitly says she had a medical appointment. Previous answer replaced passage evidence.

Original ten-question case 6: later evidence repairs an earlier answer

Question 6 directs attention to a sentence showing the event time changed. The student realises Question 2’s answer assumed the original time. Returning to Question 2 is justified because new passage evidence has been noticed, not because of vague doubt.

Original ten-question case 7: the two-reason omission

Leonie identifies both reasons in the passage and states both orally after the paper, but Question 8 contains one written reason. Early in the set she completed similar questions. Diagnosis: late execution/working-memory loss, not failure to understand “two reasons”.

Original ten-question case 8: over-paraphrase

The passage says the manager “hesitated”. The student writes “refused”. The answer shortens wording but changes meaning. Evidence and obligation were correct; construction introduced semantic drift.

Original ten-question case 9: one wrong relation word

The student intends “after the survey” but writes “because of the survey”. The survey occurred before the change but did not cause it; it merely preceded another decision. One word converts sequence into causation.

Original ten-question case 10: direct question overthought

A direct question asks where the item was placed. Iona gives a motive explaining why it was placed there. The location is buried inside a rich answer. Diagnosis: answer-level inflation, not weak comprehension.

Original ten-question case 11: wrong paragraph, right topic

The question asks why the revised plan succeeded. The student returns to the paragraph introducing the plan rather than the outcome paragraph. The topic matches, but evidence function does not. Diagnosis: routing by topic instead of paragraph role.

Original ten-question case 12: local question triggers global reread

Question 4 asks what “they” refers to in one sentence. The student rereads the entire passage. The final answer is correct but expensive. Diagnosis: poor search-space control.

Original ten-question case 13: distributed evidence under-collected

An inference depends on behaviour in Paragraphs 2 and 5. The student uses only Paragraph 5 and reaches a weaker conclusion. Diagnosis: local evidence treated as complete when the question requires the passage arc.

Original ten-question case 14: distributed evidence over-collected

A simple cause is stated in one sentence. The student scans three paragraphs for supporting detail and writes a long response. Accuracy remains, but cost rises. Diagnosis: evidence economy failure.

Original ten-question case 15: qualification lost across questions

Question 5 asks what a trial suggests, and the student correctly writes a tentative conclusion. Question 7 asks whether the method is proven effective, and the student says yes, forgetting the same passage’s small sample limitation. The model lost its qualification over time.

Original ten-question case 16: stable model, unstable answer form

The student can map every paragraph accurately and point to correct evidence but repeatedly writes vague “this shows that…” answers with unclear referents. Diagnosis: construction and reference closure.

Original ten-question case 17: confidence inversion

The student is highly confident on wrong literal answers produced by keyword matching but low confidence on correct inferences produced through careful evidence. Training must recalibrate confidence toward process quality.

Original ten-question case 18: changed answer without evidence

During final review, Leonie changes a complete answer because it looks “too short”. The new answer adds unsupported motive. Correct-to-wrong change reveals a verification rule based on appearance, not evidence.

Original ten-question case 19: marked uncertainty used well

Question 5 remains ambiguous after a bounded review. The student writes the best-supported answer and moves. All later questions are completed normally. At the end, new time allows one final check. Even if Question 5 remains imperfect, the system protected the rest.

Original ten-question case 20: marked uncertainty used badly

The student “moves on” from Question 5 but continues thinking about it during Question 6. The next answer is incomplete. Physical movement occurred; cognitive closure did not. Recovery must include attention reset.

Original ten-question case 21: question wording primes false motive

A question asks what “her reluctance” suggests. The student assumes reluctance means fear. The passage supports uncertainty but not fear. Question wording names a behaviour, not its cause.

Original ten-question case 22: mark value triggers padding

A two-mark question can be completely answered in one compound sentence containing two distinct reasons. The student adds a second sentence merely because “two marks needs two sentences” and introduces an irrelevant claim. Mark-value superstition damages sufficiency.

Original ten-question case 23: one sentence hides two atoms

“The plan succeeded because the new timetable allowed working residents to attend and the updated notices made them aware of the change.” One sentence contains two causal answer atoms. Sentence count does not determine completeness.

Original ten-question case 24: two sentences hide one atom

“He was exhausted. He was very tired.” Two sentences repeat one idea. Quantity of writing does not create semantic breadth.

Original ten-question case 25: one answer too dependent on question wording

The student writes “because this happened” using the question’s demonstrative structure without naming the event. The answer cannot stand independently enough to make the causal relation clear. Minimal reference closure fixes it.

Original ten-question case 26: a model answer creates script dependence

After seeing several model answers beginning “This suggests that…”, the student uses the phrase for direct questions. The language frame has become stronger than the obligation. Training must restore relation-first construction.

Original ten-question case 27: question 10 remains strong

Leonie uses a bounded stop rule on two earlier uncertain items and reaches Question 10 with normal pace. Her final answer contains full scope and local verification. The success is not only finishing; it is preserving decision quality late.

Original ten-question case 28: question 10 reveals cumulative cost

The final item is straightforward, yet the student omits a second point. Earlier answers are accurate but unusually long. The cause is cumulative writing cost, not final-question difficulty.

Original ten-question case 29: later question exposes wrong chronology

Question 7 asks what happened before the warning. The student realises the event spine used for Questions 2 and 4 reversed two events. Only answers depending on that sequence need revision; unrelated answers should remain closed.

Original ten-question case 30: later question does not justify global doubt

Question 8 is difficult and refers to a new paragraph. The student begins doubting every earlier answer despite no contradiction. This is uncertainty spillover, not evidence-based model update.

Original ten-question case 31: evidence proximity trap

The question appears immediately after a paragraph in the booklet, so the student assumes the evidence must be there. The correct support is two paragraphs earlier. Page proximity is not evidence ownership.

