advanced-methodology-extensionslisted

# Advanced Methodology Extensions These techniques extend specific stages of `/evolve-solution` when the standard tool in that slot is insufficient. Each maps to exactly one stage. Use one extension at a time — stacking multiple extensions on a single run risks methodology theater. ## When to use this skill Do not invoke by default. Check first: | Signal | Extension indicated | |---|---| | Tournament variants all share the same core mechanism | Structural Diversity Enforcer (Stage 4) | | Oracle is binary; can't tell WHY a variant fails | PRM-lite intermediate scoring (Stage 4) | | Skeptic is Claude; Claude wrote the variant | Cross-Model Critic (Stage 5) | | Finalist touches security, fraud, or financial flow | Fraud Attack Simulator (Stage 5) | | Simulation produces results but you need an equation | Simulation-to-Equation extraction (Stage 6) | --- ## Extension 1 — Structural Diversity Enforcer (Stage 4 add-on) **Problem:** five variants that all use the same mechanism but with different parameters are not a tournament — they are a parameter sweep. Structural diversity means different causal paths to the goal, not different settings. **How it works:** Before scoring begins, build a Mechanism Map: ```yaml variant_id: V1 mechanism: "cache the embedding, reuse on re-query" # causal path assumption_set: [embeddings are stable, queries repeat] independent_of: [V2, V3] # yes/no: does this variant's mechanism differ from Vn? ``` If any two variants share the same me