In a G2 SEC paper, not every decision deserves the same amount of checking. A choice that can be reversed in five seconds should not consume the same attention as a choice that determines an entire essay, graph, calculation route or structured explanation. The Decision-Reversibility Check teaches one practical question: if this choice is wrong, how expensive will it be to undo later?
Use it now. Before committing, name the decision, estimate the cost of changing it, and match your checking time to that cost. Check high-impact, hard-to-reverse choices early; make low-impact, easy-to-reverse choices promptly and keep moving.
The 2027 SEAB G2 school-candidate list identifies English Language K200, Mathematics K210 and Science K223/K224/K225. Their published formats differ, so this is a general decision-control method rather than a claim that every paper uses one template. Follow the instructions printed on the actual paper and your school’s current administrative directions.
Official reference route: SEAB 2027 G2 school-candidate syllabus list; MOE’s announced SEC transition context. The subject-specific official syllabus used for this chapter is the current 2027 official syllabus source. Check those pages again if you use this guide in a later examination year.
Choose your route
Route 1 — Immediate repair: Chapters 1–4 identify the mechanism and the first high-value correction.
Route 2 — Reliable execution: Chapters 5–9 turn the idea into a repeatable working routine.
Route 3 — Practice and diagnosis: Chapters 10–13 expose common traps and build variation.
Route 4 — Support and transfer: Chapters 14–15 guide parents, tutors and independent use.
Chapter contents
1. Decision reversibility measures repair cost, not question difficulty
2. Separate commitment points from ordinary steps
3. Use a ten-second reversibility scan
4. Match checks to downstream impact
5. Know when early checking has positive value
6. Use evidence, not anxiety, to classify risk
7. Build reversible drafts for uncertain work
9. Use a decision ledger during practice
10. Calibrate with original mixed-subject drills
12. Avoid over-checking reversible choices
13. Measure improvement by avoided rework
14. Parent and tutor guidance should reveal decisions
15. Transfer the check under realistic pressure
Series and learning links
Continue with eduKateSG Examinations and Assessment Hub, the canonical owner How Exam Time Management Works, and Vol 0284 Task-Re-entry State Snapshot. This chapter adds a narrower diagnostic and practice job rather than replacing those broader owners.
For wider examination habits, use Examination Craft and the earlier PSLE Learner’s Guide series example. Teaching depth and progression are benchmarked against the immutable eduKateSG small-group tutorial reference without copying its wording.
1. Decision reversibility measures repair cost, not question difficulty
Mechanism. Decision reversibility measures repair cost, not question difficulty. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. A learner choosing between two essay positions spends one minute testing which position can support three distinct points; changing after two pages would be costly. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. A learner spends three minutes deciding whether to write 6 cm or 6.0 cm even though the value is easy to correct. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. Label the first decision high-cost and the second low-cost. The label directs checking effort without predicting marks. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
2. Separate commitment points from ordinary steps
Mechanism. Separate commitment points from ordinary steps. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. In Mathematics, choosing an equation or representation may shape several later lines, while copying a final unit is locally editable. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. Students often treat every written mark as equally permanent and reread small details while the main model remains untested. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. Circle the first step that commits the solution path, then test that step against the givens before expanding the working. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
3. Use a ten-second reversibility scan
Mechanism. Use a ten-second reversibility scan. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. Before writing a long response, ask what would have to be erased, reorganised or recalculated if the choice failed. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. A candidate begins a Science explanation with an effect, then discovers the question asked for the mechanism and rewrites the whole paragraph. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. Pause for ten seconds to identify cause, mechanism and effect. That small investment protects a much larger block of writing. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
4. Match checks to downstream impact
Mechanism. Match checks to downstream impact. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. A wrong axis scale can corrupt every plotted point; one untidy label may affect only one location. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. The common error is checking the neatness of each point while never validating the scale chosen for the graph. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. Check the scale, variable placement and origin decision before plotting the full data set, then use a local check for each point. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
5. Know when early checking has positive value
Mechanism. Know when early checking has positive value. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. A title choice, interpretation or base quantity deserves early checking when later work depends on it. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. A student believes checking should happen only at the end, so a misread restriction survives through an entire response. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. Use an early dependency check and reserve the final sweep for omissions, arithmetic, units and communication details. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
6. Use evidence, not anxiety, to classify risk
Mechanism. Use evidence, not anxiety, to classify risk. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. A decision is high-risk when many later steps depend on it or repair will consume scarce time, not merely because it feels unfamiliar. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. A familiar short question can hide a costly base error, while an unfamiliar one-mark item may be easy to revise. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. State the dependency: ‘If this is wrong, these four lines change.’ If no downstream consequence can be named, reduce the checking time. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
7. Build reversible drafts for uncertain work
Mechanism. Build reversible drafts for uncertain work. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. A compact plan, labelled diagram or symbolic equation preserves flexibility before full sentences or arithmetic are committed. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. A learner writes a polished English paragraph before testing whether its example actually answers the prompt. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. Draft the claim and evidence in keywords, test their connection, then convert the stable plan into prose. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
8. Control switching costs
