Declared, Not Detected: A Blockchain-Anchored, Agentic Framework for Academic Integrity in the Generative AI Era

Integrity enforcement in higher education increasingly rests on detectors of machine-written text, yet these detectors are defeated by paraphrase and trivial edits and disproportionately misclassify students who write in English as an additional language. This article proposes SCRIBE, a framework in which the reportable offence is a declaration that misdescribes how the work was produced. Students declare an assistance level and attest to their own authorship, while an instrumented editor commits process digests to a permissioned ledger. Seven smart contracts bind these commitments to the governing policy and encode a due-process procedure that no software agent can complete; four agents prepare each case, a human panel decides it, and abstention is always available. We evaluate the framework as a mechanism-verification exercise on 60,000 synthetic submissions whose behaviour classes, generator parameters and oral-probe response rates are stated assumptions, so the figures below describe that specified model rather than expected real-world performance. Within a false accusation budget of 0.5%, SCRIBE refers 30.6% of misdeclarations at 97.1% precision, wrongly referring 73 of 22,131 truthful submissions, whereas a simulated detector refers 0.3% at the same budget. Recall ranges from 5.0% to 52.3% across plausible oral-probe parameters. Process evidence alone cannot exceed a posterior of 0.604, and a replayed writing trace cuts referral to 7.2%. For a single institution, a witnessed append-only log provides four of the six procedural guarantees; the ledger’s distinctive value lies in multi-institution settings where no party is trusted to operate the log.