Conformance report: CDF Harness, the reference implementation
Contract 1.2.1, tag v1.2.1 at 3088e44
Implementation CDF Harness, Cognitive-Delivery/cdf-harness, branch harness/v1, commit 5cd62563 (2026-10-06, "Pin the governance contract to 1.2.1; the lease journal writes the decision record the contract specifies")
Consumed as a git submodule at contract/ (DR-156)
Signature algorithm HMAC-SHA256 over the canonical bytes of every lease field but signature (SPEC §9, item 6)
Reported by the implementation itself, from its own test suite, 2026-10-06

SPEC §9 says an implementation should publish the result of running the corpus, so that the claim is evidence rather than assertion. This is that publication for the reference implementation. It is the harness's own statement; nobody outside the project has reproduced it, and it says so here rather than leaving a reader to assume otherwise.

The adapter

test/unit/contractConformance.test.ts imports runConformance from the published runner and supplies seven hooks, each the code that runs in the product rather than a copy written for the vectors: validate (Ajv over the contract's schemas), roundTrip (parse and re-emit), canonicalise (canonicalJson from src/ledger/chain, which declared_hash, every signature and the ledger chain go through), narrow (narrowManifest from src/governance/leases/narrow, its refusals mapped to R codes), hash (manifestHash), verify (verifyLeaseSignature) and rules (checkLeaseRules from src/governance/leases/rules). It does not expose lastErrors, so the runner does not check rejection paths from that test; the contract's own CI checks them with the reference Ajv adapter.

The run

npx vitest run test/unit/contractConformance.test.ts at the pin, 2026-10-06:

contract: 40 valid and 98 invalid fixtures
✓ |kernel|  test/unit/contractConformance.test.ts (6 tests)
✓ |default| test/unit/contractConformance.test.ts (6 tests)
Test Files  2 passed (2)
     Tests  12 passed (12)

The first case asserts report.failures is empty and canonicalChecked is true; with the hooks above the report also carries narrowingChecked, signaturesChecked and rulesChecked true and errorPathsChecked false. The suite does not print the one-line summary npm test prints in the contract repository. Three further cases prove the runner can fail: a JSON.stringify canonicaliser, a validator that accepts everything and a round trip that drops the unknown field are each caught. The file runs twice because it is in both vitest projects, default and kernel; under kernel the vscode module is aliased to one that throws on import, so the run also shows the adapter reaches no editor code.

Real artefacts

test/unit/contractRealArtefacts.test.ts validates the harness's own journals. A zero count fails for the audit events, CDI signals, provenance records and assessments; an empty or absent lease journal passes, because a workspace that has never issued a lease has none. At the pin, 2026-10-06: 33,665 audit events, 19,287 CDI signals, 4,588 provenance records, 2 assessments, 6 lease records and 1 config; zero rejected.

Narrowing vectors

conformance/narrowing-vectors.json was generated from this implementation (test/unit/contractNarrowingVectors.test.ts with CDF_WRITE_NARROWING_VECTORS=1), and that test fails when the committed file differs from what the code produces now; it also runs the published runner with narrow supplied by narrowManifest. The vectors are this implementation's behaviour, byte-exact, so a second implementation that disagrees with one has found a defect in one of the two.

What this report does not show