W3C/SemWeb Completeness Ledger#

Standing goal (owner, 2026-07-09): Exemplary and full implementation of every OFFICIAL W3C RDF/semweb-related specification we touch. RDF 1.2 / RDF-star is UN-PARKED and in progress (owner directive 2026-07-16): the syntax/eval suites are landed at 212 pass, 0 fail (Turtle/TriG reified triples << >>, ~ reifiers, {| |} annotation blocks, VERSION directive, N-Triples/N-Quads triple terms; RDF.Canonical triple-term bnode recursion) behind a threaded rdf_syntax_mode (Mode_11 default, so RDF 1.1 output stays byte-identical); the rdf-semantics + c14n-1.2 suites are the next waves (#305); SPARQL 1.2 triple-term/reifier syntax landed (see the SPARQL 1.2 row below for the current count — this inline used to read "219/254" and had gone stale). Anything short of full is listed here as a failure, not softened as a skip. Focus: in-depth coverage and performance. Every capability gets an npm API and Hub documentation.

Why 100% first: complete real test coverage is the FOUNDATION FOR AUTOMATED PERF RESEARCH (owner, same date). A perf harness over a partially-conformant engine measures the wrong thing; the coverage ledger below is the prerequisite checklist for the per-engine perf program.

Update discipline: numbers here are labelled and dated at last edit; the live dashboard supersedes them the moment they diverge. Rows sorted alphabetically. "npm/Hub" = typed API + hub page exist.

RDF 1.2 / RDF-star: UN-PARKED (owner, 2026-07-16), in progress — syntax/eval suites landed 212/0 (see the header note above); rml-star unblocked in principle, still sequenced behind the 1.2 semantics/c14n waves and SPARQL 1.2 (#305). (The former "Protocols parked" note is also obsolete: Protocol 34/0 and Graph Store 19/0 pass in-process, and factoidal-http serves them — see the SPARQL row.)

Non-semweb engines (XML, XSLT, XPath, Schematron, JSON Schema, MathML, XForms, TOAN) stay measured on the dashboard but are outside this ledger's "full implementation" mandate.

Dispositions table (2026-07-16 day-closure /goal)#

Owner day-closure goal (2026-07-16): every W3C + ShEx suite either fully green or every residual fail/skip named with a one-line written disposition. Vocabulary (fixed): disputed-fixture (the vendored corpus fixture itself is wrong/contested upstream) · dependency-blocked(what) (blocked on a named engine/component not yet built) · planned-family(issue/doc) (a named family with a plan and a tracker) · environment(what) (a checkout/toolchain gap, not an engine defect) · by-design(why) (deliberately out of scope, the processor's behavior is correct).

Capability-existence rule (added 2026-07-17, owner escalation on the namespace:: axis). disputed-fixture and by-design may only be applied when the underlying capability EXISTS; a missing feature is always planned-family/gap, stated as such. Labelling a hole "implementation-defined" or "disputed fixture" implies the processor made a correct choice among valid behaviours — false when it produces nothing because the feature was never built. Full re-audit + the XPath/XSLT spec-derived coverage matrix that prompted this rule: docs/designissues/2026-07-17-xpath-xslt-coverage-matrix.md.

Scores below are cross-checked against git merge-base --is-ancestor on this worktree's HEAD (1e152504) — a score is reported as "landed" only when its source commit is a verified ancestor. Suites with zero fails and zero skips (SPARQL 1.1, RDF 1.1 six-suite, RDFC-1.0, SHACL, type-consistency 352/0, DID, geosparql, mathml, schematron, xforms, toan_matrix, vc20_api, vc_di_eddsa, vc_stage1, hdt_stage4_parity, xpath_unit, rml_core, shex_negative_syntax) are omitted — nothing to disposition.

