DRA (Document Reliability Assurance): A Deterministic, Evidence-Auditable Approach to Assessing Claim Support in Machine- and Human-Consumed Documents

DRA is a research-stage reference evaluator that assesses documents against a defined set of reliability issue classes (evidence absence, evidence inadequacy, and internal claim inconsistency, out of nine formally defined) using a deterministic, eight-stage evaluation pipeline bound to cryptographically verifiable proof receipts. This archival deposit covers DRA-GC-1, the first frozen, publication-candidate evaluator state (evaluator 0.1.2, pipeline 1.0, model/schema 0.1.0, corpus DRA-CORPUS-1.0.0), and its accumulated internal evidence base: a 33-document development corpus, a documented robustness programme, and two internal, pre-registered, contamination-blind generalisation studies (DRA-GEN-001, DRA-VAL-002) totalling 100 blindly-evaluated documents. The archived repository contains both the evaluator source code (MIT-licensed) and its accompanying research methodology, evidence base, and manuscript/documentation (CC-BY-4.0-licensed). See LICENSE and LICENSE-DOCUMENTATION in the archive, and docs/dra/DRA-LICENSING.md for the complete licensing boundary. No external, third-party validation has yet been performed. See docs/dra/DRA-PUBLIC-CLAIMS.md in the archived repository for the complete, publication-controlled statement of what has and has not been demonstrated. This system does not provide independent validation, production readiness, certification, universal document trust, guaranteed truth detection, or guaranteed real-world safety. This archived artefact corresponds exactly to the public GitHub release DRA-GC-1 (commit e9d75a1fb21528485c10d4dea0cd79ccf823878c) of github.com/Cristina925/document-reliability-assurance. Canonical aggregate digest: 77544648dcb37caf96468123de4d19123becceff48f2023f6c885899350b857b.