SIBD-63: A Dataset of A-Level Cross-Domain Structural Isomorphism Discoveries with Shared Equations and Variable Mappings

63 cross-domain structural isomorphism discoveries produced by three independent AI-assisted pipelines (V1 broad-recall embedding, V2 strict-precision embedding, V3 StructTuple + LLM rerank) operating on a 4,443-phenomenon natural-language knowledge base. By construction the pipelines have zero overlap at the A-rated tier; each contributes 24, 19, and 20 records respectively. Each record pairs two phenomena from disjoint scientific domains that share the same underlying mathematical structure, and is annotated with a proposed shared equation (for 20 records), variable mapping, suggested target journal, literature-gap assessment, and a step-by-step empirical execution plan. Records span 48 domain labels across geology, condensed-matter physics, ecology, molecular biology, financial microstructure, decentralized finance, behavioral economics, civil engineering, pharmacology, and more. Released as a seed bank for empirical follow-up papers: domain experts can pick a seed matching their expertise and convert it into a publishable paper in 3-6 months. See the companion paper for methodology details and the README for usage and collaboration instructions. Companion empirical validations covering five of the SOC-class records (earthquakes, S&P 500 returns, DeFi liquidations across Aave V2 + Compound V2 + MakerDAO, mouse neural avalanches, and null controls) are available at https://structural.bytedance.city.