G2 SEC examination decision-making improves when learners choose the next check for its information gain. When several answers, explanations or methods remain plausible, the best next step is the one that most sharply separates them—not the one that merely feels familiar.
This one-hundred-and-fifty-sixth Learner’s Guide develops information-gain control across K200 English, K210 Mathematics and K223–K225 Science. It builds on Vol 0152 Counterfactual Swap Tests and Vol 0144 Falsification Checks.
Mechanism: good checks make candidates disagree
Suppose two answers remain possible. A high-information check is one for which those candidates predict different results. A low-information check is one both candidates can pass. The mechanism is discrimination: identify the live candidates, identify their different predictions, and observe the cheapest discriminating feature.
Diagnosis: define the uncertainty before checking
Vague uncertainty produces vague review. Name the unresolved distinction: source, denominator, sign, domain, mechanism, control, timing, scope or representation. Once named, the uncertainty itself tells you which evidence could reduce it.
Smallest repair
Choose the smallest check capable of separating the candidates. If it resolves the issue, repair only the branch that depended on the losing candidate. Do not rebuild independent work whose assumptions were never in doubt.
Observation, inference and competing explanations
Keep the observed result separate from what you infer from it. A check is informative only relative to competing explanations. If two explanations predict the same observation, the observation may support both but discriminate neither. Ask what result would make their predictions diverge.
English reading and source control
1. Two referents remain
High-information move: Ask which nearby clause or sentence makes the candidates predict different meanings. Re-read only that window and substitute each noun explicitly.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
2. Two motives remain
High-information move: Use chronology and knowledge-state evidence. A motive requiring information the character did not yet possess loses immediately.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
3. Two source owners remain
High-information move: Return to quotation marks, reporting clauses and paragraph boundaries. Source ownership often resolves multiple later inferences at once.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
4. Two paraphrases remain
High-information move: Test quantifier, modality, causality and scope. One small mismatch can eliminate a fluent but unfaithful paraphrase.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
5. Two attitudes remain
High-information move: Find the strongest evaluative phrase or final qualification. Prefer evidence that one attitude explains and the other cannot.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
6. Two paragraph functions remain
High-information move: Ask what changes if the paragraph is removed. If the text loses evidence, transition, qualification or conclusion, that function is more strongly supported.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
7. Two MCQ options remain
High-information move: Identify the single word or condition that makes them non-equivalent, then search for evidence about that distinction rather than rereading broadly.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
8. Unknown vocabulary
High-information move: Use grammar, contrast, cause, examples and lexical-chain context to constrain meaning. Do not spend time comparing dictionary senses the sentence cannot support.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
9. Ambiguous negation
High-information move: Write both possible scopes in plain English. The reading that preserves surrounding logic is the higher-information candidate.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
10. Ambiguous reporting verb
High-information move: Compare the verb with neutral ‘says’. Ask whether the writer signals evidence, doubt, reluctance, opposition or confirmation.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
Mathematics structure and verification
11. Two formulas seem possible
High-information move: Use dimensions and variable meanings before numbers. Units often eliminate a whole branch more cheaply than calculation.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
12. Two roots remain
High-information move: Substitute both into the original equation and contextual domain. This directly tests whether each candidate survives the problem, not only the transformed algebra.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
13. Two rate directions remain
High-information move: Write each unit in words. Dollars per item and items per dollar answer different questions even when both calculations look tidy.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
14. Two percentage bases remain
High-information move: Ask ‘percentage of what?’ and name the denominator. The correct reference quantity usually settles the comparison.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
15. Two graph readings remain
High-information move: Check one labelled interval or axis unit. A single scale decision can change every later coordinate and slope.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
16. Two theorems seem usable
High-information move: List the condition each theorem requires. The missing right angle, parallel line or correspondence often eliminates one route.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
17. Two probability universes remain
High-information move: Name one outcome included in one universe but not the other. Universe ownership has higher information value than calculating both fractions.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
18. Two integer answers surround a decimal
High-information move: Use minimum, maximum, capacity or count wording. Constraint language determines ceiling, floor or nearest-integer interpretation.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
19. Two normalisations remain
High-information move: Write both denominators in words and ask which one matches the comparison requested. The base, not the arithmetic, controls meaning.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
20. Candidate answer feels suspicious
High-information move: Use the cheapest independent structural check: units, bounds, direction, conservation, symmetry or substitution. Pick the check whose failure would force a different answer.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
Science mechanisms and experimental evidence
21. Cause versus correlation
High-information move: Ask what manipulation or control would have to exist for a causal interpretation. Design evidence can be more informative than another descriptive observation.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
22. Two mechanisms remain
High-information move: Find a result they predict differently: time course, location, control response, boundary condition or secondary measurement.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
23. Two variable roles remain
High-information move: Trace what was deliberately changed, what was measured and what was kept comparable. Procedure ownership resolves the labels.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
24. Two control explanations remain
High-information move: Name the alternative cause each control is supposed to remove. The more specific explanation usually has greater information value.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
25. Negative result: absent or undetected
High-information move: Check sensitivity, duration, range and positive control. Measurement capability separates true absence from a weak test.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
