TraceAGTAdapter.build_trust_record() always claims an affirming appraisal it never performed

[Yatsuiii]

appraisal.status is a verifier-owned field — spec/trace-v0.2.md §3.3.1: "A verifier MUST record the depth it actually checked in appraisal.provenance_depth_verified" and "the verifier MUST set appraisal.status to contraindicated" when evidence fails. TraceAGTAdapter.build_trust_record() sets it unconditionally to "affirming" at record-construction time, before signing and before any verifier has run:

# src/agentrust_trace/adapters/agt.py
"appraisal": Appraisal(
    status="affirming",   # hardcoded; no constructor parameter to change it
    ...

TraceAGTAdapter.__init__ has no appraisal_status parameter. Every record this adapter produces claims an affirming appraisal regardless of whether one happened.

Reproduction

inspect.signature confirms no such parameter exists; the docstring's own usage pattern is followed exactly, no override:

adapter = TraceAGTAdapter(model_provider="x", model_id="y", build_provenance_digest="sha256:"+"e"*64, transparency="https://x")
record = adapter.build_trust_record(session)
record["appraisal"]["status"]  # 'affirming'
signed = sign_record(record, key)
verify_record(signed, public_key_or_jwk=key.public_key())  # passes

Produces a cryptographically valid, verified Trust Record asserting an appraisal that never ran.

What the sibling adapter already does correctly

TraceSandboxAdapter (adapters/sandbox.py) already handles this: appraisal_status is a constructor parameter defaulting to "none", with tests (test_appraisal_status_defaults_to_none, test_appraisal_status_is_configurable_when_one_actually_happened) proving both the default and the override. TraceAGTAdapter has neither.

tests/test_agt_adapter.py's 14 tests cover every other field this adapter produces but never assert anything about appraisal.status.

---

🤖 Generated with Claude Code

[imran-siddique]

Confirmed, fixed in #336, and the comparison you drew is what made it quick: agt.py:163 was the only remaining hardcoded status in the package, and TraceSandboxAdapter already had the shape to copy, including the docstring explaining why the default has to be none.

You are right about ownership rather than merely about the default. models.Appraisal carries the same point in a comment on the field below it, "What this verifier ran, not what the issuer claimed", so the model and section 3.3.1 already agreed and only this adapter did not.

Two things your report did not reach, both worth knowing because they say how long this sat there. docs/integration/agt.md documented the behaviour as intended: the adapter "populates an affirming appraisal without independently evaluating the session". And docs/tutorials/agt-adapter.md carried a post-hoc record["appraisal"]["status"] = "none" line, which is to say the gap was known well enough to be worked around in a tutorial rather than fixed in the adapter. Both are corrected, the tutorial line is gone because the default now does it, and tests/test_docs_quickstart.py executes that tutorial so the edit is checked rather than asserted.

One consequence stated rather than discovered later: this changes what records this adapter emits. An unappraised record now says none where it said affirming, and a caller who really did appraise declares it with appraisal_status="affirming".

Coverage is the default, the override, every status the model allows reaching the record and validating, and a control that an unappraised record still signs and verifies, so the new default costs a caller nothing. Seven of those fail on the previous code.