How to answer inference questions is to move beyond what is stated directly without moving beyond what the evidence can support. Strong inference combines a specific clue with relevant context or background knowledge to produce the most defensible conclusion.
Students often confuse inference with guessing. The difference is evidence. A valid inference is not printed word-for-word, but it can be justified from details in the text, source, graph, image or situation.
This guide explains how to identify inference questions, distinguish evidence from assumption, use clue-plus-reasoning structures, compare competing inferences and avoid overclaiming.
This guide is written for students, parents and educators who want a practical answer to how to answer inference questions in reading and exams. The aim is to turn the command word into a visible operating system: decode the task, select the right evidence or mechanism, produce the answer at the right depth, check it against the question and then practise again on a changed prompt.
The 60-Second Answer
Read the exact inference question. Find the most relevant clue or clues. Ask what those details suggest, not what they could possibly mean. Add only the background knowledge needed to connect clue to conclusion. State the inference clearly and support it with evidence. If two inferences are possible, choose the one best supported by the source and scope.
Working definition: An inference is a conclusion that is not stated explicitly but is supported by available evidence combined with appropriate reasoning or contextual knowledge.
Why Inference Questions Matters
Inference is central to reading because writers, speakers and data displays do not state every relationship directly.
The skill requires restraint: the learner must go beyond literal copying but stop before speculation.
Evidence-based inference also supports Science, Humanities and data interpretation because conclusions often depend on patterns rather than explicit statements.
The useful standard is not whether the learner has memorised an answer frame. It is whether the learner can recognise the command in an unfamiliar question and build a response that satisfies it without wasting time or drifting into a different task.
The Hidden Problem: Inference Is More Than Retrieval but Less Than Imagination
A literal answer copies the clue without interpreting it.
A speculative answer adds an unsupported story.
A strong inference sits between them: it states the conclusion that best explains the evidence available.
The Operating Model
Decode → Locate Clue → State What It Suggests → Add Context → Infer → Support → Check Alternatives → Qualify.
Each stage has a different job. Interpretation identifies the required thinking. Selection chooses the relevant mechanism, evidence or criterion. Construction turns that thinking into a markable response. Checking verifies scope, completeness and precision. Retesting shows whether the routine transfers beyond one familiar question.
How to answer inference questions: Step by Step
1. Identify the inference target
Clarify whether the question asks about motive, attitude, relationship, cause, likely outcome or unstated meaning.
Do not infer beyond the requested scope.
2. Locate the strongest clue
Find the detail, phrase, behaviour, pattern or evidence most relevant to the target.
Avoid collecting every possible clue.
3. Translate the clue
State what the clue suggests in plain language.
This prevents copying from masquerading as inference.
4. Add necessary context
Use only the background knowledge required to make the connection.
Do not build a large story.
5. State the inference
Give the conclusion directly and at an appropriate level of certainty.
Avoid absolute language if evidence is limited.
6. Support with evidence
Quote, paraphrase or reference the clue precisely.
Show the reader why the inference follows.
7. Check alternatives
Ask whether another explanation fits the evidence better.
Choose the most supported inference.
8. Stay within the text or data
Do not import unsupported motives, facts or assumptions.
The inference must remain anchored.
What Strong Performance Looks Like
- The target is clear.
- A strong clue is selected.
- The answer goes beyond copying.
- The inference remains evidence-based.
- Context is limited and relevant.
- Alternative explanations are considered.
- Certainty matches evidence.
- Support is explicit.
These are observable behaviours, so they can be modelled, practised and corrected. The learner does not need a special exam instinct; the learner needs reliable decisions that become faster with repetition.
Common Failure Modes
1. Copying only
The answer repeats the source.
Explain what the detail suggests.
2. Guessing
The learner invents a possibility with no evidence.
Anchor every inference to a clue.
3. Too many clues
The response lists evidence without a conclusion.
State the inference clearly.
4. Overclaiming motive
A behaviour is treated as proof of an internal state.
Use cautious language where appropriate.
5. Outside knowledge dump
Background information overwhelms the source.
Use only what links clue to conclusion.
6. First inference wins
The first plausible idea is accepted.
Compare alternatives.
7. Evidence mismatch
The chosen quotation does not actually support the inference.
Select the strongest clue.
8. Scope drift
The learner infers beyond the person, paragraph, period or data requested.
