When one claim is wrong, isolate it, trace what depends on it, and recheck the remaining claims independently instead of discarding or defending the whole response. This guide is for a Primary 6 learner who can often complete the easy part of a question but loses accuracy when a familiar word, number relationship, observation or piece of evidence is used too broadly. The difficulty is not usually a lack of effort. It is a control problem: the mind recognises something useful and stops checking before the task contract has been fully satisfied.
The repair is to separate three moves. First, identify exactly what the information establishes. Second, identify what it does not establish. Third, connect only the supported conclusion to the wording of the question. That three-part habit is small enough to use under examination pressure and strong enough to prevent many confident but unsupported answers.
This article develops that habit through mechanism-led explanation, worked examples, misconception repair, independent practice, delayed transfer and examination control. It belongs to the complete Learner’s Guide volume directory and continues the earlier PSLE Learning Guide foundation. The current official subject documents can be reached through SEAB’s 2026 PSLE formats page; always use the document for the year in which you will sit the examination.
The learner problem
A learner discovers one error and reacts in one of two unhelpful ways. The learner may cross out an entire response, including correct independent points, or may defend every sentence because one part was right. Both reactions treat a bundle of claims as if it had only one truth value.
Multi-part responses often contain different relationships. An English explanation can identify a character’s action correctly but overstate the motive. A Mathematics solution can use a correct method after copying one value wrongly. A Science answer can name the correct observation but attach an unsupported mechanism.
The dependable control is claim-level checking: split the response, test each claim against its own evidence or rule, draw dependency arrows, repair the first affected point, and keep every independent claim that still survives.
Localise the error before you decide how far the repair must travel.
The mechanism: what is really happening
A claim is a statement that can be checked. One sentence may contain several claims joined by and, because, therefore or a relative clause. If those claims rely on different evidence, success or failure does not automatically spread across the sentence.
Dependencies create legitimate spread. If part b uses the value from part a, a changed value may require part b to be recomputed. If a conclusion rests on a false premise, the conclusion loses support even if it happens to be true. The task is to trace actual dependence, not emotional confidence.
Independence stops unnecessary spread. A spelling correction in one sentence does not alter an unrelated comprehension answer. A wrong perimeter does not invalidate a separately calculated area if the area used correct original dimensions. A weak mechanism does not erase an accurately reported observation.
Good repair preserves diagnostic information. Mark the precise claim, name the error type, revise affected dependants, and reread the surviving claims. This makes review faster and teaches which early decision caused the difficulty.
A compact decision routine
- Split the response at each distinct claim, calculation result or conclusion.
- Test every part against its own evidence, definition, operation or unit.
- Draw arrows from premises and intermediate values to the claims that use them.
- Repair the earliest wrong or unsupported node and recompute only genuine dependants.
- Reread the surviving independent claims and leave one clear final response.
Say the routine aloud during practice until it becomes internal: split, test, trace, repair, preserve. The purpose is not to make every answer slow. It is to make the checking stage reliable enough that speed can grow safely.
Precise worked examples
Worked example 1: English inference
Prompt. “Lena closed the window, so she was angry and wanted silence.”
Reasoning. Closing the window is stated; anger and motive require separate evidence.
Answer. Keep the action, qualify or remove unsupported motives.
Control check. Do not reject the stated action merely because the inference is weak.
Worked example 2: English grammar pair
Prompt. A sentence has correct tense but wrong pronoun case.
Reasoning. The two features use different checks.
Answer. Repair the pronoun while preserving the correct tense.
Control check. Reread the whole sentence after the local edit.
Worked example 3: Mathematics copied value
Prompt. A pupil copies 48 as 84, then applies the correct division method.
Reasoning. The method may be valid but the numerical result depends on wrong data.
Answer. Correct the input and recompute dependent lines.
Control check. Record both data error and method status.
Worked example 4: independent geometry
Prompt. Perimeter is wrong, but area uses correct side lengths independently.
Reasoning. Area does not use the perimeter result.
Answer. Repair perimeter only and recheck area independently.
Control check. Do not assume the shared diagram creates dependence.
Worked example 5: Science observation and mechanism
Prompt. “The water level fell, proving the plant drank all the water.”