26. Positive result: target or false positive
High-information move: Check specificity, blank and negative control. A signal is informative only if competing sources are constrained.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
27. Baseline drift or treatment effect
High-information move: Inspect a concurrent control over time. Shared movement suggests drift; divergence after treatment supports a treatment-specific effect.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
28. Lag or no effect
High-information move: Add a later measurement chosen from the expected mechanism timescale. An immediate repeat can have little information if both models predict the same early state.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
29. Path dependence or current-input effect
High-information move: Reset the system or use a fresh sample. Different outcomes after different histories become testable only when starting states are controlled.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
30. Interaction or simple effect
High-information move: Measure A at another level of B. One crossed condition can reveal whether A’s effect is conditional.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
31. Bottleneck or non-limiting factor
High-information move: Increase the suspected limiting factor while holding other conditions stable. No response shifts attention to another constraint or a measurement limit.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
32. Saturation or instrument ceiling
High-information move: Change detector range or use an independent measurement. Biological/physical plateaus and measurement ceilings predict different cross-checks.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
33. Spatial gradient or local anomaly
High-information move: Measure a neighbouring location and, if possible, the opposite side. A coherent spatial pattern contains more information than another repeat at one point.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
34. Transfer between compartments or local production
High-information move: Measure both compartments over time. Reciprocal or linked changes can separate transfer from independent production/consumption.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
35. Dynamic balance or inactivity
High-information move: Measure opposing flows, not only the stable stock. Equal non-zero flows reveal turnover hidden by a flat level.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
Review, repair and transfer
36. Several unresolved questions
High-information move: Choose the check that can repair the most descendants. A shared source or denominator error has greater information gain than a leaf-level typo.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
37. Time-limited review
High-information move: Rank checks by expected ability to change the answer multiplied by mark value, divided by time cost. Do not equate anxiety with information value.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
38. Low confidence with a hard constraint available
High-information move: Apply the hard constraint first. An impossibility check can resolve uncertainty faster than full re-solving.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
39. High confidence without a falsifier
High-information move: Ask what observation would force you to change the answer. If no such observation can be named, confidence may be under-tested.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
40. Writing-plan uncertainty
High-information move: Identify the paragraph function or task bullet that remains unfulfilled. Clarifying the missing function is often more useful than generating more content.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
41. Listening uncertainty
High-information move: Re-enter the audio stream. Future content may disambiguate; mentally replaying an unheard detail creates no new evidence.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
42. Uncertain model scope
High-information move: Test the nearest boundary or assumption. A model can be well supported inside its range and still fail outside it.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
43. Uncertain sample generalisation
High-information move: Inspect selection, location, time and population coverage. Sampling design can matter more than another calculated average.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
44. Uncertain anomaly
High-information move: Repeat the measurement and inspect method provenance. This distinguishes random error, method error and real exception more effectively than deleting the point.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
45. Uncertain completion
High-information move: Check command, requested count, source, scope, unit and final form. If every proof obligation is paid, further work may have near-zero information gain.
For practice, create two plausible candidates and state in advance what each predicts about the chosen check. Perform or reason through the check, remove only the candidates that genuinely fail, and keep any unresolved alternative visible.
Then transfer the method to a changed surface—different names, numbers, source, graph, apparatus or context. The learner should generate a new discriminating check from the structure rather than reproduce the original check from memory.
Information gain versus check cost
The most discriminating check is not automatically the best if it consumes most of the remaining paper. Examination control combines information value with cost. A one-minute source check that can repair three answers may dominate a five-minute re-solve of one low-mark item. A ten-second unit check can outrank a detailed derivation when the unit alone can eliminate the wrong formula.
A useful ranking rule is: prefer checks that are cheap, capable of changing the answer, and attached to high-value or high-dependency work. Avoid checks that merely repeat the original reasoning, confirm facts both candidates already predict, or polish a response whose proof obligations are already complete.
Model limits and omitted candidates
Information gain depends on the candidate set. A check can look decisive only because a relevant alternative was never considered. Keep plausible competing explanations alive long enough to ask whether the chosen test would distinguish them. This is especially important in Science mechanisms, English source interpretation and Mathematics model selection.
Delayed transfer and measurement
Return later with a mixed problem where the framework is not named. Measure success by whether the learner can independently identify the uncertainty, generate competing candidates, choose the cheapest high-information check, repair the dependent branch, state the remaining model limit and close the question without prompts.
Internal learning links
Use the Examination Craft hub, the Secondary English Learning Hub, the Mathematics Hub, the Science Hub, and the PSLE Learner’s Guide.
MOE and SEAB current framework
MOE states that Full Subject-Based Banding is fully implemented and that, from 2027, the Singapore-Cambridge Secondary Education Certificate (SEC) replaces the former N- and O-Level examinations, with students sitting subjects at their respective G1, G2 or G3 levels. See the official MOE Full SBB / SEC announcement. For current 2027 G2 school-candidate subject codes and syllabuses, use the official SEAB G2 syllabus directory, which lists K200 English Language, K210 Mathematics and K223–K225 Science combinations.
Final rule
When uncertainty remains, do not collect evidence at random. Name the surviving candidates, choose the cheapest check that makes them disagree, update the answer, and stop when the uncertainty that matters has genuinely collapsed.
G2 SEC Learner’s Guide: Open the Vol 0132–0175 hub and subject index.
