RFC 008 slice 0b: registries, report schema, and severity policy

[zkwentz]

Checkpoint

Stacked on #1044 (slice 0a: types + manifest + fixtures), which stacks on #1041 (RFC). Second half of the slice-0 contracts — the original slice-0 PR (#1042) was split in two for reviewability.

PYTHONPATH=src:envs uv run pytest tests/test_validation/ -v   # 59 passed
python scripts/sync_validation_schemas.py --check              # schemas in sync

Every contract is exercised by a test the day it lands — the "mock API" exists before any implementation.

How to review (~550 lines)

Read in this order:

1. src/openenv/validation/signature.py — well-known-file detection table, SignatureError / UnsupportedPackageError, and the unsupported-categories list (exit-code-2 contract). GPU is not a category: GPU packages validate on providers with the GPU capability.
2. parsers/__init__.py, providers/__init__.py, graders/__init__.py — protocols plus working ParserRegistry and GraderRegistry. The load-bearing rule: grader selection reads the manifest only (a conformance test proves two manifests differing only by signature select identically). Third-party graders register via the openenv.validation.graders entry-point group. Providers start subjects under the manifest's declared network policy (default public); enforcing no-network/allowlist needs the NETWORK_POLICY capability.
3. report.py + policy.py — report models and severity-policy application: hub-lane entries filtered out entirely for local runs (an author never sees a check they can't red-to-green); ERROR fails closed; unknown ids are an internal error.
4. tests/test_validation/test_policy.py — the verdict semantics, one behavior per test.

Skim only:

• policies/severity-v1.json — the data table: all 45 check ids (33 local + 12 reserved hub/statistical); tests/test_validation/conftest.py holds the same table as a literal so the file can't drift silently.
• schemas/report.schema.json — generated, CI-checked.
• test_protocols.py / test_report.py / test_checkpoints.py — fakes and mechanical assertions; test_checkpoints.py is the accretion point every later slice appends to.

Flagged judgment calls (not in the approved design docs)

1. severity-v1.json bounds are placeholders: max_oracle_tolerance 0.1 · min_floor_margin 0.1 · max_variance_tolerance 0.2 · max_episode_timeout_s 3600.
2. Reserved hub-lane severities taken from the RFC table; cosign kept warn.
3. A judge pin declared while llm_judged: false is rejected as a contradiction (strict iff).

Refs #778, #898.

🤖 Generated with Claude Code

---

Note

Low Risk
Contract and test-only slice with no runtime CLI wiring yet; policy bounds are documented placeholders and verdict logic is heavily unit-tested.

Overview
Adds the second half of RFC 008 slice 0: executable contracts for parsers, graders, providers, structured reports, and versioned severity policy—ahead of a full openenv validate pipeline implementation.

Protocols and registries define Parser/ParserRegistry (signature → manifest), Grader/GraderRegistry (manifest-only selection, entry-point group openenv.validation.graders), and ValidationProvider/RunningSubject for sandbox exec and network policy. Signature module ships detection tables, SignatureError / UnsupportedPackageError, and unsupported categories (GPU is a provider capability, not a reject category).

ValidationReport and CheckResult models embed the normalized manifest; severity-v1.json maps all 45 check ids (local + reserved hub/statistical) with apply_policy as the sole verdict path (ERROR fails closed; local lane hides hub-only ids). Packaging includes policy JSON; schema sync adds report.schema.json.

Tests lock policy completeness against EXPECTED_POLICY, verdict semantics, protocol fakes, checkpoint 0, and CI schema sync.

^{Reviewed by Cursor Bugbot for commit af44c21. Bugbot is set up for automated code reviews on this repo. Configure here.}

[bot-ci-comment]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit e1a5f68. Configure here.}