Warrant as State: A Regulative Architecture for Scientific Evaluation

Scientic evaluation rests on conventions that are rarely part of the object evaluated: sig-nicance thresholds xed by tradition, verdicts coerced into pass or fail, normalizationand missing-data policies left implicit, and error-correction treated as an afterthought.We take this diagnosis as established by two decades of metascience and ask a nar-rower questionwhat form should an adequate response take? We argue the responseshould be architectural rather than statistical: because the degrees of freedom an ana-lyst enjoys are a structural feature of inquiry and not a remediable defect of method,the leverage lies in making them auditable, not in engineering them away. We propose aregulative framework of four commitmentsa contract xed before the evidence, a thirdindeterminate verdict alongside conformance and nonconformance, disciplined revision,and an append-only recordand show that they compose not as a list but as a layeredarchitecture in which each component's content stands alone while its epistemic forceis grounded in the others. We then defend the framework against the two objections itmost invites: that its organizing principle is covertly vericationist and self-undermining,and that its apparent delivery of analyst-independent verdicts is defeated by the theory-ladenness of the conventions it freezes. Both are answered by drawing a distinctionthe surface claim elidesan uncorroborated conjecture from a meaningless utterance,auditor-independence from author-independenceand in each case the defensible claimis the smaller one: the framework eliminates no researcher freedom, but renders it au-ditable. None of the four commitments is individually novel. The contribution is theirintegration under an explicit fallibilist reading, the demonstration that they cohere as anarchitecture, and a falsication program to which, by its own standard, the frameworksubmits itself as a proposal not yet corroborated.