Keep within the question boundary.
How the Strategy Changes Across Subjects
English comprehension
Infer character feelings, writer attitude, relationships, unstated motives or implied meaning from textual clues.
The subject standard remains decisive. General exam technique can organise the answer, but it cannot replace the discipline-specific knowledge needed to make the reasoning correct.
Literature
Use language, action, dialogue and narrative context to support interpretive inferences while avoiding unsupported psychology.
The subject standard remains decisive. General exam technique can organise the answer, but it cannot replace the discipline-specific knowledge needed to make the reasoning correct.
History and source work
Infer purpose, attitude or situation from source content and provenance, then cross-check with context.
The subject standard remains decisive. General exam technique can organise the answer, but it cannot replace the discipline-specific knowledge needed to make the reasoning correct.
Science and data
Infer trends, relationships or likely explanations from evidence while distinguishing inference from proof of causation.
The subject standard remains decisive. General exam technique can organise the answer, but it cannot replace the discipline-specific knowledge needed to make the reasoning correct.
Primary School, Secondary School and Examination Years
Primary school
Teach clue plus what it suggests. Use one or two clear details and make the evidence visible before asking for deeper interpretation.
Keep the structure simple enough that the child can remember it independently. Adults can model the sequence first, then fade prompts as accuracy grows.
Secondary school
Students should practise competing inferences, tone, motive, source purpose and data-based conclusions with explicit evidence.
Secondary learners should increasingly decide response depth, evidence selection and scope without relying on a teacher to signal what belongs in the answer.
Before major examinations
Use unseen passages, sources and data so inference depends on transfer rather than memorized content. Mark both the conclusion and evidence link.
Near major exams, practise the command word in mixed sets so the learner must identify the required response type rather than seeing ten identical questions in a row.
A Focused 60-Minute Practice Session
- 10 minutes — classify literal versus inference questions.
- 10 minutes — underline the strongest clue in five prompts.
- 15 minutes — write clue-plus-inference answers.
- 10 minutes — compare two plausible interpretations.
- 10 minutes — practise qualified language.
- 5 minutes — record one recurring inference error.
The exact minutes can change. What matters is that the student experiences decoding, independent construction, feedback, targeted repair and a fresh retest.
Diagnostic Checklist
- Can the learner identify the target?
- Can they locate a strong clue?
- Can they explain what it suggests?
- Is background knowledge limited?
- Is the conclusion supported?
- Are alternatives considered?
- Does certainty match evidence?
- Does the answer stay within scope?
Use the checklist to identify the first decision that failed. A weak explanation may actually begin with a missing command word; a weak evaluation may begin with no criterion; a weak inference may begin with evidence that was never read closely.
Practice Laboratory
Practice 1: Clue-to-inference
Pair each clue with the most defensible conclusion.
Reject tempting unsupported alternatives.
Practice 2: Literal-to-inference repair
Take a copied sentence and add what it suggests.
Keep the new claim evidence-based.
Practice 3: Evidence ranking
Choose the strongest of three clues for one inference.
Explain why.
Practice 4: Alternative explanation
Generate two possible inferences and decide which has better support.
Use explicit evidence.
Practice 5: Certainty ladder
Rewrite definitely as likely, suggests or may indicate where evidence is limited.
Match language to support.
Practice 6: Motive boundary
Infer motive from behaviour, then list what remains unknown.
Avoid pretending internal states are directly observed.
Practice 7: Source inference
Use a historical or media source to infer audience or purpose.
Connect provenance and content.
Practice 8: Fresh-passage transfer
Answer the same inference type on an unfamiliar text.
Check whether the method transfers.
Three Student Cases
Case 1: Maren copies the clue
Maren identifies the right sentence but repeats it as the answer.
Her tutor asks what the sentence suggests that is not written directly.
Her responses begin making genuine inferences.
Case 2: Iona invents stories
Iona is imaginative and creates plausible motives not supported by the text.
She adopts a rule that every inference must point back to a clue.
Creativity becomes constrained by evidence.
Case 3: Leonie overstates certainty
Leonie writes definitely even when the passage only suggests a possibility.
She practises certainty language tied to evidence strength.
Her answers become more defensible.
