Skip to content

External read-only audit: replay assurance, crypto capability boundaries, dependency and CI qualification (507995ba) #6

Open

Description

@Knapp-Kevin

Review scope and revision

External, read-only source and CI review performed 2026-11-15 at main commit 507995ba. Reviewed README.md, AGENTS.md, Cargo manifests, code in the crypto/FHE/tensor/replay/verification surfaces, conformance vectors, active issues/PRs, and historical GitHub CI/release logs. No local build, live dependency-resolution exercise, penetration test, independent crypto validation, or signed Qortara gate was performed.

Findings

Significant downstream compatibility constraint, already tracked

Cargo.toml caps blake3 at >=1.5,<1.6 to preserve the Rust 1.83 minimum. Issue #2 describes a real conflicting downstream resolution path through Hologram; PR #3 proposes broadening the dependency requirement while preserving the older local lock and toolchain. This is not a new duplicate fix request.

Acceptance: independently resolve and build both an MSRV 1.83 locked consumer and a newer-toolchain downstream consumer requiring recent BLAKE3; maintain exact lock/source provenance, cargo-deny and no_std checks.

Assurance boundary: structural replay is not source-integrity or producer-authenticity verification

crates/uor-prism/src/replay.rs re-exports uor_foundation::enforcement::replay::certify_from_trace. Its documented operation validates the trace structure and reconstructs a certificate from a fingerprint already stored in the trace, deliberately without recomputing the source digest or invoking the author's deciders. uor-prism-verify is a thin façade without independent cryptographic behavior.

This is an intentional design scope, not an observed exploit. An integrator must not turn a successful trace replay into an unqualified claim about authenticated original source, model semantics or trusted producer.

Acceptance: explicitly document certificate assurance level and upstream fingerprint/producer trust requirements in consumer-facing examples, then test that forged/unbound provenance is not mislabeled as independently authenticated.

Capability boundaries: declarations/reference kernels versus operational implementations

  • crypto/src/signature.rs and crypto/src/curve.rs expose axis traits, not bundled production signature or elliptic-curve kernels.
  • fhe/src/fhe.rs explicitly calls OneTimePadFhe an XOR-based test reference, not cryptographic FHE.
  • tensor/src/tensor.rs provides a concrete bounded CPU integer matmul reference, but explicitly says operational substrate-Term folding across the complete kernel is follow-on. crypto/src/hash.rs likewise uses established library hash implementations with conformance test vectors; full substrate-Term hash-round composition is a follow-on.

Acceptance: publish an explicit tested matrix distinguishing: trait/API only, conformance reference, production operational implementation, formal proof scope, and independent authentication evidence. Do not advertise reference FHE as confidentiality-preserving homomorphic encryption or a trait as an implemented cryptographic algorithm.

Governance/supply-chain readiness and exact-commit CI

  • CI uses actions/checkout@v4, Swatinem/rust-cache@v2, and EmbarkStudios/cargo-deny-action@v2; release workflow includes mutable version references including dtolnay/rust-toolchain@stable. Issue #4 / PR #5 already tracks repository-standard hardening. Confirm pinned action identities and restricted release permissions as part of that change.
  • Main CI run 26347437165 and release run 26347477413 succeeded for this exact main revision on May 24, 2026. A later CI run 43546211681 succeeded on a different PR commit on Nov 14, 2026 and must not be substituted as a new main acceptance.
  • Regular Wiki Drift Check success is positive architectural-link hygiene but not a current whole-suite security/conformance pass.

Acceptance: make final-revision CI, dependency/MSRV resolution, Cargo packaging, no_std compilation, cryptographic vectors, and independent release identities available in a source-bound report. Pin action source revisions through the existing governance PR.

Integration considerations for Qortara Logic

Potential fit: bounded deterministic computations, content hashes/Merkle commitments and trace schema checking under clear evidence-source trust boundaries. Not yet qualified: independent proof of source correctness, authenticated provenance, production FHE, curve signature implementation, or general substitution for agent reasoning. Benchmark cost, correctness, and latency against existing Qortara paths before any integration decision.

Verdict

CONDITIONAL for scoped typed Rust primitive use after dependency and authority-boundary qualification; NO-GO for claiming this repository alone independently certifies source provenance or supplies production homomorphic encryption. Hosted Qortara Logic tools only evaluate caller-supplied evidence/structure and cannot execute this audit or issue a release approval.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions