test: hermetic W3C N3 reasoning spec test suite

[jeswr]

🚧 Draft — agent-generated PR. Written by an AI agent working on the issue backlog; please review before considering it for merge.

Closes #336

What this adds

A jest runner for the reasoning part of the W3C N3 test suite (tests/N3Tests/manifest-reasoner.ttl), runnable with:

npm run test:spec

Per review, the upstream test suite is not committed to this repo — it is fetched at a pinned commit and cached in CI.

• scripts/fetch-n3-tests.ts (npm run spec:fetch) — downloads tests/N3Tests from w3c/N3 at a pinned commit (N3_TESTS_COMMIT) into the git-ignored __tests__/spec/.w3c-n3-tests/. It is a no-op once the directory is present, so local re-runs and CI cache hits skip the download. npm run test:spec runs it first.
• CI cache — the spec-tests job wraps the run in actions/cache@v4, keyed on hashFiles('scripts/fetch-n3-tests.ts') (which holds the pin), so runs are deterministic and a cold cache only downloads once; bumping the pin busts the cache automatically. The job is intentionally not in the release job's needs — a newly surfacing deviation should be triaged into the skip list, not block releases.
• __tests__/spec/spec-test.ts — parses the fetched manifest with the already-present n3 dependency, runs each test:TestN3Reason entry through n3reasoner, and compares the derived graph to the reference by RDF isomorphism (jest-rdf / rdf-isomorphic, already a devDependency). The cwm-style test:options map onto EYE output modes (conclusions → derivations, data → deductive_closure, neither → deductive_closure_plus_rules), and output/reference graphs are rebased onto a common base IRI (n3reasoner loads its input as data_0.n3s while the references resolve relative IRIs against the original input file name). If the suite is missing it throws with a pointer to npm run spec:fetch.
• __tests__/spec/skip-list.json — every known EYE deviation recorded with a reason, so the suite stays green while deviations are triaged one by one. Removing an entry re-enables the test.
• Separate jest project (jest.spec.config.json) + testPathIgnorePatterns in the main config, so the coverage-thresholded unit suite is unaffected.

Upstream manifest typo

The pinned manifest concatenates two entries without whitespace in mf:entries — :cwm_includes_t4:cwm_includes_t6 — which Turtle parses as a single, undefined prefixed name (: is legal in local names), so both tests would silently vanish. Since the suite is now fetched unpatched, spec-test.ts applies a small self-healing in-code patch (insert the missing space) to the manifest text before parsing; it becomes a no-op once upstream fixes it. Filed upstream as w3c-cg/N3#232.

Current results

	count
manifest entries	89
passing	50
skipped (known deviations, with reasons)	39
failing	0

Local run (both cold-fetch and warm-cache paths): Tests: 39 skipped, 51 passed, 90 total (the extra test asserts the manifest loaded), ~20 s. The 39 skips cluster into: WASM sub-reasoner gaps (log:conclusion needs shell/2/exec — same root as #873 / eyereasoner/eye#177), tests that dereference sibling files via log:semantics (not exposed through the single-document n3reasoner API), math: numeric edge cases, EYE output constructs the n3.js parser rejects, and reference files that aren't standalone-parseable. Each entry in skip-list.json carries its specific reason.

Notes

• No changes to any file touched by the in-flight fix: Parse quad results from the reasoner and drop the obsolete RDF-star workaround #1950–test: Add browser dist e2e test and document RDF 1.2 support level #1954/fix!: bump swipl-wasm from 7.0.10 to 8.0.1 #1945 work.

---

@jeswr will review — expect a delay; active review happens Wed–Fri.

🤖 Generated with Claude Code

[github-actions]

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'EYE JS Benchmark'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.25.

Benchmark suite	Current: b1b5bef	Previous: d129aee	Ratio
Run socrates query	16.01 ops/sec (±4.93%)	22.2 ops/sec (±2.39%)	1.39
Executing the socrates query	117 ops/sec (±19.58%)	3890 ops/sec (±1.67%)	33.25
Run deep taxonomy benchmark [10]	6.92 ops/sec (±0.43%)	18.83 ops/sec (±6.56%)	2.72
Run deep taxonomy benchmark [50]	0.26 ops/sec (±0.38%)	19.63 ops/sec (±7.26%)	75.50
Run deep taxonomy benchmark [100]	0.04 ops/sec (±0.69%)	18.67 ops/sec (±6.95%)	466.75
Run deep taxonomy benchmark [10] [reasoning only]	7.8 ops/sec (±16.64%)	4107 ops/sec (±1.65%)	526.54
Run deep taxonomy benchmark [50] [reasoning only]	0.65 ops/sec (±23.07%)	4147 ops/sec (±1.85%)	6380
Run deep taxonomy benchmark [100] [reasoning only]	0.17 ops/sec (±24.95%)	4136 ops/sec (±1.60%)	24329.41
Run timbl + foaf + rdfs rules	2.05 ops/sec (±2.66%)	17.23 ops/sec (±6.18%)	8.40
Run timbl + foaf + owl2rl rules	1.22 ops/sec (±1.56%)	17.03 ops/sec (±3.91%)	13.96
Run timbl + rdfs rules	4.7 ops/sec (±2.46%)	18.28 ops/sec (±6.60%)	3.89
Run timbl + owl2rl rules	6 ops/sec (±0.27%)	17.1 ops/sec (±6.85%)	2.85
Run timbl + foaf + rdfs rules [string]	2.13 ops/sec (±0.21%)	18.45 ops/sec (±7.08%)	8.66
Run timbl + foaf + owl2rl rules [string]	1.26 ops/sec (±0.13%)	18.54 ops/sec (±6.64%)	14.71
Run timbl + rdfs rules [string]	4.81 ops/sec (±0.30%)	18.77 ops/sec (±6.77%)	3.90
Run timbl + owl2rl rules [string]	6.19 ops/sec (±0.26%)	18.98 ops/sec (±6.47%)	3.07

This comment was automatically generated by workflow using github-action-benchmark.

CC: @jeswr