Original ten-question case 32: keyword matching trap

The question uses “change”. The student scans for the word “changed” and finds an irrelevant sentence. The real answer is expressed through before-and-after behaviour without that keyword. Passage model beats lexical hunting.

Original ten-question case 33: evidence and answer use different wording

The passage says “she remained beside the doorway until everyone else had entered”. The answer proposition is that she was reluctant to enter first. The student must preserve meaning without copying the whole sentence or inventing fear.

Original ten-question case 34: one true detail weakens answer

The correct cause is that the bus route changed. The student adds that the day was rainy. Rain is true but irrelevant. Extra truth can still reduce precision.

Original ten-question case 35: before-state omitted because it feels obvious

A change question asks how participation developed. The student writes only that the character later volunteered. The earlier avoidance is obvious from the passage but still needed to express change.

Original ten-question case 36: later-state omitted because answer starts too early

The student begins writing a detailed account of initial reluctance and runs out of sentence control before reaching the final volunteering. Overdevelopment of one slot crowds out another.

Original ten-question case 37: wrong evidence granularity

The question asks what specific action showed concern. The student answers “She helped him”, which is too broad when the passage says she returned to bring his medication. The evidence is directionally right but insufficiently precise.

Original ten-question case 38: overprecision

The passage suggests the character was concerned. The student writes that she “feared he would collapse”, adding a specific outcome never mentioned. Detail can become unsupported precision.

Original ten-question case 39: correct world knowledge, wrong evidence

A student explains that plants wilt because they lack water, but the passage attributes the change to heat exposure during transport. Scientific truth does not replace passage evidence.

Original ten-question case 40: answer from title instead of passage

The passage title suggests “A Second Chance”. The student uses redemption language in several answers even though the text supports only a practical correction. Titles can guide orientation but do not license themes beyond evidence.

Original ten-question case 41: wrong referent infects three questions

A pronoun in Paragraph 2 is assigned to the wrong character. Questions 3, 4 and 6 then appear wrong for different reasons. The first recurring cause is reference, not three separate question-type weaknesses.

Original ten-question case 42: paragraph-role map prevents oversearch

Question 8 asks whether the intervention worked. The student knows the outcome paragraph is Paragraph 5 and goes there first. The exact evidence is found quickly. Passage architecture reduces search cost.

Original ten-question case 43: one difficult word is not structural

An unfamiliar adjective appears in a descriptive sentence. The student spends two minutes resolving it even though the ten questions never depend on that detail. The local uncertainty budget was misallocated.

Original ten-question case 44: one difficult word is structural

The word “reluctantly” describes the central action targeted by two later inference questions. Here vocabulary is structural and deserves resolution. Same surface problem, different value.

Original ten-question case 45: local checking catches slot loss

Before moving on, Maren rereads the question, notices “two ways”, and adds the missing second behaviour. A five-second local verification saves a mark that final review might miss.

Original ten-question case 46: local checking creates harmful rewrite

The answer is complete. The student rewrites for elegance and changes “after” to “because”. Checking has become composition editing. Verification needs a stopping rule.

Original ten-question case 47: answer shorter than model answer but complete

The model answer uses two sentences. The student uses one precise sentence containing the same required propositions. Do not force expansion merely to resemble the model.

Original ten-question case 48: answer longer than model answer but still valid

A developing writer needs one extra clause to make the referent clear. The answer remains faithful and relevant. Economy is not minimum word count; it is absence of unnecessary cost and meaning.

Original ten-question case 49: one question becomes a model-repair signal

Two answer options for an inference both seem impossible under the student’s passage model. Instead of forcing one, the student revisits the relevant paragraph and discovers a missed contrast. The question exposes model error productively.

Original ten-question case 50: one question is simply hard

The passage model remains coherent, surrounding questions work, and evidence for one item is genuinely subtle. Do not overreact by rebuilding the entire reading system. Sometimes a hard question is local.

Original ten-question case 51: evidence distance grows late

Question 9 requires combining Paragraphs 1 and 6. The student’s map has decayed and only Paragraph 6 is used. Training should preserve long-range structure late in the sequence.

Original ten-question case 52: question position creates false urgency

A hard Question 10 causes panic because it is last. The student believes it must be solved immediately. In reality, all earlier items are complete and time remains. Emotional position, not evidence, drives the urgency.

Original ten-question case 53: the last answer is too short but correct

Fatigue makes the student suspicious of a concise final answer. Verification confirms obligation, evidence and clarity. No expansion is needed. Short is not synonymous with incomplete.

Original ten-question case 54: the last answer is short because one slot vanished

The question asks for cause and effect. The student writes only the cause. Here brevity reflects incompleteness. Verification should be semantic, not visual.

Original ten-question case 55: model update should be localised

The student discovers that the meeting happened Tuesday, not Monday. Only answers depending on day sequence need revision. Other answers about motive, outcome and comparison remain closed.

Original ten-question case 56: global update is genuinely needed

The student discovers that “they” refers to the organisers rather than residents throughout a central paragraph. Several actor-dependent answers must be reconsidered. The update is global because the underlying entity map changed.

Original ten-question case 57: answer proposition correct, grammar awkward

The answer’s meaning is clear despite minor language awkwardness. The tutor should distinguish comprehension proposition from language surface when diagnosing the first weak link, while still teaching clear expression.

Original ten-question case 58: grammar changes meaning

A misplaced pronoun makes the wrong person appear to act. Here language surface becomes semantic, so the construction error can cost comprehension credit.

Original ten-question case 59: answer copied exactly but wrong scope

The copied sentence is factually perfect but includes both cause and later consequence when the question asks only for cause. Copy accuracy is not answer relevance.