Reasoning. The fall is an observation; the exclusive mechanism is unsupported because evaporation may occur.
Answer. Keep the observation and repair the explanation.
Control check. Separate what was seen from why it happened.
Worked example 6: Science classification
Prompt. “The animal has six legs, so it is an insect and can fly.”
Reasoning. The leg evidence supports insect classification at this level; flight is a separate claim.
Answer. Keep the supported classification and remove the universal flight claim.
Control check. Not all insects fly.
Worked example 7: part dependency
Prompt. Part b says use the answer from part a.
Reasoning. A change in part a travels into part b.
Answer. Recompute part b with the corrected value.
Control check. The instruction makes dependency explicit.
Worked example 8: independent subpart
Prompt. Parts a and b share a passage but ask unrelated details.
Reasoning. Shared source does not make one answer depend on the other.
Answer. Check each directly against the passage.
Control check. Do not erase b because a was wrong.
Worked example 9: true conclusion from bad reason
Prompt. A correct option is selected for an incorrect reason.
Reasoning. Truth of the conclusion does not validate the reasoning.
Answer. Keep the option only after rebuilding a valid reason.
Control check. Distinguish answer accuracy from method accuracy.
Worked example 10: mixed final response
Prompt. A final sentence contains one correct value and a stale unit.
Reasoning. The quantity claim includes both number and unit.
Answer. Repair the unit because it is part of the same dependent claim.
Control check. Independence must not be invented inside one quantity.
Common misconceptions and targeted repairs
Misconception 1: one error means start everything again
Why it happens. Anxiety replaces dependency analysis.
Targeted repair. Localise the first error and trace arrows.
Micro-test. Which later lines actually use it?
Misconception 2: one correct fact saves the whole answer
Why it happens. Relevance is confused with completeness.
Targeted repair. Test every joined claim separately.
Micro-test. Does the evidence support each side of and?
Misconception 3: same question means all parts depend
Why it happens. Layout is mistaken for logic.
Targeted repair. Read each command and source relationship.
Micro-test. Could part b be answered without part a?
Misconception 4: a correct final number proves the method
Why it happens. Coincidence and compensating errors are ignored.
Targeted repair. Verify every transformation.
Micro-test. Would the method work with new numbers?
Misconception 5: a wrong explanation erases an observation
Why it happens. Claim and mechanism are fused.
Targeted repair. Label observation and explanation separately.
Micro-test. What was directly seen?
Misconception 6: a correct method preserves a result from wrong data
Why it happens. Method quality and input accuracy are merged.
Targeted repair. Recompute all numeric dependants.
Micro-test. Which values came from the copied input?
Misconception 7: preserving work means never changing it
Why it happens. Effort is confused with evidential value.
Targeted repair. Keep only claims that survive independent checking.
Micro-test. Can the claim still be justified now?
Misconception 8: every neighbouring sentence is independent
Why it happens. The opposite shortcut ignores references and therefore links.
Targeted repair. Trace pronouns, values and logical connectors.
Micro-test. What premise does “therefore” use?
Graded independent practice
Attempt each item before opening its explanation in your mind. Write the smallest complete answer that proves the required idea. The questions rise from recognition to explanation and transfer; accuracy matters more than rushing.
Practice 1: recognition
Question. What is the first action after finding one error?
Explained answer. Split the response and identify the exact claim or value that failed.
Practice 2: dependency
Question. Part b explicitly uses part a. Part a changes. What follows?
Explained answer. Recompute part b.
Practice 3: independence
Question. Two answers use the same passage but different evidence. One is wrong.
Explained answer. Check the other independently; do not discard it automatically.
Practice 4: English
Question. A character action is stated, but motive is guessed.
Explained answer. Keep the action and repair or qualify the motive.
Practice 5: English
Question. A sentence has correct vocabulary but wrong agreement.
Explained answer. Repair agreement while preserving the suitable word choice.
Practice 6: Mathematics
Question. Correct operation, wrong copied number. Is the method wrong?
Explained answer. Not necessarily; correct the input and recompute the dependent result.
Practice 7: Mathematics
Question. A wrong area is not used in a later perimeter calculation.
Explained answer. The perimeter may remain valid, but verify it from original dimensions.
