Auditable Conformance and Cross-Library Interoperability Testing for ML-KEM and ML-DSA

An implementation that passes internal round-trip tests may still produce artifacts that another library cannot consume. We present an auditable workflow for selected-vector conformance and raw-artifact interoperability testing of ML-KEM and ML-DSA in liboqs and PQMagic. The workflow links FIPS 203/204 Automated Cryptographic Validation Protocol (ACVP) projections, bidirectional producer–consumer cases, and deterministic negative tests to vector and binary hashes. For pinned liboqs 0.15.0 and PQMagic-SHAKE builds, both adapters matched all 690 expected implementation-vector outcomes across ML-KEM-512/768/1024 and ML-DSA-44/65/87. All 1200 bidirectional positive cases and 90 deliberately sparse deterministic one-bit mutation cases produced their specified outcomes. A supporting regression check passed 12 persisted-artifact replay cases between two PQMagic builds from the same commit; it does not establish cross-version compatibility. Performance measurements provide a descriptive single-session snapshot on one Windows host, with 50 timed batch observations per operation. Aigis and SHAKE/SM3 comparisons are supporting configuration analyses, not extensions of the standardized cross-library compatibility claim. The accompanying deployment considerations are a taxonomy and checklist, not a validated decision procedure. The contribution is an auditable record of specified artifact-level behavior, bounded by the selected vectors, recorded commits, host, and interfaces. It does not establish product certification, protocol interoperability, or stable performance rankings across sessions.