CSVW csv2rdf — 264 pass, 6 fail (out of 270)#

Linked: docs/designissues/2026-07-15-csvw-csv2rdf-fail-triage.md + #297. The ledger's own headline (row above, "235 pass, 35 fail") is stale — round 3 (6ee10225) and round 4 (c91fd4c7) are both ancestors of this worktree's HEAD and bring the real current score to 264/6. Named residuals, from the triage doc's round-4 remaining-detail section:

• test036, test037, test102 — planned-family (#297 family C: @id real-IRI table/group-node identity, not yet a CSVW.Conversion signature change; test036/037 additionally need annotation-object common properties).
• test148, test149 — planned-family (#297 family A residue: title-LANGUAGE-aware column-name derivation, a decode-shape change to csvw_decode_titles).
• test194 — dependency-blocked(regex engine phase 3, #304) — the datatype.format duration-pattern facet needs a regex match; #304 phase 2 (Brzozowski-derivative Exec + XSD pattern parser) has landed (e2e2dbb9, ancestor-confirmed) but the CSVW/OWL consumer swap (phase 3) has not.

eecc_interop — 4 pass, 0 fail, 51 skip (out of 55)#

Vendoring commits (third_party/testing/eecc/PROVENANCE.md): vc-verifier-rules at b165d68393485c0fd3c5285eaa32fa9c3602a264 (2026-06-22), webuild-attestations at 1f1096b06b40b19c3711d635a2e9bada8f8584ef (2026-07-06), both Apache-2.0, both retrieved 2026-07-13.

• 4 skips (crypto-verify on the 4 real DataIntegrityProof credentials) — dependency-blocked(live HTTPS did:web resolution; this runner is offline/network-free and no fixture bundles the Ed25519 public key material).
• 47 skips (40 webuild-attestations files + 6 vc-verifier-rules JSON Schemas + 1 JWT-VC example) — by-design(not W3C VCDM JSON-LD credentials — SD-JWT/mdoc/JWT-VC serializations Factoidal has no parser for; vendored as reference data only, per the suite's own yaml comment).

GRDDL stage 1 — 15 pass, 53 fail, 0 skip (out of 68)#

Linked: #301 + the stage-2 census already in this ledger's GRDDL row (landed 7d1c36e4/5c54fd21, ancestor-confirmed). Four named fail-buckets, all planned-family(#301, stage 3 — parked as prioritization not prohibition per owner comment 2026-07-16 on #301: "Do what you outlined but not grddl - more polish/optimize our xpath, xslt ... first"):

• 30 fail-graph-mismatch — consumed XSLT-engine fidelity gaps (side-effect improvement expected from the now-prioritized XSLT burndown — the #301 comment itself notes "~30 of the 53 fails are XSLT-engine fidelity items").
• 6 fail-known-gap-xslt-feature — inline-rdf4/5/6/8/9/10, xml:base/ xml:lang handling in embeddedRDF.xsl.
• 10 fail-no-transformation-discovered — namespace/profile second-document not vendored offline (two-transforms, three/ four-transforms, ns-*, sq1/sq2, glean/hcarda/card5na/ grddlProfileBase3).
• 7 fail-transform-not-vendored — a discovered stylesheet (.sxsl or a redirect-base path) not present offline.

hub_browser_bundle — 234 pass, 1 fail (out of 235)#

• post18 "the C demo transcript quoted in this post is real" — environment(the compiled KaRaMeL C binary formal/fstar/c-output/deltalog/demo/delta_log_demo is listed in formal/fstar/.gitignore:23 and is not built in this checkout — the .c sources are present, the compile step was never run here). Verified: the assertion failure is literally "expected the compiled C delta-log demo binary to exist", not a content mismatch.

JSON-LD 1.1 — 1193 pass, 0 fail, 17 skip combined (of 1210)#

toRdf 461/0/6, fromRdf 53/0/1, expand 379/0/6, compact 244/0/2, flatten 56/0/2 — all by-design(JSON-LD-1.0-version fixtures, named per-ID in each runner; frame/html API suites are NOT RUN, already flagged in the existing ledger row as counting as failed until measured, not a skip). Note: this session found no unmerged branch touching JSON-LD skips despite the task brief's "concurrent agents ... jsonld skips" warning — treated as current, not in flight, but re-verify before citing if a jsonld-* branch lands after this commit.

