0.9.0 schema rejects §3.1.2 references (additionalProperties: false)

[mohdibrahimaiml]

Gap

TRACE spec §3.1.2 (trace-v0.2) defines an OPTIONAL references array, including registered rel value behavior-trace.

The published 0.9.0 JSON Schema (trace-v0.2.json) does not declare references. Because the object schema uses additionalProperties: false, a Level 0 record that includes a spec-valid references block fails schema validation (Additional properties not allowed).

This is distinct from #226 (registry governance for further rel values). The issue here is mechanical: normative text already names the block, the released schema drops it.

What I observed

jsonschema / iter_errors on a record that otherwise verifies at Level 0:

• property references present
• schema 0.9.0: not listed
• result: rejected as additional property

Ask

Treat this as a schema/release alignment bug against §3.1.2, not a new profile. Either:

1. ship a schema that includes the references block already specified, or
2. document that 0.9.0 consumers must ignore references until a schema bump, without treating that as a producer defect.

I am not requesting a new rel. behavior-trace is already in the registered set in §3.1.2.

[github-actions]

🔴 Contributor Check: HIGH

Check	Result
Profile	HIGH
Credential	LOW
Overall	HIGH

Automated check by AgenTrust Contributor Check.

[imran-siddique]

Tracking the pattern this belongs to in #247: the normative spec, the JSON Schema, the reference model and the docs state the rules and nothing says which wins. This issue is one of five instances found in three days. The ruling above stands on its own; #247 is about stopping the next one.

[imran-siddique]

Confirmed, and I checked the published artifact rather than the repository, since that is what your report is about.

Downloading agentrust-trace 0.9.0 from PyPI and reading the schema inside the wheel:

agentrust_trace/schema/trace-v0.2.json
  additionalProperties: False
  has 'references': False
  properties: appraisal, build_provenance, cnf, data_class, delegation, eat_profile,
              iat, model, origin, policy, runtime, signature, subject,
              tool_transcript, transparency

And §3.1.2 at 25013c59d607860a4a4d2608476280286f688243 does register the value you are using:

behavior-trace. A behavioural record of what the agent did, of which this record is the environment evidence.

So a producer emitting a spec-valid references block is rejected by the schema we ship. Your diagnosis is right, and separating it from #226 is right: nothing here is asking for a new rel.

Option 1, and the schema is already fixed. schema/trace-claim.json on main carries references with a full sub-schema, including a description that states the two verifier obligations the schema cannot express. Nothing needs writing.

Option 2 is the one I want to decline explicitly. Documenting that 0.9.0 consumers must ignore references would write a workaround into the spec for a problem whose fix already exists, and it would put us in the position of telling producers that conforming to the normative text is a defect. It is not; the package is stale.

The real answer is a release, and this is not the only thing behind it. main is 92 commits and 175 changed files ahead of v0.9.0, and pyproject.toml on main still says 0.9.0. Waiting behind that same tag:

• this references block
• schema/trace-revocation.json and schema/trace-revocation-bundle.json, landed 19 August, which @aeoess reported against in aeoess-aps: revocation mapping, two TRACE surfaces, three questions integrations#140 as "current main, untagged"
• the nine reference-implementation defect fixes merged today in fix: nine defects in the reference implementation, found by a package wide sweep #236, including a revocation check that read a non-bool store answer by truthiness

Three separate contributors this week have reported problems whose answer is the same version bump. That is the item, and I am treating it as such rather than closing this one on "fixed on main", which is not fixed for anyone who installed the package.

Leaving this open until a release ships rather than closing it against main. Thank you for reading the schema in the package instead of the one in the repository; that distinction is what made this reproducible.

[aeoess]

When the release is cut I can run the published wheel from the integration side before tested_against in aeoess-aps moves: the packaged schema against a §3.1.2 record carrying references, and the revocation paths from agentrust-io/integrations#140 through verify_record with the store answers #236 changed, all against the installed artifact rather than a checkout. Exact version, artifact hash and results back here.