Practice 8: Science
Question. An observation is correct but cause unsupported.
Explained answer. Preserve the observation and replace the causal explanation.
Practice 9: Science
Question. A classification is correct but a universal behaviour claim is false.
Explained answer. Keep the classification and remove or qualify the behaviour claim.
Practice 10: logic
Question. Does a true conclusion make its reason true?
Explained answer. No. Rebuild the evidence chain.
Practice 11: logic
Question. Does a false reason guarantee the conclusion is false?
Explained answer. No. It means that reason does not support the conclusion.
Practice 12: recording
Question. A number is correct but unit wrong.
Explained answer. The final quantity is wrong until the unit is repaired.
Practice 13: trace
Question. What does a dependency arrow represent?
Explained answer. One claim, value or premise being used to produce another.
Practice 14: repair
Question. Where should recomputation begin?
Explained answer. At the earliest wrong or unsupported node that feeds later work.
Practice 15: preservation
Question. When may a claim remain?
Explained answer. When it still has valid independent evidence and meets the command.
Practice 16: scope
Question. Why test each side of “and”?
Explained answer. The conjunction can bundle two claims with different support.
Practice 17: transfer
Question. A graph description is right but causal explanation wrong.
Explained answer. Keep the accurate description and repair the causal claim.
Practice 18: transfer
Question. An essay paragraph has one inaccurate date but a separate valid argument.
Explained answer. Correct the date and check whether the argument actually depends on it.
Practice 19: delayed retrieval
Question. State the five verbs in the routine.
Explained answer. Split, test, trace, repair, preserve.
Practice 20: exam control
Question. What is the stopping condition?
Explained answer. Every active claim is supported, and every genuine dependant of a correction has been rechecked.
Delayed transfer: use the idea after the surface changes
Do not repeat these transfer tasks immediately after the worked examples. Leave at least one study session between learning and transfer. Delay removes the comfort of short-term imitation and reveals whether the mechanism can be reconstructed from memory.
Transfer 1: marking feedback
New setting. A teacher circles one phrase in a long response.
What transfers. Identify its scope before rewriting everything.
Answer check. Ask which later claims depend on that phrase.
Transfer 2: spreadsheet
New setting. One input cell is corrected.
What transfers. Trace formulas that reference it.
Answer check. Independent cells do not need manual alteration.
Transfer 3: source evaluation
New setting. An article contains one wrong statistic.
What transfers. Correct it and reassess only arguments that use it.
Answer check. Do not grant or deny credibility wholesale.
Transfer 4: oral answer
New setting. A learner self-corrects one word.
What transfers. Preserve the rest if grammar and meaning remain supported.
Answer check. Reread the complete active sentence.
Transfer 5: experiment report
New setting. One reading is anomalous.
What transfers. Investigate its effect on summaries and conclusions.
Answer check. Do not delete unrelated observations.
Transfer 6: delayed mixed set
New setting. Responses contain local errors and dependency chains.
What transfers. Draw a claim map before editing.
Answer check. Compare repair extent with actual arrows.
Build a proof-quality answer
A correct thought can still produce an incomplete written answer. To make your reasoning visible, build the response from three parts: a claim that answers the exact question, evidence drawn from the given sentence, quantities, observation or condition, and a bridge explaining why that evidence supports the claim. The bridge is often the missing part. Learners commonly copy a relevant fact and expect the examiner to infer the connection. Instead, write the relationship explicitly. In a short-response item, each part may be only a phrase; in an explanation item, it may take a complete sentence.
The claim should be no broader than the evidence. If the information proves only what happens in one stated case, do not turn it into a universal rule. If a comparison identifies which value is greater, do not invent the size of the difference. If a word modifies an adjective in one sentence, do not claim that it must have the same job in every sentence. Precision is not timid writing. It is the discipline of matching the strength of the conclusion to the strength of the support.
The evidence should be selected rather than copied wholesale. Ask, “Which detail changes the answer?” A complete prompt may contain setting, names, decorative numbers and contextual sentences that make the question readable but do not justify the conclusion. Circle the smallest decisive detail. Then test it by removal: if deleting the detail would make your answer unsupported or change it, the detail is probably doing real evidential work. If deleting it changes nothing, it may be background rather than proof.
