The deterministic governance library
Long-form thinking on why AI governance must be enforceable, reproducible, and provable.
The deterministic governance library
Long-form thinking on why AI governance must be enforceable, reproducible, and provable. The pieces below build one argument: that in regulated settings, a governance layer is only worth its name if it can stop an action, reproduce its decision, and prove what happened to someone who wasn't there.
Where to start
Three threads run through the library. Follow the one that matches the question you're holding.
The case for determinism
Start with What Deterministic Means and Why Probabilistic Fails, then the comparison analysis. The through-line: a probability is not a control, and "probably safe" is not an answer an examiner accepts.
The architecture of refusal
The Infrastructure of No and the control-plane pillar make the structural argument — that the defensible primitive in regulated AI is the ability to deterministically refuse, and to prove the refusal afterward.
The evidence underneath
The whitepaper and benchmark cover the threat model, the signing and replay mechanics, and how enforcement behaves under load — the engineering behind the claims the rest of the library makes.
New writing is published on the EVE blog and surfaced here as it lands. For the principles these pieces argue from, see Governance Principles.
Book a Governance Assessment
A working session mapping your highest-risk AI workflows to deterministic controls and the evidence your examiners will ask for.