Original ten-question case 60: paraphrase shorter and better

The student converts a long passage sentence into one precise causal proposition while preserving actor, condition and certainty. This is efficient answer construction, not loss of detail.

Original ten-question case 61: passage map survives interruption

After a difficult Question 5, the student can still state “P2 problem, P4 turning point, P6 outcome” and route Question 6 directly. The map is resilient.

Original ten-question case 62: map collapses after writing

The student understands the passage immediately after reading but loses paragraph roles after writing three answers. Map decay, not initial comprehension, becomes the training target.

Original ten-question case 63: evidence is correct, certainty is wrong

Two clues support that a character was probably reluctant. The answer says she “definitely hated” the activity. Evidence selection works; epistemic calibration fails.

Original ten-question case 64: certainty is cautious when it need not be

The passage explicitly states the venue moved. The answer says it “might have moved”. Overcaution weakens a literal fact. Calibration means matching evidence in both directions.

Original ten-question case 65: strong evidence, wrong obligation

The student gives excellent proof that the plan succeeded when the question asks why it was introduced. Evidence quality cannot compensate for task mismatch.

Original ten-question case 66: weak evidence, right obligation

The student understands that the question asks for motive but selects a vivid outcome sentence rather than behavioural clues. Obligation works; routing/evidence fails.

Original ten-question case 67: answer-independent reference

The student writes “because she had moved it” when two female characters are possible. Replacing “she” with “the teacher” costs two words and removes ambiguity. Minimal closure increases robustness.

Original ten-question case 68: unnecessary question repetition

“The reason why he left early was because…” repeats the question but does not add meaning. A direct “He left early because…” is cheaper and clearer. Independence is not repetition.

Original ten-question case 69: model answer worship

The student changes a correct answer because it does not resemble the teacher’s wording. Training should compare semantic content, not lexical similarity. Many fitted answers can express the same supported proposition.

Original ten-question case 70: one question never deserved active repair

A rare unusual mistake occurs once across several papers and does not recur. The tutor records it but does not reorganise the programme. Pattern evidence matters.

The sequence-profile library

Across many students, several recurring ten-question profiles appear.

• Flat stable: accuracy and cost remain consistent.
• Early weak, later stable: slow orientation or initial model formation.
• Midpoint shock: one difficult item creates downstream errors.
• Late decay: fatigue, pacing or map decay reduces quality.
• Alternating instability: specific question mechanisms rather than state drift.
• High accuracy, high cost: strong knowledge but inefficient evidence/verification.
• Fast but brittle: quick answers with poor qualifier/evidence control.

The profile helps decide whether the next intervention should target skill, integration or execution.

Flat stable profile

A student with stable performance may not need more strategy. Training can shift toward harder unseen passages, maintenance and broader English development. Do not keep adding technique because the system already works.

Early-weak profile

If Questions 1–2 are slow or error-prone but later questions improve, the first passage model may take too long to stabilise. Work on first-pass orientation, not late-paper endurance.

Midpoint-shock profile

If one difficult question reliably creates a cluster, train stopping, marking, resetting and later return. The original hard item may remain difficult; the objective is to contain its reach.

Late-decay profile

If quality declines gradually after Question 7, compare writing length, earlier time cost and map retention. Late errors may be downstream consequences of early inefficiency.

Alternating-instability profile

If every inference question fails but direct questions remain stable regardless of position, the pattern is likely mechanism-specific. Do not diagnose fatigue merely because errors are distributed across the sequence.

High-accuracy high-cost profile

This student often looks excellent in untimed tuition and less reliable in the real paper. Train economy: smaller evidence sets, proposition-first writing, local stop rules and selective verification.

Fast-but-brittle profile

This student finishes comfortably but misses qualifiers, conditions and multipart slots. Speed is not the problem to remove; it is the resource to redeploy. Add selective slowing at high-risk triggers rather than slowing every question.

Selective slowing

World-class performance is not uniformly fast. It is fast on low-risk operations and deliberate at high-risk boundaries.

• quantity word → slow enough to preserve slots;
• two plausible interpretations → compare evidence;
• pronoun ambiguity → resolve actor;
• cause/sequence boundary → verify relation;
• answer change → demand new evidence.

This makes time allocation intelligent rather than equal.

Selective acceleration

Students should also learn what not to overwork.

• direct detail with exact evidence;
• simple reference with one clear antecedent;
• complete answer with no risk signal;
• question whose evidence is already known from the passage map.

Saving time on cheap items creates reserve for genuinely expensive ones.

The expected-value test for another reread

Before another reread, ask: what new information might I obtain? If the student can name the missing element—actor, cause, second clue, time boundary—the reread is targeted. If the answer is “maybe I will feel more certain”, marginal value is low.

The expected-value test for another sentence

Before adding another sentence, ask what required meaning it adds. If no new slot, evidence relation or necessary reference appears, the sentence may be cost without return.

The expected-value test for another check

Check when there is a risk signal. Rechecking a clean direct answer three times is low value. Rechecking a two-part late-paper answer is higher value. Verification should follow risk, not habit.

The ten-question compression ladder

• Stage 1: explicit passage map + written obligation + evidence marking + full verification.
• Stage 2: explicit map + mental obligation + selective evidence marks.
• Stage 3: compressed map + mental proposition + quick local check.
• Stage 4: mostly internal system under mixed unseen sets.
• Stage 5: exam-day routine with selective slowing only at risk signals.

Training becomes less visible as control improves.

The independence threshold

The student should not need the tutor to ask “What does the question want?”, “Where is your evidence?” or “Did you answer both parts?” on every item. Those cues must transfer inward.

Independence is proved when the student initiates the relevant check without prompting across unfamiliar questions.

