Skip to content

Latest commit

 

History

History
104 lines (76 loc) · 3.25 KB

File metadata and controls

104 lines (76 loc) · 3.25 KB

CRG Self-Assessment — standards

1. Current Grade: C

Assessed: 2026-04-09

2. Category Checklist

Category Status Evidence

Unit tests

Present

a2ml/bindings/rust (47), k9-svc/bindings/rust (45), mcp-repo-guardian (36), axel-protocol (14), groove-protocol (10), avow-protocol/src (8)

Smoke tests

Present

just build compiles all sub-projects; just test runs full suite

Property-based (P2P)

Present

proptest in a2ml + k9-svc Rust crates (parse/render roundtrip)

End-to-end reflexive

Present

Real manifest parsing + roundtrip stability in mcp-repo-guardian + a2ml

Contract (pre/post)

Present

a2ml + k9-svc pre/post-condition validation; avow-lib Idris2 dependent-type proofs

Aspect (security)

Present

TruffleHog secret scanning, Trivy vulnerability scanning, CodeQL, OpenSSF Scorecard

Benchmarks (baselined)

Present

criterion for a2ml + k9-svc; zig bench for groove-protocol grv6

All 7 CRG categories are present. Grade C requirements satisfied.

3. Gap Analysis: C → B

Grade B requires all categories to be robust (not just present), with consistent pass rates and integration into CI.

Requirement Status Gap

All tests run in CI

Partial

groove-protocol Zig tests and avow-lib Idris2 tests are not run in CI (toolchains missing from runners)

Avow protocol tests

New

ReScript unit tests for ProvenResult added (8 tests). Idris2 tests exist but are compile-time only. Need CI integration.

Test count > 200

Gap

Currently 160+. Need ~40 more tests across existing suites.

No continue-on-error on quality gates

Partial

Several CI steps still use continue-on-error: true (YAML lint, markdown lint, secrets scan). These should be enforced.

Benchmark regressions block CI

Gap

Benchmarks exist but are not gated — regressions do not fail the build.

Fuse3 blocker resolved

Gap

repo-guardian-fs cannot compile on Rust stable >= 1.80 due to fuse3 v0.7.3 incompatibility. Offline tests exist as workaround.

Self-referential dogfooding

Partial

hypatia-scan.yml workflow exists but is not regularly triggered. CRG self-assessment now exists (this file).

  1. Add Zig and Idris2 to CI runners (or use Nix/Guix shells in workflows)

  2. Add AVOW protocol deno task test to Justfile test suite

  3. Write 40+ additional tests: expand axel-protocol coverage, add groove-protocol edge cases, add consent lifecycle tests for avow-lib

  4. Remove continue-on-error from YAML lint, markdown lint, and secrets scan CI steps

  5. Gate benchmark regressions in CI (fail on > 5% regression)

  6. Resolve fuse3 dependency (upgrade to v0.9.0 or replace with fuser crate)

  7. Schedule hypatia-scan to run weekly and enforce its output

Grade A requires comprehensive coverage, formal verification, and external audit readiness.

  1. All lol/ postulates proven (currently 4 of 9 are provable but unproven)

  2. ECHIDNA proof verification running in CI

  3. VeriSimDB populated and queryable

  4. PanLL compliance dashboard operational

  5. External security audit completed

  6. All downstream repos at CRG Grade C or above