Effect-Transparent Governance for AI Workflows — Rocq Proof
Researchers formalized AI workflow governance in Rocq, proving that safety controls on side effects (memory, external calls, LLM queries) don't reduce computational expressivity.
The formalization spans 12,000 lines, 36 modules, and 454 theorems—all machine-checked, zero admitted lemmas.
Key result: a decidability boundary where governance rules stay decidable while semantic program properties remain undecidable.