Document ID: SF0037Version: 1.3Status: Active / PublicDocument Type: Methodological ProtocolApplication: AI-Assisted Academic Research The Citation Verification Protocol (CVP) v1.3 is a methodological framework designed to support researchers using AI systems by providing a structured process for verifying citation integrity, provenance consistency, and source traceability in AI-assisted scholarly workflows. CVP addresses two core failure modes commonly observed in AI-assisted research: Confirming that cited sources actually exist and are accessible Determining whether those sources substantively support the specific claims for which they are cited New in v1.3: Public Verifiability Constraint (PVC) – a mandatory exclusion gate requiring all citations to be freely accessible full-text sources (no paywalls, authentication, or partial access) Persistence and Archiving Requirement (PAR) – capturing archival snapshots and content hashes to preserve auditability Evidence anchors – requiring specific page/section pointers and excerpts for claim support verification Explicit support ratings (FULL/PARTIAL/NONE/AMBIGUOUS) with clear disposition rules While originally developed within the Synthience Institute, CVP is published as a general-purpose research tool for any researcher employing AI systems in scholarly work across scientific, technical, policy, legal, and humanities domains. The protocol is platform-agnostic by design, but requires AI systems with persistent, real-time web browsing access. Systems relying solely on training data, cached results, or simulated behavior cannot execute this protocol reliably. CVP explicitly does not replace scholarly responsibility. It requires researchers to personally read and evaluate source materials. CVP serves as a verification safeguard, helping detect non-existent citations, mischaracterized sources, and unsupported claims before publication. The protocol defines a clear, auditable workflow for citation existence verification, relevance verification, claim-level support verification, detection of common AI failure modes, and tiered verification rigor appropriate to publication risk. Applied Example: CVP was applied to verify all 12 citations in the RICO (Relationally-Induced Coherence Organization) technical report (10.5281/zenodo.18086834). The complete verification is documented in the RICO Citation Verification Report (SR001-VR1) (10.5281/zenodo.18082749), achieving 100% verification rate.
Thomas Gantz (Fri,) studied this question.