The bridge names the mechanism. Useful bridge language includes “because this means…”, “therefore the two values must…”, “this modifies… rather than…”, “so the rate changes while the starting amount remains…”, and “this fact is relevant but does not establish…”. Do not copy these stems mechanically. Use the one that expresses the actual relationship. A bridge earns its place when it would help a careful classmate see why the evidence, and not some other fact, leads to the claim.
The boundary sentence
After drafting the answer, add a silent boundary sentence in your mind: “This does not yet prove that…” Complete it with the most tempting overstatement. The boundary sentence exposes hidden leaps. For example, an observation over a fixed time does not necessarily prove what happens until completion; a parity condition does not by itself identify two numbers; a familiar suffix does not settle word class; and a relevant detail does not guarantee that all required information is present. You do not always write the boundary sentence in the final response, but thinking it protects the sentence that you do write.
The opposite-case test
Try to imagine a case in which the given evidence remains true but your proposed conclusion becomes false. If such a case is possible under the question’s conditions, the conclusion is too strong or something is missing. This is a powerful test because it turns vague doubt into a concrete counterexample. Keep the counterexample small. Change one feature only, preserve the stated information and see whether the answer survives. When it fails, repair the conclusion or name the additional information that would be needed.
The language-strength ladder
Notice the strength of common answer words. “May” expresses possibility. “Is likely to” expresses a supported tendency, not certainty. “Will” states a definite outcome. “Always” and “never” claim no exceptions within the stated domain. “More than” gives an order; “twice as much” gives a multiplicative comparison; an exact value gives still more information. Choose the lowest rung that fully answers the question and is fully supported. Moving higher without evidence creates error; staying lower when the evidence is decisive creates an incomplete answer.
A diagnostic laboratory for independent learners
Use the following laboratory with four questions from your own schoolwork. Choose one that you answered correctly with confidence, one that you answered correctly by guessing, one that you answered incorrectly, and one that you left incomplete. This mixture matters. Studying only wrong answers can hide fragile success, while studying only correct answers can hide the point at which the method breaks down. Place the four responses on separate pages or separate sections so that you can annotate them without rewriting history.
Pass 1: reconstruct the first decision
Cover the marking and solution. For each item, write the first meaningful decision you made after reading: the word you classified, the relationship you selected, the variable you treated as fixed, or the detail you believed was sufficient. Do not write “I started solving”. Name the decision precisely. Many later errors are consequences of an early classification. Finding that point is more useful than circling the final wrong word because the early decision is what must change next time.
Pass 2: separate knowledge from control
Ask whether the error came from missing knowledge or from failing to control knowledge you already possessed. A knowledge gap means you could not state or reconstruct the needed concept even without time pressure. A control gap means you knew the concept but selected it too quickly, ignored a condition, overgeneralised it or failed to check the conclusion. The repairs differ. Knowledge needs explanation, examples and retrieval. Control needs contrasting cases, decision routines and deliberate checking. Calling both problems “carelessness” prevents either from being repaired.
Pass 3: mark the evidence chain
Use three simple labels beside the response: C for claim, E for evidence and B for bridge. If a label is missing, add the missing part in a different colour. If the response contains two claims, check each one separately; evidence for the first claim may not support the second. If a sentence contains several facts, underline only the decisive one. This annotation makes completeness visible and teaches you to produce an answer whose logic can be checked line by line.
Pass 4: write a minimal repair rule
Finish each analysis with a repair rule of no more than twelve words. It must describe an action you can perform, not a feeling you hope to have. “Be more careful” is not usable. “Name the modifier’s target before naming its class”, “check same parity before halving”, “separate rate from starting amount”, and “list what is still missing” are usable because they tell the brain what to do. Reuse the rule on a new question within two days, then revise it if it does not trigger the right behaviour.
Create your own contrast set
Deep understanding grows when you can create neighbouring cases that require different answers. Start with one correct example from this guide. Make Version A by changing a surface detail that should not change the reasoning. Make Version B by changing one decisive condition that should change the answer. Make Version C by removing a necessary detail so that the answer can no longer be determined or justified. Write the answer and the reason for all three versions before asking anyone to check them.
