The analogy is not the idea.
Break it anyway.
Every useful analogy is a temporary scaffold. Map a hard idea onto something you already know, restate the real structure without the story, then find where the map lies, and what wrong belief that lie would plant.
1
Why generate
A structure-preserving map lets you inherit inferences from a domain you already trust. Surface vibes don't.
2
Why restate
If you can only speak through the metaphor, you learned the story, not the target. Force the real vocabulary.
3
Why break it
Every analogy lies somewhere. Naming the break stops a fluent misconception from hardening.
4
Why transfer
A second mapping, then a problem in new dress, proves you kept the relation, not the costume.
Worked example: natural selection
INTRO BIO · BORROWED FROM A SIEVE
I. BORROW
Variation in a population is like mixed gravel poured into a sieve. The mesh (environment) lets some sizes through and holds others back. What passes is what 'fits.'
II. RESTATE, NO SIEVE
Individuals vary. Some variants leave more offspring in this environment. Those traits become more common across generations, if they are heritable.
III. WHERE IT BREAKS
A sieve is passive and one-shot; selection is repeated across generations. Gravel doesn't reproduce. The mesh doesn't change because of what it filtered; environments and populations do.
THE WRONG BELIEF IT PLANTS
That nature 'chooses' individuals the way a person sorts stones (intentional, tidy, finished) instead of differential reproductive success over time.
IV. SECOND ANALOGY
A filter in a river: continuous flow, repeated sorting, debris that seeds the next pass. Still imperfect, but it fails in different places than the kitchen sieve.
V. DROP THE SCAFFOLD
Explain selection using variation, heritability, and differential reproductive success only. If you reach for gravel, you are not done.
Getting a better one
- Reject the cliché. Security-team and filing-cabinet defaults are usually surface. Ask for three structural maps with a two-column table.
- Rank by causal roles. Keep the map where causes and constraints align, not the one that sounds cleverest.
- Aim it at your priors. Name the familiar domain you actually know. A5 works best when the source is yours.
Breaking it properly
- Ask where, not whether. Every analogy breaks. The lesson is the boundary.
- Name the wrong prediction. For each break: what error would a fluent student make under pressure? (A6)
- Mark what it hides. Omissions matter more than small inaccuracies. What does the story make invisible?
Making it stick
- Restate cold. Close the chat and write the target with no metaphor. Gaps are the curriculum.
- Transfer once. One far problem that needs the same relation. If you can't, you kept the costume. (F3)
- Drop on purpose. Decide the scaffold has done its job. Leave with the structure, not the story.
Prompt · Generate + map
Give me three analogies for [TOPIC] aimed at someone who knows [FAMILIAR DOMAIN]. For each: a two-column mapping table (analogy part → real part). Prefer structural similarity over vividness. Rank them by how well causal roles match. If one is weak, say so.Prompt · Stress-test
I am using this analogy for [TOPIC]: [ANALOGY]. (1) Ask me to restate the target idea without using the analogy. (2) Tell me three places the analogy breaks and what misconception each could plant. (3) Give a second analogy that fails differently. (4) Give one transfer problem that needs the same relation in new dress.