When a Repository-Preregistered AI-Mediated Adjudication Ends in a Recorded Global Non-Decision: A Repository-Scoped Process Trace

This descriptive technical case report reconstructs one purposively selected, partly mid-course, author-controlled AI-mediated research adjudication from a commit-pinned repository record. Receiver expectations were committed before the recorded source issuances and receiver result; source roles could refuse, repair, and issue evidence packets; role records declared prior information exposure; one exposed comparison leg was excluded; and deterministic checks included both matches and failing directions. The historical harness executed the encoded checks but did not execute the prose outcome map. The contemporaneous results and a later internal referee record treated the run as authorizing no global substantive outcome under each of three named, non-exhaustive construals. The paper's contribution is a bounded process trace and an explicitly unvalidated adjudication-provenance profile that foregrounds authority, exposure, exclusion, deviations, and decision licensing alongside familiar provenance fields. Its central distinction is between successful execution and permission to claim scientific success. It does not establish workflow efficacy, agent or reviewer independence, a statistical null, comparative superiority, prevalence, object-level scientific validity, authenticated wall-clock chronology, or exhaustive interpretation. The deposit includes the publication PDF, a checksum-bound evidence archive, and editable source. The evidence archive preserves the frozen expectation map, source-issued and receiver-imported packets, executable fixtures and harness, stored outputs, role and chronology maps, deviations, referee corrections, manifests, and an offline verifier. The verifier checks package integrity and recorded consistency; it is not an independent adjudicator or external replication.