How Parents Can Help Without Taking Over
Parents can support inference questions by asking for the learner’s reasoning rather than supplying the answer. Useful prompts include: “What is the command word?”, “What does the question actually want?”, “Which evidence matters?”, “What links this point to the result?”, “What criterion are you using?”, and “How will you know the answer is complete?”
When the learner can perform the decision reliably, remove the prompt. The goal is independent control under exam conditions.
How Tutors Can Use a Three-Student Small Group
In a three-student tutorial, inference questions becomes visible because one learner can answer, another can challenge the reasoning, and the third can check the response against the command and mark allocation. Roles then rotate.
After discussion, every learner completes a fresh item independently so group explanation becomes personal performance.
How to Measure Progress
- Literal copying decreases.
- Evidence links become explicit.
- Unsupported speculation falls.
- Strong clues are selected faster.
- Certainty becomes calibrated.
- Alternative explanations are considered.
- Inference transfers to unseen material.
- Marks on higher-order reading questions improve.
Marks are the final indicator, but process changes often appear first: more direct openings, fewer irrelevant sentences, stronger evidence links, clearer criteria, fewer missing steps and faster recognition of command words.
A Four-Week Implementation Plan
Week 1 — Build clue-plus-inference
Use short passages and one clear inference target.
Make evidence explicit.
Week 2 — Add alternative interpretations
Compare plausible conclusions and rank support.
Use certainty language.
Week 3 — Vary sources
Practise prose, images, historical sources and data.
Keep the same evidence logic.
Week 4 — Simulate
Use unseen exam-style material under time.
Review whether errors began with clue selection or reasoning.
Evidence and Responsible Use
Princeton’s active-reading guidance encourages readers to question, infer and connect evidence rather than read passively, while Cornell’s exam guidance stresses decoding question type before answering. Inference questions therefore require a deliberate shift from literal retrieval to evidence-supported interpretation.
Command-word routines should be adapted to the actual subject, school and examination board. Words such as explain, evaluate, compare and infer have common patterns, but local mark schemes and curriculum definitions remain the final standard for the assessment.
- Princeton: Active Reading Strategies
- Cornell: How to Tackle Exam Questions
- Princeton: Preparing for Exams
Frequently Asked Questions
What is an inference?
A conclusion supported by evidence even though it is not stated directly.
How is inference different from guessing?
Inference has specific evidence and reasoning; guessing may not.
Should I quote the text?
Use brief relevant evidence where the format expects support.
Can there be more than one valid inference?
Yes, but some are better supported than others.
How certain should I sound?
Match certainty to the strength of evidence.
Can I use background knowledge?
Only where it helps connect the clue to a defensible conclusion and the task permits it.
What if I cannot find a clue?
Recheck the question target and scan for behaviour, word choice, contrast, pattern or context directly related to it.
How do I practise?
Use unseen material and require every inference to cite its evidence.
Helpful Reading on eduKateSingapore
- How to Read Exam Questions
- How to Answer Short-Answer Exam Questions
- How to Use Mark Schemes Properly
- How to Check Your Work
- How to Answer Source-Based Questions
- How to Answer Data Interpretation Questions
Inference Lives Between Copying and Guessing
The answer should say more than the source states, but never more than the evidence can support.
Find the clue. State what it suggests. Add only necessary context. Compare alternatives. Then make the strongest supported inference.
Properly taught kids shine a bright light into the future.
Extended Diagnostic Workshops
Workshop 1: Identify the inference target × Guessing
Set up one fresh exam-style question where the learner must practise identify the inference target while watching specifically for guessing. Clarify whether the question asks about motive, attitude, relationship, cause, likely outcome or unstated meaning. Require an independent response before any model is shown so the controlling decision becomes visible.
The learner invents a possibility with no evidence. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Choose the strongest of three clues for one inference. Explain why. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 2: Locate the strongest clue × Outside knowledge dump
Set up one fresh exam-style question where the learner must practise locate the strongest clue while watching specifically for outside knowledge dump. Find the detail, phrase, behaviour, pattern or evidence most relevant to the target. Require an independent response before any model is shown so the controlling decision becomes visible.
Background information overwhelms the source. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Rewrite definitely as likely, suggests or may indicate where evidence is limited. Match language to support. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 3: Translate the clue × Scope drift
Set up one fresh exam-style question where the learner must practise translate the clue while watching specifically for scope drift. State what the clue suggests in plain language. Require an independent response before any model is shown so the controlling decision becomes visible.