The pressure threshold

A skill that appears only untimed remains incomplete for examination performance. The ten-question loop should survive realistic pressure, especially after one disruption and late in the set.

The delay threshold

Return to the system after attention has moved to writing, vocabulary or another subject. If obligation, evidence and verification routines survive delay, consolidation is stronger.

The genre threshold

Use narrative, informational and mixed-purpose texts. A student who can coordinate ten questions only on familiar stories has not yet built a transferable system.

The unfamiliar-topic threshold

Topic novelty raises reading load. The system should still preserve passage mapping and question isolation. If unfamiliar vocabulary causes global collapse, route back to foundational reading support rather than blaming open-ended technique.

The evidence-distance threshold

Near evidence is easier than distributed evidence. Test whether the student can maintain question obligation while combining clues from separated regions without overcollecting the whole passage.

The answer-length threshold

Students should be able to produce complete short answers and complete longer answers as obligation requires. If every answer expands to the same length, construction is template-driven rather than meaning-driven.

The final-control threshold

Question 10 is the stress test of the system. By then the student has read, located, written, doubted and verified repeatedly. Stable final-question performance indicates that the architecture has become affordable enough to last.

How to review a ten-question set without drowning in corrections

Do not spend equal teaching time on every error. Group them by first weak link. Three different red marks may share one obligation problem. Two answer-construction issues may disappear once time pressure is fixed. One model error may explain multiple downstream failures.

The review should identify the smallest number of mechanisms that explain the largest amount of lost performance.

The 80/20 repair decision

Ask which one or two repairs would recover the largest reliable share of marks and time. This is not a literal guarantee that 20% of causes create 80% of errors. It is a prioritisation habit: structural constraints before cosmetic ones.

The repair queue

• Active repair: one dominant mechanism.
• Secondary watch: one possible next constraint.
• Maintenance: recently repaired behaviour being checked lightly.
• Ignore for now: low-frequency, low-cost noise.

This keeps feedback focused enough to change behaviour.

The ten-question transfer portfolio

Keep evidence that the integrated system works across different conditions:

• narrative set;
• informational set;
• unfamiliar-topic set;
• timed set;
• set with one disruption;
• delayed retest;
• mock-paper integration.

Improvement is more credible when the system survives several surfaces.

Why the ten-question section deserves its own canonical owner

Question-reading pages, inference pages, literal-detail pages and answer-construction pages each own important mechanisms. None of them alone owns the coordination problem created by ten consecutive open-ended items sharing one passage environment.

This page owns that integration: resource sharing without contamination, repeated local contracts, sequence dynamics, cross-question updates, recovery, late-item stability and the conversion of ten separate marks opportunities into one controlled performance system.

The student should not feel ten systems

The final irony is that the more complete the training architecture becomes, the simpler the student’s experience should feel.

Understand the passage. Read this question exactly. Find its evidence. Decide the meaning. Write only what is needed. Check the risk. Move.

Then repeat. The complexity belongs to training and diagnosis. The examination routine belongs to the student.

Final deep standard

A world-class open-ended system must solve both the local problem and the sequence problem. Locally, each answer must be relevant, evidence-grounded, complete and faithful. Across the sequence, the student must preserve the passage model, allocate time intelligently, contain uncertainty, prevent contamination and arrive at the final question with the same basic control used at the first.

Maren’s local contract prevents broad understanding from becoming broad answers. Iona’s evidence discipline prevents one elegant inference from becoming a false premise. Leonie’s recovery architecture prevents one difficult question from becoming three lost marks.

The section is not ten tricks.

It is one passage, ten contracts, ten proof decisions and one performance system that must remain coherent from beginning to end.

The open-ended answer state machine

A ten-question open-ended section becomes easier to control when each question is treated as moving through a sequence of states. The student does not need to name these states in the examination, but training benefits from making them explicit because many errors occur when the student jumps ahead or reopens a question without a reason.

• Unread: no local task has been formed.
• Obligation specified: the question’s semantic contract is known.
• Evidence routed: the likely passage region is identified.
• Evidence secured: enough support has been selected.
• Proposition set: the answer meaning is clear before prose.
• Drafted: the proposition has become written language.
• Verified: scope, evidence, meaning and completeness have been checked proportionately.
• Closed: the answer is released and attention resets.
• Reopened: new evidence or a concrete defect justifies returning.

The important transition is from closed to reopened. Closure should be stable. A question should not drift back into active attention merely because the next item feels difficult or because a later answer uses similar language.

Illegal state transitions

Training can identify several expensive jumps:

• Unread → Drafted: writing before obligation or evidence is stable.
• Obligation specified → Drafted: proposition formed while evidence is still uncertain.
• Drafted → Closed: no local verification on a high-risk answer.
• Closed → Reopened: no new evidence or named defect.
• Reopened → Redrafted: whole-sentence rewrite when one word would repair the answer.

The state machine does not make the student slower. It exposes where unnecessary work enters the system so the final routine can become faster.

Question closure is a resource decision

Closing a question releases working memory. The student no longer needs to carry the local obligation, evidence set and wording problem forward. This is why unresolved doubt can be expensive even when no extra time is visibly spent: it occupies attention inside the next question.

Leonie’s training should therefore measure not only how long she spends on a difficult item, but whether the item remains mentally open after she leaves it.

The local-state reset

Close previous answer → clear local cache → read new obligation → rebuild only the evidence needed now.

The passage model persists. The local answer state does not.

The question graph

The ten questions form a graph over the passage. Some questions point to the same evidence zone. Some depend on different time layers. Some share an actor but ask different relations. Some later questions expose a passage feature that matters to an earlier answer.

