MCRP: A Prospective, Claim-Versioned Verification Lifecycle for Computational Research

Abstract

Computational research records can expose code and outputs without identifying which execution supports a claim, which dependencies remain trusted, or who authorized acceptance. We propose the Minimum Credible Reproducibility Protocol (MCRP), a prospective, externally enforced, claim-versioned lifecycle for an immutable Scientific Record Release. It composes established claim, provenance, archival and review mechanisms while distinguishing machine evidence, reviewer-controlled execution and qualified human scientific disposition. The specification defines conservative change propagation and carry-forward eligibility bound to exact claim contracts, current approval and separately authorized scope-delta review. This revision adds a bounded transition model and conditional invariants for append-only history, known-event invalidation, late-result replay and authority separation. A consolidated table relates assessment modes to assurance levels. Agent delegation and manual trust leaves have explicit boundaries. Agent-native dissemination, computational checks and consumer-policy-bounded reuse may proceed while human scientific disposition remains pending; hosting or successful execution does not supply the missing endorsement. Executable traces include deliberately undetected missing events. The models assume trusted identities, issuance epochs and serialized updates; they do not implement authentic authority, complete event discovery or production transactions. Existing assessment, resource, transport and evaluation examples remain bounded. All development and internal adversarial review were author-directed AI work. These synthetic checks do not establish scientific validity, measured resource savings, field effectiveness or independent interoperability. We retain a proposed independent study with simpler review baselines and explicit missing outcomes. MCRP remains a testable policy proposal, not a certification standard or validated intervention.

Keywords