The learner infers beyond the person, paragraph, period or data requested. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Use a historical or media source to infer audience or purpose. Connect provenance and content. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 4: Add necessary context × Too many clues
Set up one fresh exam-style question where the learner must practise add necessary context while watching specifically for too many clues. Use only the background knowledge required to make the connection. Require an independent response before any model is shown so the controlling decision becomes visible.
The response lists evidence without a conclusion. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Pair each clue with the most defensible conclusion. Reject tempting unsupported alternatives. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 5: State the inference × First inference wins
Set up one fresh exam-style question where the learner must practise state the inference while watching specifically for first inference wins. Give the conclusion directly and at an appropriate level of certainty. Require an independent response before any model is shown so the controlling decision becomes visible.
The first plausible idea is accepted. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Choose the strongest of three clues for one inference. Explain why. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 6: Support with evidence × Copying only
Set up one fresh exam-style question where the learner must practise support with evidence while watching specifically for copying only. Quote, paraphrase or reference the clue precisely. Require an independent response before any model is shown so the controlling decision becomes visible.
The answer repeats the source. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Rewrite definitely as likely, suggests or may indicate where evidence is limited. Match language to support. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 7: Check alternatives × Overclaiming motive
Set up one fresh exam-style question where the learner must practise check alternatives while watching specifically for overclaiming motive. Ask whether another explanation fits the evidence better. Require an independent response before any model is shown so the controlling decision becomes visible.
A behaviour is treated as proof of an internal state. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Use a historical or media source to infer audience or purpose. Connect provenance and content. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 8: Stay within the text or data × Evidence mismatch
Set up one fresh exam-style question where the learner must practise stay within the text or data while watching specifically for evidence mismatch. Do not import unsupported motives, facts or assumptions. Require an independent response before any model is shown so the controlling decision becomes visible.
The chosen quotation does not actually support the inference. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Pair each clue with the most defensible conclusion. Reject tempting unsupported alternatives. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 9: Identify the inference target × Guessing
Set up one fresh exam-style question where the learner must practise identify the inference target while watching specifically for guessing. Clarify whether the question asks about motive, attitude, relationship, cause, likely outcome or unstated meaning. Require an independent response before any model is shown so the controlling decision becomes visible.
The learner invents a possibility with no evidence. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Choose the strongest of three clues for one inference. Explain why. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 10: Locate the strongest clue × Outside knowledge dump
Set up one fresh exam-style question where the learner must practise locate the strongest clue while watching specifically for outside knowledge dump. Find the detail, phrase, behaviour, pattern or evidence most relevant to the target. Require an independent response before any model is shown so the controlling decision becomes visible.
Background information overwhelms the source. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Rewrite definitely as likely, suggests or may indicate where evidence is limited. Match language to support. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 11: Translate the clue × Scope drift
Set up one fresh exam-style question where the learner must practise translate the clue while watching specifically for scope drift. State what the clue suggests in plain language. Require an independent response before any model is shown so the controlling decision becomes visible.
The learner infers beyond the person, paragraph, period or data requested. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Use a historical or media source to infer audience or purpose. Connect provenance and content. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 12: Add necessary context × Too many clues
Set up one fresh exam-style question where the learner must practise add necessary context while watching specifically for too many clues. Use only the background knowledge required to make the connection. Require an independent response before any model is shown so the controlling decision becomes visible.
The response lists evidence without a conclusion. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Pair each clue with the most defensible conclusion. Reject tempting unsupported alternatives. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 13: State the inference × First inference wins
Set up one fresh exam-style question where the learner must practise state the inference while watching specifically for first inference wins. Give the conclusion directly and at an appropriate level of certainty. Require an independent response before any model is shown so the controlling decision becomes visible.
The first plausible idea is accepted. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Choose the strongest of three clues for one inference. Explain why. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 14: Support with evidence × Copying only
Set up one fresh exam-style question where the learner must practise support with evidence while watching specifically for copying only. Quote, paraphrase or reference the clue precisely. Require an independent response before any model is shown so the controlling decision becomes visible.
