Analogical Mapping Prompt¶

Structured elicitation prompt — instantiates Transfer Scaffolding

Elicits an explicit correspondence map between one familiar source and one new target — what lines up, what has no counterpart, and which similarities are false friends — so an analogy carries structure rather than surface.

An analogy in the head is a hunch; Analogical Mapping Prompt forces it onto the table. It is a structured prompt that walks a person through building an explicit correspondence map between one source they already understand and one target they don't: which relations in the source line up with which relations in the target, which source elements have no counterpart at all (missing correspondences), and which surface resemblances are false friends that must not be carried across. Its defining move — the thing that separates it from every sibling — is that it operates on a single source paired with a single target and insists the mapping be made relational and explicit, so the analogy is judged on shared structure rather than on how alike the two things look.

Example¶

A data scientist has to monitor a fraud-detection model in production but has never done ML monitoring. A colleague offers an analogy: "it's like statistical process control on a factory line." The prompt makes them lay it out instead of nodding. Source (SPC): a process makes parts, a control chart tracks a measured dimension, a point outside the limits signals the process is out of control, an operator hunts for an assignable cause. Target: the model emits scores, so — chart of the score distribution ↔ control chart; a distribution shift ↔ an out-of-limit point; a changed data pipeline ↔ an assignable cause. Then the harder lists. Missing correspondence: SPC assumes the spec is fixed, but a model's "spec" (the concept it predicts) can itself drift, and there is no clean SPC counterpart for that. False friend: in SPC an out-of-limit point almost always means something broke; for a model a shift can be the world legitimately changing, which the model should adapt to, not a defect to repair.

The output is not "the analogy works." It is a map that says precisely which SPC intuitions to import (chart a stable signal, alarm on drift) and which to distrust (treating every alarm as a defect). That map is the transferable part; the factory imagery stays behind.

How it works¶

The prompt is built to elicit relations, not shared attributes, and to demand three lists rather than a verdict:

State the source's core relations in abstract terms ("A regulates B," "X triggers Y"), stripped of surface detail.
Pair each relation with its target counterpart, leaving a blank where none exists.
Flag the blanks — source relations with no target match — as the places the analogy runs out.
Flag the false friends — target features that resemble the source but do not reflect a shared relation.

A map that produces no blanks and no false friends is treated as a warning sign, not a success: it usually means the mapper stopped at surface similarity.

Tuning parameters¶

Grain of correspondence — attribute-level ("both have alarms") versus relation-level ("both alarm when a monitored signal leaves a stable band"). Relation-level is what actually transfers, but it is harder to elicit.
Directionality — map source→target only, or also probe target→source to surface what the target has that the source never did.
False-friend strictness — how aggressively the prompt challenges each surface match. Higher strictness catches more overconfident carryover but slows the exercise.
Number of anchors — a single source analogy, or a second competing source invited to expose where the first misleads.
Output form — correspondence table, node-link diagram, or prose; the table makes blanks and false friends unmissable.

When it helps, and when it misleads¶

Its strength is speed and honesty: for almost no cost it converts a vague "it's like X" into a relational map and drags the false friends — the deceptive similarities that produce overconfident analogy — into the open. It is the natural first move when someone is about to reason from a single strong analogy.

Its central trap is that a filled-in table feels validated when it is not: mapping is not proof the analogy is sound, only a clearer statement of it.^[1] The classic misuse is motivated mapping — running the prompt to justify an analogy already chosen, inventing correspondences to fill the table. The discipline is to treat the completed map as a hypothesis to be tested by real practice, and to require the missing-correspondence and false-friend lists to be non-empty before trusting it.

How it implements the components¶

Analogical Mapping Prompt fills only the structural-mapping subset of the archetype's machinery:

shared_structure_map — its primary output: the explicit table of relational correspondences that genuinely carry from source to target.
surface_difference_map — the missing-correspondence and false-friend lists are the record of visible resemblances that must not be transferred.

It does not decide when the mapped knowledge actually applies (that gate is Boundary Condition Checklist's boundary_condition_set / transfer_readiness_check), extract a principle from lived expertise (Apprenticeship Reflection's source_abstraction_prompt), or provide any practice — the map is a static hypothesis, not a rehearsal.

Instantiates: Transfer Scaffolding — it supplies the explicit source-to-target correspondence the rest of the scaffold builds on.
Sibling mechanisms: Case Comparison and Contrast · Worked Example Variation · Apprenticeship Reflection · Boundary Condition Checklist · Negative Transfer Red Team · Bridging Examples · Cross-Context Practice · Near-to-Far Transfer Sequence · Implementation Coaching · Transfer Debrief · Transfer Prompt

Notes¶

This prompt maps one source onto one target. When the goal is to induce a general principle from several examples — and to read off where it stops holding — that is Case Comparison and Contrast, a different mechanism that abstracts across cases rather than mapping a single pair.

References¶

[1] Structure-mapping theory (Gentner) holds that a good analogy is one of shared relational structure, not shared surface attributes — which is why this prompt privileges relations over resemblance. The theory describes what a valid mapping looks like; it does not certify that any particular mapping is correct, which is exactly why the map here is a hypothesis, not a verdict. ↩