The graph is not the same as a sequence. Question 7 may share evidence with Question 2 even though five items sit between them. A student who understands this can reuse the passage map without recycling answer logic.

The passage graph and the question graph are different

The passage graph describes entities, events, causes, changes, contrasts and paragraph roles. The question graph describes which of those relationships each item asks the student to extract.

Confusing the two creates two opposite problems. The student may answer questions too globally from the passage graph, or may read each question so locally that the shared structure is lost. PCEP-009 owns the coordination between those graphs.

Cross-question dependency

Questions are operationally independent but can depend on shared passage facts. If a central referent is wrong, several answers can fail. If one local motive inference is wrong, only one or two answers may be affected.

This leads to the idea of impact radius: how many answers depend on the same underlying passage assumption.

Small impact radius

A mistaken interpretation of one descriptive phrase affects only Question 4. The repair should remain local. Reopening unrelated answers creates unnecessary doubt.

Large impact radius

A pronoun in a central paragraph is assigned to the wrong actor, and Questions 3, 5 and 8 depend on that paragraph. A model correction now justifies a wider review because the dependency is structural.

Impact-radius drill

During review, choose one discovered error and ask: which answers actually depend on this fact? Students learn to revise by dependency rather than emotion.

Model coherence versus answer isolation

A strong passage model should be coherent across the paper. Individual answers should still be isolated by obligation. These principles are not contradictory.

Coherence means the same actor, chronology and stated facts are represented consistently unless the passage itself changes perspective or time. Isolation means one answer’s inferred motive or wording does not become evidence for another question automatically.

When consistency is a useful alarm

If Question 2 says the teacher moved the model and Question 6 seems to require the student to have moved it, the inconsistency is worth checking. One answer may contain an actor error.

When consistency is a trap

If Question 4 suggests a character was reluctant while Question 7 explicitly states she left for an appointment, forcing both answers to use “reluctance” would destroy fidelity. Different questions can legitimately describe different relationships.

Local uncertainty versus structural uncertainty

Local uncertainty concerns one answer: a phrase meaning, one motive, one second reason. Structural uncertainty concerns the shared model: who acted, what happened first, which intervention caused which result, or whose perspective the paragraph follows.

Structural uncertainty deserves more attention because its impact radius is larger. Local uncertainty should be bounded so it does not contaminate the sequence.

The uncertainty-classification question

If I am wrong here, how many later answers could become wrong for the same reason?

This question helps allocate time rationally during practice and mocks.

Marks-at-risk geometry

Marks are not only attached to questions; they are exposed to mechanisms. One actor-model error can place several marks at risk. One missing qualifier may place only one answer at risk. One overlong response can indirectly put several late questions at risk through time loss.

The tutor should therefore ask which mechanism has the widest and most repeatable mark exposure, not simply which red mark looks most dramatic.

Direct mark loss versus propagated mark loss

Direct loss is the mark lost on the original faulty question. Propagated loss appears later because the fault consumed time, damaged the model or carried unresolved uncertainty forward.

A student who eventually solves one hard inference but then omits two later answer slots may have converted a local difficulty into a system-level loss.

The question portfolio

A ten-item set usually contains questions with different costs. Some are cheap and direct. Some require moderate transformation. Some require distributed evidence or higher uncertainty. The student should treat the set as a portfolio rather than expect equal cost.

• Low-cost items: direct evidence, clear reference, simple detail.
• Medium-cost items: one relation or paraphrase transformation.
• High-cost items: distributed evidence, subtle inference, evaluation, model update.

The categories are dynamic. A question that is high-cost for one student may be low-cost for another. Training identifies the student’s actual cost profile.

Portfolio balance

Low-cost questions should not be allowed to inflate. Their role is to bank accurate marks cheaply. High-cost questions deserve additional attention only while that attention continues to produce useful evidence.

The hidden expensive easy question

A student can spend too long on a simple direct item because the answer looks short. This creates a poor portfolio decision: cheap mark converted into expensive mark.

The hidden cheap hard-looking question

A question stem can look complex while the passage explicitly states the required relation. Students should route to evidence before judging cost by wording alone.

Answer granularity across ten questions

Different questions demand different levels of detail. A direct detail answer can be compact. A comparison may require two balanced clauses. A causal explanation may need a mechanism. A change answer may need two states. A fixed answer length across all ten items is evidence that the student is writing from habit rather than obligation.

Semantic density

Semantic density describes how much required meaning is carried per unit of writing. High-density answers contain several necessary propositions cleanly. Low-density answers use many words to repeat one idea.

Evidence economy in PCEP-010 controls proof; PCEP-009 adds the sequence implication: low-density writing early can create late-paper time pressure.

High density is not compression at any cost

A sentence that crams several relations together and becomes ambiguous is too compressed. Density is useful only if actor, relation, scope and certainty remain readable.

Response-form diversity without templates

Students need flexible answer forms rather than one universal frame. The form should follow the semantic job.

• Direct: state the requested fact.
• Cause: event because mechanism/reason.
• Change: before state contrasted with after state.
• Comparison: common dimension + side A + side B.
• Inference: calibrated conclusion grounded in evidence.
• Evaluation: criterion + outcome + qualification where needed.
• Multi-part: one clear structure preserving every slot.

These are semantic shapes, not scripts. Word choice can vary.

Form switching

A strong student should switch from a direct one-clause response to a comparison sentence without carrying unnecessary phrases such as “This shows that”. Form follows task. This reduces both answer drift and writing cost.

Proof-to-prose conversion

Evidence is not yet an answer. The student must transform proof into the proposition the question asks for, then express that proposition in readable language.

Evidence selected → relation interpreted → proposition formed → prose written.

Skipping the relation step is why students sometimes copy a correct sentence that still fails to answer the question.

