Signal Verification-Time Dependency on a Disappearing Evaluator

Summary

AI governance and assurance often assume that a consequential model-mediated decision can be reconstructed or tested after the fact, but that assumption may fail once the evaluator that produced it is no longer accessible in the same version and execution context. This paper develops three verification-time constructs derived from Execution Governance (EG) 3.0: Decision-State Commitment, Independent Verifiability, and Counterfactual Auditability. Independent reprocessing of released Study 2 artifacts reproduced two original within-family behavioral comparisons: a 52.0% modal-decision reversal rate for Llama 3.1 8B versus Llama 3.3 70B (26/50) and 30.0% for GPT-OSS 20B versus GPT-OSS 120B (15/50), with the corrected baseline establishing these as within-family rather than provider-established succession comparisons. Post-hoc re-pairing against Groq-designated migration paths yielded 64.0% and 38.0% reversal, but these figures remain descriptive because the cross-family invocation parameters were asymmetric. A 22-event retirement census independently recomputed to a median of 16.45 months and a mean of 18.72 months (range 3.9-40.3 months), with 17 of 22 intervals below 24 months, also showing that evaluator availability can differ by service surface. The joint contribution is an operational verification-time protocol and an optional Verification-Time Preservation Package (VTPP) specifying what evidence to bind at authorization time and what a separately trusted verifier can substantiate later, described as downstream and non-authorizing since it does not alter the EG Core Formula or state jurisdiction-specific legal admissibility.