A useful Version A tests whether you are distracted by appearance. You might change names, rearrange a sentence without changing the modifier’s target, use different whole numbers with the same structural relationship, replace one material with another while keeping the relevant variables controlled, or put a needed clue in a different paragraph. If your method changes merely because the surface changed, you were remembering a picture of the example rather than its mechanism.
A useful Version B changes exactly one controlling feature. The discipline of changing one feature prevents a muddled comparison. After making the change, complete the sentence, “The answer changes because…”. If you cannot name the causal or logical difference, the pair is not yet teaching you anything. Revise it until the difference is visible. This activity is especially valuable for rules with boundaries, because it shows both where the rule works and where another idea must take over.
A useful Version C teaches sufficiency. Remove a value, condition, referent, observation or comparison that the answer requires, but keep enough context for the question to look tempting. Then state exactly what can still be concluded and exactly what cannot. If additional information would make the question answerable, name it. The goal is not to refuse every difficult question. The goal is to distinguish a demanding inference from an impossible one.
Accuracy before speed, then speed with evidence
Do not time the first stage of repair. Work slowly enough to carry out the full routine and produce proof-quality answers. Record which step feels effortful. After three accurate sessions, begin timing small mixed sets, but preserve a separate checking interval. This order matters: speed is repeated decision-making, so practising fast guesses makes the wrong decision more automatic. Practising accurate decisions first gives speed a reliable path to follow.
Track two measures rather than one. The first is answer accuracy. The second is decision accuracy: did you identify the correct target, relationship, controlled variable, evidence gap or boundary even if a later calculation or spelling slip occurred? A learner whose decision accuracy is improving is building a durable method. A learner who gets occasional answers right while decision accuracy remains low is still depending on recognition and luck. Use the two measures to choose the next practice task.
During a timed paper, use a two-pass system. On the first pass, solve questions whose task and evidence chain are clear. Mark uncertain items with a small symbol and move on before one question consumes disproportionate time. On the second pass, diagnose the uncertainty: missing knowledge, missing information, competing interpretations, or an unchecked calculation. Give each type a specific response. This is more efficient than rereading the whole question repeatedly without knowing what you are looking for.
End every practice set with one retrieval prompt: write the central rule from memory, one example where it works, one near-miss where it does not, and one checking question you will ask next time. This four-part exit record takes only a few minutes, yet it converts completed work into a reusable mental tool. On the following day, reconstruct the record before looking at it. The delay shows whether the idea has become available independently or remains tied to the page.
Examination control
1. Split bundled claims. Check every assertion joined by and, because or therefore.
2. Test local support. Match each claim to evidence, operation, definition or unit.
3. Trace dependencies. Follow reused values, references and premises.
4. Repair upstream first. Correct the earliest affected node before recomputing dependants.
5. Preserve independent work. Keep claims that still stand on valid evidence.
6. Reread the active response. Ensure no stale claim or duplicate final answer remains.
A strong examination routine is deliberately modest. It does not ask you to become brilliant in the final minute. It asks you to expose one hidden assumption, one missing unit, one unsupported word or one untested condition before it costs a mark. When time is short, check the exact sentence that carries the conclusion.
A seven-day repair cycle
Day 1 — Diagnose
Complete five short items and mark the exact point at which your reasoning became too broad. Do not merely write “careless”. Name the missing distinction.
Day 2 — Contrast
Place a correct and an almost-correct answer side by side. Underline the word, number or condition that separates them.
Day 3 — Explain
Teach the central rule to an imaginary younger pupil without using the worked example. If you need the original wording, the rule is not yet portable.
Day 4 — Mix
Combine this topic with two unrelated question types. Mixed practice trains recognition because no heading tells you which method to use.
Day 5 — Delay
Return to three earlier items without notes. Reconstruct the routine and record which step was forgotten.
Day 6 — Time
Complete a short set under a calm time limit, then spend a separate minute on conclusion checks. Keep solving time and checking time distinct.
Day 7 — Transfer
Create one new example, one counterexample and one explanation. A learner who can generate all three usually understands the boundary of the rule.
A deeper claim-map and dependency-repair laboratory
Boundary map
Sort twelve cases about claim-level checking and dependency repair into correct, incorrect, conditionally correct and undecidable. For each borderline case, name the exact word, value or condition that controls the decision.