Answer latency

Answer latency is the delay between having enough evidence and beginning a stable response. Long latency can indicate proposition uncertainty, language formulation difficulty or overchecking. Short latency with high error rates can indicate premature commitment.

Tracking latency during practice helps distinguish reading cost from writing cost.

Search latency versus construction latency

If most time is spent before evidence is found, locating is expensive. If evidence is found quickly but writing takes long, answer construction is expensive. The repair should target the dominant latency rather than calling the whole question “slow”.

Late-sequence decision quality

Question 10 is not merely another item. It is a test of whether the system remained affordable. By then the student has repeated the cycle of reading, locating, deciding, writing and checking many times. Late quality exposes hidden costs accumulated earlier.

Late quality can fail without fatigue

A student may be mentally capable but simply have too little time because earlier answers were overlong or overchecked. Diagnose resource depletion separately from cognitive fatigue.

Late quality can fail without time shortage

A student can have minutes remaining but show weak late answers because the passage model decayed or attention became fragmented. Time on the clock is not the only resource.

Late quality can remain high under uncertainty

If one difficult middle item is bounded and released, later answers can remain normal even when that earlier item is still unresolved. This is evidence of strong recovery, not avoidance.

The ten-question control dashboard

• global passage-map quality;
• average obligation accuracy;
• wrong-zone routing rate;
• evidence under/overcollection;
• answer latency;
• local verification rate;
• reopen triggers;
• changed-answer outcomes;
• recovery latency;
• Question 8–10 stability.

The dashboard is diagnostic. It should collapse to one active repair decision, not become a permanent reporting burden.

Original ten-question case 71: illegal jump to writing

The student recognises a familiar topic and immediately begins a sentence before deciding whether the question asks cause or outcome. Halfway through, the answer must be rewritten. The lost time begins at the state transition, not the handwriting.

Original ten-question case 72: proposition set before evidence

Iona decides the character is jealous, then searches the passage for confirmation. Evidence collection becomes biased. The system should secure evidence before committing to the narrow motive.

Original ten-question case 73: answer closes cleanly

Question asks for one reason. Direct evidence is found, proposition is clear, one sentence is written, local verification confirms scope. The student moves on without carrying doubt. This low-cost closure is a model worth preserving.

Original ten-question case 74: answer never closes

Leonie writes a supported answer but keeps glancing back at it during Questions 5 and 6. No visible reread occurs, yet attention remains split. Training should include cognitive closure, not only physical movement.

Original ten-question case 75: reopened by new evidence

Question 8 directs the student to a sentence showing the event happened after the warning, not before. An earlier chronology answer is legitimately reopened because the passage model changed.

Original ten-question case 76: reopened by discomfort only

Question 2 was direct and well-supported. The student reopens it at the end because the answer is only six words. No defect is found. This is a low-value state transition.

Original ten-question case 77: small impact radius

A phrase in Paragraph 4 is misread, affecting one inference. Other questions rely on different evidence. Only that answer needs repair.

Original ten-question case 78: large impact radius

The student assigns “they” to residents instead of organisers. Three questions use the same paragraph. The referent repair justifies reviewing all three dependent answers.

Original ten-question case 79: consistency alarm saves actor error

One answer says the teacher moved the model; another implies the student did. The inconsistency triggers a passage check that repairs the second answer.

Original ten-question case 80: forced consistency creates error

A student makes two answers use the same motive even though the passage gives different reasons at different times. Coherence was confused with repetition.

Original ten-question case 81: structural uncertainty deserves time

The student is unsure whether a paragraph is flashback or current time. Several later questions depend on chronology. Resolving the structure has high expected value.

Original ten-question case 82: local uncertainty deserves containment

One metaphor has two plausible nuances, but no later question depends on it. A bounded answer is written and the paper continues.

Original ten-question case 83: cheap item stays cheap

A direct reference question has one clear antecedent. The student answers and releases in seconds, preserving reserve.

Original ten-question case 84: cheap item becomes expensive

The same reference question triggers a whole-paragraph reread and answer rewrite because the student distrusts simple questions. Cost inflation appears without accuracy gain.

Original ten-question case 85: hard-looking item is direct

The stem contains several clauses, but the relevant instruction is “According to Paragraph 5, what caused…?” Paragraph 5 explicitly states the cause. Routing compresses apparent complexity.

Original ten-question case 86: easy-looking item is distributed

“How did her attitude change?” looks simple but requires evidence from early and late passage regions. One local sentence cannot prove the arc.

Original ten-question case 87: high semantic density

“She first avoided speaking, but after successfully explaining the diagram she volunteered to present.” The sentence carries before-state, transition and after-state without redundancy. It is longer than a direct answer because the obligation is richer.

Original ten-question case 88: low semantic density

Three sentences say the character was nervous in slightly different ways while omitting the later change. More words create less coverage.

Original ten-question case 89: overcompression creates ambiguity

“She changed after it worked.” The answer hides actor, event and nature of change. Compression crossed the clarity threshold.

Original ten-question case 90: flexible form beats template

A direct location answer uses one clause. The next comparison uses “while” to balance two methods. The student does not force both into “This shows that…” frames.

Original ten-question case 91: evidence copied without relation

The student copies “the old timetable remained at the bus stop” but never states how this kept residents unaware of the new hours. Proof was found; proof-to-prose conversion failed.

Original ten-question case 92: relation expressed without overcopying

“Residents remained unaware of the new hours because the outdated timetable was still displayed.” The evidence meaning is converted into the required causal proposition.

Original ten-question case 93: search latency dominates

The student takes two minutes to locate a sentence but writes the answer in ten seconds. Passage routing, not sentence construction, is the bottleneck.