Mechanism. Control switching costs. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. Leaving a task is sensible only when the return state is recorded well enough to resume without rebuilding the reasoning. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. A student skips a calculation but writes only a star, then rereads the entire question later. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. Record known values, the chosen method and the unresolved step. This makes the skip reversible and keeps its repair cost bounded. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
9. Use a decision ledger during practice
Mechanism. Use a decision ledger during practice. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. A ledger records the choice, predicted repair cost, actual outcome and better checkpoint. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. After a timed paper, the learner remembers only wrong answers and cannot see which early decisions caused clusters of errors. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. Review causal decisions, not just final marks. One repaired scale choice may prevent several future plotting errors. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
10. Calibrate with original mixed-subject drills
Mechanism. Calibrate with original mixed-subject drills. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. Use one planning choice, one representation choice and one interpretation choice in a short set. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. Practising only one subject can make the method feel like subject knowledge rather than a transferable control skill. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. Alternate English, Mathematics and Science examples, but keep the same reversibility question and compare the repair costs. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
11. Protect the final minutes
Mechanism. Protect the final minutes. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. Late in the paper, prefer checks that can expose large omissions or high-impact dependencies quickly. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. A candidate recomputes a secure one-mark answer while a whole subpart remains blank. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. Scan unanswered parts, page turns and major dependencies first; perform cosmetic or low-impact revisions only if time remains. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
12. Avoid over-checking reversible choices
Mechanism. Avoid over-checking reversible choices. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. The method must reduce wasted checking as well as prevent expensive errors. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. A student repeatedly changes a synonym in an English response even though either word is accurate and clear. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. Set a stop rule: once a low-cost choice is correct, relevant and readable, move to a decision with greater downstream value. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
13. Measure improvement by avoided rework
Mechanism. Measure improvement by avoided rework. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. Faster completion is useful only when it comes from fewer destructive restarts, not careless rushing. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. A learner finishes earlier but makes the same base-selection error twice. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. Compare how many lines, paragraphs or graph points needed rebuilding. A falling rework count shows better early decisions. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
14. Parent and tutor guidance should reveal decisions
Mechanism. Parent and tutor guidance should reveal decisions. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. Adults should ask what the learner checked before committing, rather than supplying the choice. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. A tutor silently corrects a plan, producing a clean answer but no independent checkpoint. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. Use prompts such as ‘What becomes expensive to change after this?’ and fade the prompt when the learner begins asking it independently. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
15. Transfer the check under realistic pressure
Mechanism. Transfer the check under realistic pressure. This matters because a visible decision can be checked and corrected, while an unspoken habit can produce a plausible answer for the wrong reason. Name the decision in plain language before compressing it into a fast examination routine.
Original worked example. Transfer means selecting the checkpoint without a label, across new content and with a time limit. The values and wording here are invented for learning. Work through the relationship, show the decisive step and then test whether the final statement answers exactly what was asked.
Misconception diagnosis. The learner performs the routine only on worksheets that announce ‘reversibility check’. This response is useful evidence: it shows which decision failed, not merely that the final answer differs. Mark the first unstable step so the learner does not waste time rewriting everything that followed correctly from it.
Repair. Use a fresh mixed mini-paper, require a brief decision mark, and ask for one post-task explanation of why that decision deserved early checking. After the correction, close the model and reproduce the step from memory. Then change one surface feature—number, wording, graph direction, speaker or context—so the repair becomes a rule the learner owns rather than a copied answer.
Practice progression. Begin untimed with the key decision labelled. Next remove the label, vary the surface details and add one nearby distractor. Then use a short realistic time boundary. Record whether the learner selected the method independently, carried it through accurately and stopped when the task was complete.
Explained check. A secure response preserves the governing relationship, the relevant detail and any required unit or language constraint. If two methods appear possible, compare their assumptions against the prompt instead of choosing the one that looks most familiar. The explanation should make the decision auditable in one or two sentences.
Transfer. Give a fresh item at least a day later without naming this chapter. Success means the learner notices the same structure under different wording, completes the task and can explain why the move fits. If the skill disappears, return to the earliest failed decision rather than adding more identical questions.
A seven-day practice plan
Day 1: learn the mechanism and complete two guided examples.
Day 2: diagnose three deliberately flawed responses.
Day 3: complete a varied set without chapter labels.
Day 4: explain one solution aloud and shorten the explanation without losing the reason.
Day 5: use a timed mini-set and record the first failed decision.
Day 6: repair that decision, then attempt a changed transfer item.
Day 7: complete a delayed mixed check and write one next-step target.
Final learner checklist
□ I can state the task before I start.
□ I can identify the decision that controls the later steps.
□ I can show enough working or notes to make that decision visible.
□ I can diagnose the first wrong step instead of only comparing final answers.
□ I can complete a changed example without copying the model.
□ I follow the current paper and school instructions rather than treating this guide as an examiner rule.