Control Integrity Regression Evidence
Synthetic/internal benchmark. This is regression evidence, not a production accuracy study.
Why v2 was blocked
A pre-hardening corruption rehearsal exposed four known misses. Receipt v3 added envelope-integrity binding, missing-balance checks, contradiction checks and deterministic receipt/input reproduction.
Integrity controls retained in v4
The registered corruption corpus detects 10 of 10 positive mutations and leaves 2 of 2 clean controls clean. Receipt v7 retains those controls and adds explicit event-specific procedure applicability so structurally irrelevant balance/document checks return NOT_APPLICABLE instead of false exceptions.
Applicability abuse checks
The separate applicability regression currently passes 10 of 10 cases, including required-balance omissions, required facility-document omissions, explicit missing-document states, source-only events, and a source-change event interleaved between balance events.
Important limits
These are small deterministic regression corpora, not statistical accuracy estimates. An unsigned SHA-256 digest is not source-origin authentication. Receipt v7 does not claim digital signature, notarization, source authenticity, non-repudiation, legal truth, engineering judgment, or credit approval.