MIYOG: A Trust-Separated Architecture for Computational Discovery in Finite Formal Worlds

MIYOG is an implemented architecture for computational discovery in finite formal worlds. It investigates a set of computational obligations associated with scientific reasoning: establishing what can happen, what must happen, what cannot happen, and what kinds of evidence justify each conclusion. The architecture separates candidate generation from evidential adjudication. A Creator proposes equational laws, a Destroyer searches for counterexamples, and a Sustainer maintains, derives, and revalidates laws. A fingerprinted trusted core evaluates terms, checks proof certificates, and assigns evidence-dependent outcome categories using an append-only record. MIYOG uses ASK, a bounded formal-world language with hidden state, equality-based observations, and an execution interface aligned with the trusted evaluator. The study examines 73 recorded benchmark runs over small generated worlds. Results include adversarial falsification, bounded mechanism identification, and treatment of resource exhaustion as distinct from falsification. Important limitations remain: several historical benchmarks lack retained generating code and have not been independently reproduced; guarantees are bounded; and trust separation is incomplete outside the equational-law system. This work does not claim general intelligence, discoveries about physical systems, or solutions to open mathematical problems. Research companion: https://github.com/amogh-code/miyog-research Code availability: The full research engine remains private. The public repository contains documentation and illustrative materials, not a reproducible implementation of the reported system. Status: Research preprint; not peer reviewed.