[snkmcb]

Implements the P0 roadmap item that baselines the reference runtime, and rewrites the vector priority around the identities its provider already publishes.

Schemas

schemas/ defines the three machine-readable documents this repository produces:

• runtime-metadata.v1.json — canonical target identity, immutable identities, components, and capability providers for one composed target.
• release-evidence.v1.json — the immutable acceptance result committed per released target. It requires each acceptance check and the measurements that make the claim auditable, and it accepts the existing v0.1.0 record unchanged.
• acceptance-report.v1.json — the generated report tools/accept.py writes.

Generated runtime metadata

runtime-metadata.windows.json is generated by tools/runtime_metadata.py from committed inputs only — the manifest, its lock, and the newest accepted evidence for the same target. It decomposes the canonical target string, joins every resolved component and capability to its OpenStrata and OCI digests, and derives whether a Python interpreter is bundled from the lock's install list rather than asserting it. Nothing in it is hand entered.

One validation entry point

tools/validate_metadata.py replaces the inline Python in both workflows and fails when:

• a manifest and its lock disagree, a resolved component is unpinned, or a candidate artifact is unused;
• an evidence record is not schema valid, contradicts the lock it claims, or has no release record;
• the committed runtime metadata is stale;
• an identity published in README.md or a release record no longer matches the generated one;
• tools/accept.py would emit a report that does not match its schema.

--release <tag> additionally gates VERSION, the release record, and evidence for that tag.

Acceptance report shape

The report loses its version-specific field names: complete replaces full_v0_1_0_acceptance and checks replaces commands. A tier2 verification failure now marks the tier2 check failed instead of only failing the run. Probe execution is unchanged.

No dependencies

tools/jsonschema_lite.py implements the keyword subset these schemas use and raises on any unsupported keyword, so a schema cannot silently stop being enforced. CI installs no packages.

Tests

tests/tooling/test_metadata_tools.py (9 tests) is negative-case first: malformed digests, a missing required check, non-canonical targets, an unsupported schema keyword, tier2 evidence that should be rejected, a stale README locator, and evidence that contradicts the lock. CI runs it next to the validator.

Documentation

• New: docs/reference/acceptance-contract.md (implemented checks, generated report, committed evidence shape) and docs/design/composition-layout.md (proposed target-aware layout, with the digest-stability condition that decides whether the move is a relocation or a release).
• Updated: architecture overview, support matrix, README, category indexes.
• docs/roadmap/current.md drops the completed P0 and rewrites the vector priority around usd-vector-plugins v0.1.0, which already declares usd-fileformat:geojson and requires only capabilities this composition resolves. The remaining blockers are recorded as such: the product has no published OCI locator yet, its provenance records verification against a different OpenUSD runtime artifact than this composition pins, and it installs no acceptance probe under share/.

Verification

python tools/validate_metadata.py --release v0.1.0 passes all 46 checks and python tests/tooling/test_metadata_tools.py passes 9 of 9. Committed release evidence and the v0.1.0 release record are untouched.

🤖 Generated with Claude Code