Published February 8, 2026 | Version v1

Codex Benchmark: Governed Evolution vs Ungated Baselines

Description

This document specifies a minimal, reproducible benchmark that tests a single architectural claim:

Claim (empirical): For matched seeds, compute, and training schedule, an admissibility-

before-commit update rule reduces long-horizon drift and fabrication-like failure modes

compared to an ungated baseline, without collapsing task performance.

This benchmark is intentionally system-agnostic. It does not require disclosure of proprietary

invariants. It evaluates the observable consequence of governed updates: bounded behavior under

long-horizon stress.

Files

Benchmark_Proof_Toy.pdf

Files (191.9 kB)

Name Size
md5:1be105214c2fc98c360d7ffa00c110ec
191.9 kB Preview Download