Original ten-question case 94: construction latency dominates

Evidence is identified instantly, but the student spends two minutes trying to phrase a comparison elegantly. Construction economy is now the target.

Original ten-question case 95: answer latency reflects unresolved proposition

The student has the right lines highlighted but cannot decide whether they show sequence or cause. Relation interpretation must be repaired before more writing practice.

Original ten-question case 96: answer latency reflects perfectionism

The proposition is clear, but the student delays because the first sentence does not sound polished. Comprehension answers need faithful executable language, not composition-level stylistic optimisation.

Original ten-question case 97: late failure from early verbosity

Questions 1–6 are correct but unusually long. Question 9 omits a second reason under time pressure. The visible late error is downstream of early writing cost.

Original ten-question case 98: late failure from map decay

Time remains, but the student cannot remember which paragraph contains the turning point and starts global rereading. The resource loss is structural memory, not clock time.

Original ten-question case 99: late quality survives disruption

A difficult Question 5 is marked and released. Questions 6–10 retain normal scope, evidence and verification. Recovery successfully contains propagation.

Original ten-question case 100: late quality collapses after disruption

Question 5 consumes four minutes. Questions 6 and 7 become terse and incomplete. The direct cost of Question 5 is only part of the loss.

Original ten-question case 101: reopen radius correctly limited

A chronology correction affects Q2 and Q8. The student revises those two only. Other closed answers remain untouched.

Original ten-question case 102: reopen radius spreads emotionally

The same chronology correction causes the student to reread all ten answers. No additional errors are found. Dependency-aware revision would have saved the time.

Original ten-question case 103: portfolio reserve used well

Five direct questions are answered economically. The saved time supports careful distributed evidence on two later items. Whole-section accuracy improves without rushing.

Original ten-question case 104: portfolio reserve squandered

The student finishes early questions quickly but uses the reserve to rewrite clean answers for style before reaching a difficult final inference. Time reserve existed but was allocated badly.

Original ten-question case 105: response form matches cause

“The organisers changed the schedule because working residents could not arrive before closing.” The causal relation is explicit and economical.

Original ten-question case 106: response form obscures cause

“Working residents could not arrive before closing. The organisers changed the schedule.” Both facts are present, but the relation is left implicit. A connector would increase clarity at low cost.

Original ten-question case 107: comparison dimension stable

“The first method required advance booking, whereas the second allowed walk-ins.” Both sides compare access procedure.

Original ten-question case 108: comparison dimension drifts

“The first method required booking, while the second was cheaper.” The answer gives two differences on different dimensions and may fail the intended comparison.

Original ten-question case 109: evaluation preserves criterion

The programme aimed to reduce waiting time; average waits fell. The answer evaluates success on the stated criterion.

Original ten-question case 110: evaluation swaps criterion

The programme reduced waiting time but cost more. If the question asks whether it met the waiting-time goal, cost may be secondary unless the criterion includes efficiency.

Original ten-question case 111: direct answer stays direct

The question asks where the document was found. The student writes the location only, with sufficient clarity. Strong readers should accept low-complexity obligations.

Original ten-question case 112: direct answer becomes inference

The same question receives a sentence about why the document was probably hidden there. Additional reasoning answers a question that was not asked.

Original ten-question case 113: inference remains calibrated

Two clues support reluctance, so the answer says the character “appeared reluctant”. The claim sits below the evidence ceiling.

Original ten-question case 114: inference contaminates next question

The student writes “she was afraid” in Q4 and then uses fear as the cause in Q5 without checking the explicit appointment evidence. Local inference became global premise.

Original ten-question case 115: question 10 gets full local check

Despite time pressure, the student verifies the second required slot on Q10. The check takes seconds because the risk trigger is known.

Original ten-question case 116: question 10 gets no check

The answer is rushed and omits the comparison side. Earlier overchecking consumed the verification budget.

Original ten-question case 117: delayed retest preserves state control

Three weeks after direct coaching, the student still closes each question cleanly, reopens only with evidence and keeps late items stable. The integrated routine has consolidated.

Original ten-question case 118: delayed retest reveals scaffold dependence

Without a tutor saying “new question, new obligation”, the student returns to broad passage summaries. The local reset cue has not transferred inward.

Original ten-question case 119: narrative transfer succeeds

The student applies the same obligation/evidence/proposition loop to a character-driven narrative despite different surface language.

Original ten-question case 120: informational transfer fails

The student who was stable on stories loses the model when reading an informational passage with claim–evidence–qualification structure. Genre transfer becomes the next target.

Original ten-question case 121: unfamiliar topic raises reading cost but system holds

Scientific vocabulary slows the first read, but question isolation, evidence routing and answer construction remain stable. The system tolerates topic novelty.

Original ten-question case 122: unfamiliar topic causes global collapse

The student abandons passage mapping and keyword hunts every question. Foundational vocabulary/reading support may be the first weak link, not open-ended technique.

Original ten-question case 123: correct answer with expensive wording search

The student spends ninety seconds replacing ordinary accurate words with sophisticated synonyms. The mark is safe; the sequence cost is not.

Original ten-question case 124: plain wording preserves difficult relation

A simple sentence clearly expresses cause and condition. Technical elegance is unnecessary when semantic fit is strong.

Original ten-question case 125: multi-part answer integrated well

Two requested ideas appear in one sentence with distinct clauses. Both slots are easy to identify and evidence-grounded.

Original ten-question case 126: multi-part answer split well

The same two ideas are clearer in two short sentences because one contains a complex condition. Form choice serves readability, not a fixed rule.

Original ten-question case 127: multi-part answer fused badly

Two ideas are crammed into one sentence with ambiguous pronouns. The attempt at economy reduces executability.

