← ClaudeAtlas

analogical-transferlisted

Structure-mapping across domains: abstract a problem to its relations, constraints and dimensionless regime, retrieve structurally (not superficially) similar problems from other fields using a seed catalogue and search, build an explicit correspondence, check which relations survive the mapping, transfer the source domain's canonical method, and validate the transfer by derivation, precedent, simulation or adversarial check. Load when a problem looks stuck within its own field or when another field may already have solved its structure.
rustcircle/c101 · ★ 0 · AI & Automation · score 65
Install: claude install-skill rustcircle/c101
# Analogical transfer An analogy is useful when the **relations** between things carry over, not when the things look alike. Structure-mapping theory (Gentner, 1983) states the rules this skill operationalises: map relational structure and drop object attributes; prefer **systematic** mappings, where connected systems of higher-order relations (causes, constrains, trades off against) carry over together; keep correspondences one-to-one and connected. Two empirical facts drive the procedure: people (and models) retrieve analogues by surface similarity while structural similarity is what makes them useful (Gentner, Rattermann and Forbus, 1993), and spontaneous transfer is rare without abstraction or two worked analogues (Gick and Holyoak, 1980, 1983). So: abstract first, retrieve by structure, map explicitly, check for breakage, then validate as you would any hypothesis. ## 0. When to use | Signal | Use analogy? | |---|---| | in-field methods exhausted or clearly misfit | yes | | a recognisable dynamic: oscillation, contention, cascade, diffusion, drift, selection, matching | yes; strong prior that another field has a canonical method | | need a novel design option for an ensemble or debate | yes, as idea generation, labelled speculative until validated | | a known, well-tested in-field method exists | no; use it | | decision needs evidence, not ideas | no; analogies generate hypotheses, they never prove | ## 1. Abstract the target Write the relational skeleton. Drop every