Teach-back
Close the guide and explain claim-level checking and dependency repair with a rule, mechanism, worked example, counterexample and independent check. Reopen only after finishing and repair the earliest missing link.
Minimal pairs
Create six pairs about claim-level checking and dependency repair. Change one decisive feature in each pair, then explain why the conclusion changes or remains stable while every other relevant feature is held fixed.
Error classes
Classify eight wrong responses about claim-level checking and dependency repair as knowledge, interpretation, execution, evidence, recording or checking errors. Give each error class a different repair and a fresh retest.
Delayed reconstruction
After forty-eight hours, reconstruct the central rule, routine, three examples and two boundaries for claim-level checking and dependency repair. Compare only after writing and practise the first decision you could not recall.
Mixed retrieval
Hide three claim-level checking and dependency repair questions among six unrelated English, Mathematics and Science items. Remove headings, delay the attempt and record the cue that correctly selected the method.
Claim audit
Rewrite five explanations about claim-level checking and dependency repair as claim, evidence and bridge. Pressure-test strong words such as always, only, impossible, proves and complete before accepting the final sentence.
Alternative cases
For three correct answers about claim-level checking and dependency repair, create a surface change, a condition change and a missing-information version. State which conclusions persist, reverse or become undecidable.
Confidence calibration
Attempt twelve prompts about claim-level checking and dependency repair and rate confidence before checking. Repair confident-wrong responses first, because they reveal a shortcut that currently feels trustworthy.
Exam simulation
Place one claim-level checking and dependency repair item near the beginning, middle and end of a mixed page. Compare selection, execution, checking and recording under untimed and calm timed conditions.
Adversarial example
Ask a partner to invent a misleading claim-level checking and dependency repair case. Reject the surface clue, name the controlling evidence and state the smallest change that would make the opposite answer correct.
Transfer log
Keep a one-week log of every claim-level checking and dependency repair encounter: cue, method, check and error. At week end, locate the earliest repeated failure and build a targeted contrast that tests it.
Topic laboratory 1
Take four responses from different subjects and split every sentence into atomic claims. Give each claim its own evidence line, then mark supported, unsupported, false or undecidable without assigning one label to the whole paragraph.
Topic laboratory 2
Draw dependency graphs for two multi-step calculations. Introduce one input error at a time and colour only the nodes that must change. Compare this with a separate subpart that uses the same diagram but not the wrong value.
Topic laboratory 3
Create English and Science examples containing a correct observation, a plausible interpretation and an overstrong conclusion. Practise preserving the observation while calibrating or replacing the later claims.
Topic laboratory 4
Run a five-minute correction drill. Locate the earliest error, trace genuine dependants, repair them and stop. Afterwards, count unnecessary changes and missed stale claims to measure repair precision rather than editing volume.
For parents and tutors
When a child finds one mistake, ask “What used this?” before suggesting a full restart.
Praise preservation of independently correct work as well as accurate correction.
Use claim maps with arrows to make hidden dependencies visible.
Separate answer accuracy, method accuracy and recording accuracy during review.
Frequently asked questions
Should I ever restart the whole solution?
Yes, if the earliest premise controls everything or the structure is too tangled, but decide from dependencies rather than panic.
Can a conclusion be true when its reason is wrong?
Yes. The conclusion then needs a different valid reason.
Can a correct method produce a wrong answer?
Yes, when input, execution or recording is wrong.
Can a wrong explanation accompany a correct observation?
Yes. Preserve the observation and repair the mechanism.
How do I know whether two subparts are independent?
Ask whether one answer, value or premise is required to produce the other.
What is the fastest check?
Draw arrows from the corrected item and change only the claims those arrows reach.
Next routes
Return to the complete volume directory to choose another precise learner problem, or revisit the original PSLE Learning Guide for the earlier foundation. For subject-specific support, continue through the verified PSLE English route on eduKateSG, the verified PSLE Mathematics route on eduKateSG, the verified PSLE Science route on eduKateSG. These routes are for extending the mechanism, not for replacing your school’s instructions or the official examination document for your year.
Final reminder: An error has a scope: find it, trace it, repair it, and preserve every independent claim that still earns its place.