Warren Smith

OMEGA · verification PROVEN

Verification

The formal artefacts do not prove real-world correctness or safety. They establish properties of a defined formal model and name the limits of that model. Two checkers: the Lean 4 kernel, and a SymPy necessity harness.

What is proven, and what is not

The Lean model establishes internal consistency of the Governed conjunction, projection, absence and joint-sufficiency lemmas per conjunct, and constructive tamper-evidence, including at the JSON encoding level, with collision-resistance carried as an explicit hypothesis rather than an axiom. A SymPy harness establishes failure-mode necessity for nine schema fields; the rest are design rationale. The model does not establish real-world correctness, agent safety, deployment behaviour, or that the formal definitions match the prose specification.

Verified in the evidence inventory

From a fresh public clone at commit 2fab5d1, on toolchain leanprover/lean4:v4.27.0 with no Mathlib: lake build completed successfully (16 jobs), and #print axioms over the root theorems reported 0 user-declared axioms. Every theorem depends only on Lean's standard axioms (propext, Classical.choice, Quot.sound) or on none.

One wording note carried from the repository: the constructive core theorem tamper_implies_collision is described in the repository README as "axiom-free", but #print axioms reports it depends on the three standard Lean axioms. The precise statement is that it uses 0 user axioms and takes no collision-resistance hypothesis, not that it is free of Lean's standard axioms.

Main theorems

Reproduce

git clone https://github.com/repowazdogz-droid/omega-lean-proof
cd omega-lean-proof
lake build                              # Build completed successfully (16 jobs)
lake env lean probes/AxiomProbe.lean    # 0 user-declared axioms

This pass. Re-verified this pass: lake build and the axiom probe were re-run from a fresh public clone at 2fab5d1 (16 jobs green; the axiom listing reproduced), not only read from source.

Honest limits

The repository names structurally unclosable limits, among them: the model cannot distinguish genuine reasoning from a perfect post-hoc rationalisation (it does not access model internals); the external anchor can be poisoned by an actor who rewrites the whole chain; a physical staleness gap remains between the sealed record and the world; and accountability for an emergent, multi-actor failure is not located by a factual record alone. Each is stated with what its closure would require.

The specification · Back to OMEGA