AI may propose. Evidence decides.

RuleOak uses AI as a tool for generation, explanation and analysis without treating the model's confidence as authority over consequential actions or its own correctness.

Separate proposal from authority

A model can propose code, a classification, a plan or an action. The system should still bind execution to policy/approval and bind correctness claims to deterministic or externally observable evidence where available.

Preserve provenance

Generated output should remain distinguishable from source facts, deterministic calculations, user decisions and verified external evidence.

Abstain instead of invent

When evidence is insufficient, generated explanation should preserve uncertainty. A polished answer is not a substitute for a missing source or unobserved effect.

Model independence

Core authority and verification semantics should not depend on one model/vendor. Models can be replaceable components around a stable evidence/decision contract.

Privacy boundary

Sending data to a cloud model is a real provider transfer. Local-first or privacy-sensitive systems should make that action explicit and minimize the payload.