The answer repeats the source. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Rewrite definitely as likely, suggests or may indicate where evidence is limited. Match language to support. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 15: Check alternatives × Overclaiming motive
Set up one fresh exam-style question where the learner must practise check alternatives while watching specifically for overclaiming motive. Ask whether another explanation fits the evidence better. Require an independent response before any model is shown so the controlling decision becomes visible.
A behaviour is treated as proof of an internal state. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Use a historical or media source to infer audience or purpose. Connect provenance and content. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 16: Stay within the text or data × Evidence mismatch
Set up one fresh exam-style question where the learner must practise stay within the text or data while watching specifically for evidence mismatch. Do not import unsupported motives, facts or assumptions. Require an independent response before any model is shown so the controlling decision becomes visible.
The chosen quotation does not actually support the inference. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Pair each clue with the most defensible conclusion. Reject tempting unsupported alternatives. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 17: Identify the inference target × Guessing
Set up one fresh exam-style question where the learner must practise identify the inference target while watching specifically for guessing. Clarify whether the question asks about motive, attitude, relationship, cause, likely outcome or unstated meaning. Require an independent response before any model is shown so the controlling decision becomes visible.
The learner invents a possibility with no evidence. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Choose the strongest of three clues for one inference. Explain why. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 18: Locate the strongest clue × Outside knowledge dump
Set up one fresh exam-style question where the learner must practise locate the strongest clue while watching specifically for outside knowledge dump. Find the detail, phrase, behaviour, pattern or evidence most relevant to the target. Require an independent response before any model is shown so the controlling decision becomes visible.
Background information overwhelms the source. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Rewrite definitely as likely, suggests or may indicate where evidence is limited. Match language to support. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 19: Translate the clue × Scope drift
Set up one fresh exam-style question where the learner must practise translate the clue while watching specifically for scope drift. State what the clue suggests in plain language. Require an independent response before any model is shown so the controlling decision becomes visible.
The learner infers beyond the person, paragraph, period or data requested. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Use a historical or media source to infer audience or purpose. Connect provenance and content. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 20: Add necessary context × Too many clues
Set up one fresh exam-style question where the learner must practise add necessary context while watching specifically for too many clues. Use only the background knowledge required to make the connection. Require an independent response before any model is shown so the controlling decision becomes visible.
The response lists evidence without a conclusion. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Pair each clue with the most defensible conclusion. Reject tempting unsupported alternatives. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 21: State the inference × First inference wins
Set up one fresh exam-style question where the learner must practise state the inference while watching specifically for first inference wins. Give the conclusion directly and at an appropriate level of certainty. Require an independent response before any model is shown so the controlling decision becomes visible.
The first plausible idea is accepted. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Choose the strongest of three clues for one inference. Explain why. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 22: Support with evidence × Copying only
Set up one fresh exam-style question where the learner must practise support with evidence while watching specifically for copying only. Quote, paraphrase or reference the clue precisely. Require an independent response before any model is shown so the controlling decision becomes visible.
The answer repeats the source. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Rewrite definitely as likely, suggests or may indicate where evidence is limited. Match language to support. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 23: Check alternatives × Overclaiming motive
Set up one fresh exam-style question where the learner must practise check alternatives while watching specifically for overclaiming motive. Ask whether another explanation fits the evidence better. Require an independent response before any model is shown so the controlling decision becomes visible.
A behaviour is treated as proof of an internal state. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Use a historical or media source to infer audience or purpose. Connect provenance and content. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 24: Stay within the text or data × Evidence mismatch
Set up one fresh exam-style question where the learner must practise stay within the text or data while watching specifically for evidence mismatch. Do not import unsupported motives, facts or assumptions. Require an independent response before any model is shown so the controlling decision becomes visible.
The chosen quotation does not actually support the inference. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Pair each clue with the most defensible conclusion. Reject tempting unsupported alternatives. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.
Workshop 25: Identify the inference target × Guessing
Set up one fresh exam-style question where the learner must practise identify the inference target while watching specifically for guessing. Clarify whether the question asks about motive, attitude, relationship, cause, likely outcome or unstated meaning. Require an independent response before any model is shown so the controlling decision becomes visible.
The learner invents a possibility with no evidence. After correction, change one meaningful feature—topic, wording, evidence, mark allocation, data, source or context. Then use this related exercise: Choose the strongest of three clues for one inference. Explain why. Finish with a delayed retest so success cannot be explained only by immediate memory of the correction.