Original ten-question case 128: uncertainty budget allocated by impact radius

The student spends extra time resolving a central actor because three questions depend on it, but releases one local metaphor after a bounded check. This is rational allocation.

Original ten-question case 129: uncertainty budget allocated by emotion

A vivid metaphor receives three minutes while a central chronology ambiguity is ignored. The most uncomfortable item is not always the highest-impact item.

Original ten-question case 130: complete system under pressure

The student builds one passage model, routes each item locally, keeps previous answers from becoming evidence, spends more time only where impact or information value justifies it, preserves flexible answer forms, locally verifies high-risk responses and reaches Question 10 with normal control. No single technique explains the result. Coordination does.

Drill 31: answer-state tracking

For one set, mark when each question reaches obligation, evidence, proposition, drafted, verified and closed states. Identify illegal jumps and unnecessary reopenings.

Drill 32: closure test

After a student moves on, ask the next question immediately. Observe whether attention returns to the previous item. Train the reset cue until closure becomes clean.

Drill 33: impact-radius mapping

Give a model error and ask which of ten answers depend on it. Students learn surgical revision.

Drill 34: question-graph mapping

Link questions that share evidence zones, actors or structural relations. Then explain why shared evidence does not mean shared answer wording.

Drill 35: local versus structural uncertainty

Classify ten uncertainties by expected impact radius. Allocate a limited review budget accordingly.

Drill 36: portfolio cost audit

Record time by item and classify low/medium/high cost after the fact. Compare perceived difficulty with actual cost.

Drill 37: cheap-item compression

Practise direct questions with strict evidence sufficiency and immediate release. The goal is not rushing but preventing cost inflation.

Drill 38: high-cost bounded review

Give difficult questions with a fixed information target and review window. Students must articulate what another minute could add.

Drill 39: semantic-density edit

Take long answers and identify distinct answer atoms. Remove repetition while preserving every required atom.

Drill 40: overcompression repair

Take short vague answers and restore actor, relation or evidence meaning with the fewest additional words.

Drill 41: response-form switching

Use one passage with direct, cause, change, comparison and evaluation questions. Students must choose a form that matches each job without relying on one sentence frame.

Drill 42: proof-to-prose separation

First list the evidence meaning; then state the relation; only then write the answer. This exposes where copying replaces interpretation.

Drill 43: latency decomposition

Measure search time and writing time separately. Target the dominant cost.

Drill 44: late-sequence integrity

Place known operations at Questions 8–10 after demanding earlier items. Compare slot completion, evidence and latency with early versions.

Drill 45: model-update dependency

Reveal new evidence late and ask which earlier answers should reopen. Penalise unnecessary global rereading.

Drill 46: consistency alarm

Plant one true inconsistency between two answers caused by actor or chronology error. Students identify when cross-answer comparison is useful.

Drill 47: consistency trap

Plant two answers that legitimately differ because questions ask different times or perspectives. Students resist forcing them together.

Drill 48: three-resource audit

For a full set, record time, attention disruption and writing volume. Identify whether late failure is caused by the clock, state shock or construction cost.

Drill 49: delayed sequence retest

After several weeks, run an unseen set without reminders about resets or closure. Observe whether the integrated loop is self-initiated.

Drill 50: full ten-question postmortem

Reconstruct the sequence: model quality, obligation, evidence, proposition, answer state, verification, time cost, confidence, reopen events and recovery. Then choose one active repair that explains the largest share of lost performance.

Mock-paper postmortem: reconstruct the sequence, not just the marks

A mock-paper review should create a timeline. Mark where the student was on pace, where a state shock occurred, where answer latency rose, where local verification disappeared and whether later questions recovered.

Then separate direct losses from propagated losses. A hard Question 5 may be responsible for its own lost mark plus two late omissions. Conversely, a wrong Question 5 with no downstream effect may deserve less training priority than a recurring scope error.

The five-question compression test

After a full mock, select five questions representing different jobs. Re-answer them with the integrated loop and compare cost. If performance improves mainly because the student remembers answers, use a new passage. The test is the process, not recall.

The repair–retest gate

A repair is not complete because the student can explain it immediately. Retest under changed wording, new passage, realistic sequence position and delayed return. The integrated system should survive all four.

Retire the repair when the sequence remains stable

Once the target behaviour survives unfamiliar items and late positions without prompts, move it to maintenance. Keep active teaching capacity for the next constraint.

Final-examination compression for the ten-question system

Keep the passage model. New question, new contract. Route. Prove. Decide. Write. Check the risk. Close. Reset.

No student should carry the full analytical vocabulary of this article into the examination. The long architecture exists so the short routine is reliable.

Canonical handoff to evidence economy

When the sequence is stable but evidence search remains expensive, route to SETC-PCEP-010 — Evidence Economy. PCEP-009 owns coordination across questions; PCEP-010 owns the resource economics of proof inside each question.

Canonical handoff to verification

When answers are well-constructed but checking is weak, excessive or change-prone, route to SETC-PCEP-011 — Verification. PCEP-009 owns the ten-question loop; PCEP-011 owns the release decision after an answer exists.

Final sequence standard

A mature student does not experience ten unrelated puzzles. The passage remains one coherent model while every question gets a fresh local contract. Evidence can be reused without answer logic being recycled. Uncertainty is classified by impact. Cheap questions stay cheap. Difficult questions are bounded. Answer form follows meaning. Closed questions stay closed unless evidence changes.

Maren’s broad understanding becomes locally precise. Iona’s interpretations remain passage-dependent rather than self-propagating. Leonie’s time and attention survive the difficult middle item and reach the final question intact.

The open-ended section is twenty marks distributed across ten local decisions. The performance advantage comes from making those decisions share a strong model without sharing each other’s mistakes.