Practice Drill: Reconstruct Understanding Across Representations
A no-notes drill for extracting structure, translating representations, testing an analogy, and locating a model boundary.
Chinese version: /zh-hant/practice/drills/reconstruct-understanding-across-representations/
Subskill
Demonstrate understanding of a causal claim by preserving its core relation across representations and testing its boundary.
Performance condition
Without notes, complete the six responses in 12 minutes.
Claim: “After a school replaced part of its textbook homework with short instructional videos, average quiz scores rose. The principal concludes that the videos caused the improvement.”
Required response
- Extract the structural skeleton: outcome, proposed cause, relevant actors, at least two possible conditions or competing variables.
- Write one conditional causal relation that is more precise than the principal’s conclusion.
- Translate that relation into a text-only causal diagram using arrows and labels.
- Give one analogy to a familiar situation. State the shared relation and one important place where the analogy breaks.
- Name one changed case or counterexample and predict what the revised model says will happen.
- State one item of evidence that would make you revise the model again.
Success criteria
- Separates observation from causal conclusion.
- Identifies a mechanism or condition, not only a list of factors.
- Preserves the same conditional relation in prose and the causal diagram.
- Uses an analogy with both a valid structural match and a stated limit.
- Gives a prediction that follows from the revised model.
- Names evidence that is capable of changing the conclusion.
Common error
Changing “videos caused improvement” to “videos never help.” That reverses an overgeneralisation instead of identifying the conditions and competing explanations.
Constraint
Use no external resources before writing. After feedback, repeat in two days with a communication case and explicitly separate wording from inferred intent.
Next action
- If successful: Save the unedited response as an attempt and repeat with a communication or procedural case.
- If unsuccessful: Identify the earliest missing component: structure, conditional relation, representation translation, analogy limit, boundary prediction, or revision evidence. Retry only that component with a simpler claim.