test(ontology): #204 OwlValidator live-path — silent false-pass is dead (dispatch 4rkh1s secondaire)

[jsboige]

OwlValidator live-path: silent false-pass is dead (dispatch 4rkh1s secondaire)

NEW additive file — OwlValidatorLivePathTests.cs (+5 tests). Companion to PR #481 (the OwlAdapter reader fix). Dispatch 4rkh1s secondaire: prove the production OWL validation path is LIVE.

What it proves

The production validator Tests/OwlOntologyValidationTests.cs (a runtime validator invoked by OwlValidatorConfig.Apply, NOT an xUnit suite) had a silent false-pass:

• ValidateMultilingualAnnotations (L169-174) and ValidateAIFMappings (L220-225) both early-return true ("No concepts to validate — skipping") when _ontology.GetResourcesByType(SKOSVocabulary.Concept) returns empty.
• Before fix(ontology): #204 OwlAdapter readers — RDFResource.ToString() + SKOSHelper silent-empty fallback (fixes #480 bug) #481, that reader ALWAYS returned empty (RDFResource type-mismatch) → validators reported PASS regardless of whether any annotation or AIF mapping actually existed.

These tests drive the REAL production validator via reflection (its _ontology field is private; the validation methods are public). They prove the fix delivered the actual production delta:

Test	Setup	Asserts
Annotation_Validator_Inspects_Concepts_And_Passes_When_Annotated	2 fully-annotated concepts	PASS (genuine — annotations present, not skip-path)
Annotation_Validator_Fails_When_Concepts_Lack_Annotations_After_Fix	4 concepts, zero annotations	FAIL (the check ran & found missing labels) — dead false-pass
Aif_Validator_Fails_When_Concepts_Have_No_Match_Mappings_After_Fix	concept with no match mappings	FAIL (check ran)
Aif_Validator_Passes_When_Concepts_Carry_ExactMatch_After_Fix	exactMatch mapping present	PASS (genuine)
The_Reader_The_Validators_Branch_On_Resolves_Concepts_NonEmpty	declared concepts	GetResourcesByType(Concept).Count == 2 (the predicate that gated the false-pass)

The decisive tests are the two FAIL assertions: before the fix they would have been PASS (empty concepts → skip). After #481 the concepts resolve, the check actually runs, and it correctly reports the missing annotations/mappings. That contrast IS the proof the false-pass is dead.

Tests

223 pass / 0 fail / 5 skip (218 + 5). Deterministic, key-free, release-independent. No existing file modified. AssetConverterConfig.cs / DatasetUpdaterRootConfig.cs / FallaciesLocalizationTests.cs untouched (reserved #444 β/γ po-2023).

Relation to #481

This PR depends on #481's reader fix (the validator tests assert non-empty concept resolution, which only holds after the fix). Recommend merge order: #481 first, then this. If #481 is reverted, tests #2/#3 here flip red (correctly — they'd expose the returning false-pass).

🤖 Worker po-2024 — dispatch 4rkh1s secondaire.

[clusterManager-Myia]

[NanoClaw]

Solid companion PR to #481. The fix is mechanical and correct; the test coverage is thorough. A few observations:

---

Production code (OwlAdapter.cs) - fix is correct

The core change across all 14 sites is identical in pattern: replace .Equals(string) with .ToString() == .ToString(). This works because RDFResource.Equals(object) dispatches by type and returns false when compared against a string or Uri - switching to string-string equality bypasses the type-mismatch.

Specific improvements beyond the raw .ToString() swap:

• CheckIsNarrowerConcept: The original try { return ... } catch { return fallback } pattern meant the fallback never executed if SKOSHelper returned false (no exception). The new try { if (...) return true; } catch { } return fallback correctly falls through on silent false, which is a real behavioral improvement - the annotation scanner now actually gets a chance to run.
• GetExactMatchConcepts / GetCloseMatchConcepts / GetRelatedMatchConcepts: Same fix - SKOSHelper returning null or empty no longer silently short-circuits; the annotation fallback now runs. This is the right pattern.

No issues with the production changes.

---

Regression tests (OwlAdapterRegressionTests.cs) - well-flipped

All 8 section-(A) tests that previously asserted the broken behavior (BeEmpty, BeFalse) are now proper round-trip assertions. The assertions are specific (HaveCount(3), BeEquivalentTo(...), ContainSingle(...)) rather than just NotBeEmpty, which is good.

The CheckIsNarrowerConcept test now also asserts the negative case (Unrelated returns false), and CheckHasClass asserts UndeclaredClass returns false. Good defensive coverage.

---

Live-path tests (OwlValidatorLivePathTests.cs) - well-designed with one structural concern

The test design is sound: prove the production validator actually runs (not just skips) by driving it through reflection. The 5 tests cover the two axes that matter:

1. Annotated concepts -> PASS (genuine, not skip-based)
2. Unannotated concepts -> FAIL (the silent false-pass is dead)
3. AIF mappings -> same pattern for both PASS and FAIL
4. Contract guard pinning GetResourcesByType(Concept) returns non-empty

One concern - reflection fragility: BuildValidator accesses a private _ontology field by name. If the field is renamed, these tests silently break (compile error, not a test failure - but still invisible in CI if the test project isn't rebuilt). This is acceptable for now since OwlOntologyValidationTests has no public injection seam, but worth noting that a longer-term fix should add a constructor overload or factory method that accepts an OwlAdapter directly. Not blocking.

One minor observation: InvokeValidate casts the return to Task<bool> without checking for null. Low risk but could be hardened.

---

Security

No secrets, credentials, connection strings, or API keys in the diff.

Verdict

Clean fix, thorough regression suite, well-documented. The reflection-based approach for the live-path tests is the right pragmatic choice. Ship it.