AegisLabel-FX

AegisLabel-FX is a working medical-device labeling research prototype that catches unsafe label revisions before release. A small specification change can affect multiple product labels. Our demonstration updates “Do not reuse” to “Do not reuse. Inspect packaging before use.” The platform identifies dependent labels, generates revised artwork, and blocks a deliberately faulty draft that retains the old warning. Reviewers can inspect the expected value, observed value, source revision, and triggered rule. After correction, deterministic validation makes the artwork eligible for independent reviewer approval. Approval binds the exact artwork hash and source version. A subsequent artwork change requires fresh approval. The platform generates a traceable release ZIP containing actual PDF/SVG labels, validation findings, source evidence, and approval records. Our differentiator combines change impact, evidence-linked verification, mutation-guided regression testing, and version-bound approval. Mutation Lab executes 24 isolated fault scenarios against the validation engine. The recorded synthetic benchmark detected 24/24 injected faults, with 0/5 clean controls blocked; these small experiments do not establish real-world safety performance. The application includes persistent records, PDF extraction, audit-chain verification, optional server-side Meta API adapters, and clearly labeled deterministic fallback. Separate public FDA UDI and classification reference data preserves provenance and excludes restricted terminology and contact details. Verification includes 61 backend tests, 3 frontend tests, 2 browser/accessibility tests, dependency audits, and passing GitHub Quality and CodeQL workflows. Educational prototype with bounded checks—not certified medical-device software. AI proposes. Rules decide. Humans approve.