Structural Recursive Self-Improvement: DBM-SI as a Rich Evaluator and Improvement Infrastructure

Structural Recursive Self-Improvement (SRSI) is a research framework for studying recursive improvement as a structured computational process rather than candidate generation alone. The framework centers on Rich Evaluators, Counter-Evidence, structural comparison, localization, verification, Structural Improvement Memory, and Improvement Governance. It introduces the Evaluator Bottleneck hypothesis, explores Two-Way CCC and Counter-Evidence Search as mechanisms for Anti-Goodhart recursive improvement, develops Localized RSI and recursive structural growth as alternatives to whole-system replacement, and maps existing DBM-SI mechanisms into one possible Rich Evaluator and improvement infrastructure. The project also proposes the AI-SI-RSI Gold Rush as a research hypothesis in which trustworthy machine-operable judgment, Structural Search, Evaluator Packs, and Improvement Runtimes may become increasingly important as AI candidate generation becomes more abundant.