References: add the observed-effect relation and its example set (#402)

[astrogilda]

Closes #402 by registering observed-effect in §3.1.2 as a non-breaking informative addition, in the shape of #389. The registry entry defines the referenced object as facts and limits what a verifier may conclude to rule 3.

Five signed records in examples/observed-effect/ cover a verified observation, one altered after issue, a disagreement, an unresolvable reference and an unknown observer key. Two of their statements come byte for byte from the observed-effect corpus in agent-evidence-vectors. At review I renamed the condition-appraisal example's verdicts the same way, and both example suites now test a bad signature over matching bytes.

Every commit is signed off, CHANGELOG.md is updated, and on f160107 pytest reports 2234 passed and 6 skipped.

Edited on September 24 to add the condition-appraisal rename and the current test counts.

[github-actions]

🔴 Contributor Check: HIGH

Check	Result
Profile	HIGH
Credential	LOW
Overall	HIGH

Automated check by AgenTrust Contributor Check.

[imran-siddique]

Please rename the example verdicts to describe the referenced object, rather than the effect itself. effect-confirmed currently means that the digest matches and the observer’s signature verifies; it does not establish that the reported change occurred. Likewise, altered evidence does not establish that the effect was contradicted. Keep resolution, digest match and signature verification as separate findings, consistent with the registry text.

[astrogilda]

Thanks, Imran, for catching the overclaim in effect-confirmed. In 305895c each verdict describes the referenced observation instead: verified, digest-mismatch, unresolved, unverified or signature-invalid, with the prefix observation-. Resolution, digest match and signature verification stay separate fields, each recomputed from the committed bytes.

Should I send the matching rename for the condition-appraisal example, whose appraisal-contradicted has the same overclaim?

[imran-siddique]

The revised observation verdicts address my wording concern. All 36 related tests pass. A corrupted signature with a recomputed reference digest correctly reports signature failure separately.

Please include the matching condition-appraisal rename in this PR, limited to example terminology and tests.

[astrogilda]

Hey @imran-siddique, thanks for running the suite and for the corrupted-signature case, which is now a test in both example suites. In f160107 the condition-appraisal verdicts follow the observation ones: appraisal-verified, appraisal-digest-mismatch, appraisal-unresolved, appraisal-unverified and appraisal-signature-invalid, with only example terminology and tests changed. Locally on f160107, 2234 tests pass and 6 skip.

@chernistry, only your example's verdict names moved, and I'm happy to change them if you'd prefer others.

[chernistry]

@astrogilda the new names are right and I would not change them: appraisal-verified, appraisal-digest-mismatch, appraisal-unresolved and appraisal-unverified each name the state of the referenced appraisal, which is the rule the README now states, and the separate appraisal-signature-invalid for a failing signature over bytes whose digest matches is a case the old set could not express. Nothing on my side reads the old strings. Thanks for carrying the rename.

[imran-siddique]

The requested terminology change is complete at f160107. All 38 example tests pass, including separate signature-failure and digest-mismatch cases. Existing validation rules remain unchanged. Hosted CI passes; five local Windows fixture failures also reproduce on the pre-PR base.

[imran-siddique]

@astrogilda main moved 12 commits and this now conflicts on CHANGELOG.md only. Merging main in, keep main's two new Added entries (intent_bridge.verify_bridge() successor observation, provenance.verify_record() now) and your observed-effect entry, and take your condition-appraisal wording (it says "verified" where main still says "confirmed"). I ran that resolution locally: same 28 files, +1070/-36, and the four affected test files pass 173/173. Once you push it I will merge.

[imran-siddique]

@astrogilda approving c217d83; it should have merged after your 09-28 push. #437, #432 and #401 merged today and it now conflicts on CHANGELOG.md and examples/README.md only. Merge main once more keeping both sides' entries and I will merge it the same day.

[astrogilda]

@imran-siddique, 5a6a5fe merges main (152c843), keeps both sides' entries in CHANGELOG.md and examples/README.md, and passes the full test suite locally. The old verify_bridge() successor-observation bullet is gone, since main's entry for #424 replaces it.

Pushing it dismissed your approval, so the approval check needs one on 5a6a5fe, and CI, CodeQL, ClusterFuzzLite and runtime-evidence need their runs approved. After the merge I plan a follow-up PR that tests the observed-effect examples against a pinned release of https://github.com/probityai/agent-evidence-vectors.