JSON Schema draft-07 — 712 pass, 0 fail, 58 skip (out of 770)#

All 58 — by-design(keywords outside the validator's slice-1 scope: pattern-regex matching, the format vocabulary, and remote/anchor $ref resolution — bin/jsonschema-runner's own header comment). Per-file: ref.json 38 (remote/anchor $ref), additionalProperties.json 8 + properties.json 8 (pattern-keyed properties), definitions.json 2 + propertyNames.json 2 (anchor $ref / pattern).

npm_package — 160 pass, 2 fail, 1 skip (out of 163)#

🔴 The 2 fails are not covered by any linked design doc or GitHub issue found in this pass — flagged, not confidently dispositioned:

• Test 44 openCottas + queryCottas (raw ABI): COUNT/ASK/SELECT match the fixture's known content — fails at cottas-bytes-store.test.js:94 (ask.boolean expected true, got false).
• Test 50 index.js openCottas/queryCottas: SELECT/ASK/COUNT match the heap-store answers for the same data — same file, line 173.

Both are in npm/factoidal/test/cottas-bytes-store.test.js, which compares the binary .cottas fixture (tests/unit/fixtures/store_capabilities_sample.cottas) against the same data parsed live from tests/local/data/cottas_sample.nq. Best working hypothesis, unverified: environment(the binary .cottas fixture is stale relative to the .nq source it claims to be built from, and needs regeneration) — but this is a guess, not a diagnosed root cause, and no tracking issue exists. Needs a live -v diff before it can be reclassified with confidence.

The 1 skip (test 22, "update: without npm-entry bundle rejects with pending message") is by-design(the negative-path test only applies when no npm-entry bundle is present; this checkout has the bundle, so the SKIP is the correct outcome — # SKIP npm-entry bundle present — pending-message path not reachable).

OWL 2 DL type-inconsistency (tinc) — 124 pass, 3 fail (out of 127 scored), 1 skipped (semantics-scope: RDF-Based only), zero oracle-assisted [re-measured 2026-07-17]#

Linked: #299. Score history since the 118/10 measurement: owl2-double- blocking (inverse-lifted subPropertyOf, 122 pass/6 fail — flips dl-626, dl-627, dl-026, dl-027), a regex-phase3 pattern-facet clash (123/5 — flip: Inconsistent String Pattern), and owl2-tableau-completeness's disjoint-dataproperty numeric-singleton clash (541753cb, 124 pass, 4 fail out of 128 — flip: Inconsistent Disjoint Dataproperties), all ancestor-confirmed on top of the owl2-concrete-domain wave (ed8c4a03). 2026-07-17: WebOnt-Thing-005 recategorized OUT of the DL-regime denominator (harness categorization fix, evidence + rationale below) — scored denominator moves 128 -> 127, pass/fail on the remaining 127 UNCHANGED (124/3, was 124/4 including Thing-005) — a categorization move, not a verdict change; confirmed by a fresh owl_runner type-inconsistency.rdf --regime dl run whose 3 named FAIL lines are identical to the pre-fix run's non-Thing-005 fails (Minus Infinity is not in owl:real, WebOnt-description-logic-502, WebOnt-description-logic-909).

Named residuals (the 3 scored fails):

• WebOnt-description-logic-909 — disputed-fixture, 🧭 owner decision pending (#299 comment 2026-07-16: three independent methods — hand-checked one-element model, z3 sat, clash-free tableau — agree the premise is consistent; the catalog itself marks the test Extracredit, not Approved, with a WG comment that even the approved version was defective).
• WebOnt-description-logic-502 — encoding-not-loaded (root cause re-diagnosed 2026-07-17, #299/#209 nominal-DPLL wave; see the dedicated subsection below). ⚠️ NOT a search-budget problem: the engine returns a definite FAIL/unexpected-consistency (finds a model) with ZERO refuter cap-trips — the 3-SAT constraints never reach the tableau, so there is nothing for a DPLL search to refute. The prior "complementOf/oneOf search + budget-outs" framing was wrong about the layer. The corpus test:description calls it "the classic 3 SAT problem," not a nominals test (the "nominals" label is inherited from an earlier, imprecise classification, per the 2026-07-14 tcon-fail-classification caveat).
• Minus Infinity is not in owl:real — FIXED 2026-07-29 by the XSD value-space decision procedure (XSD.Facets.fst + Tableau.Refute.fst section 5a'/5b'; design note docs/designissues/2026-07-29-xsd-value-space-decision-procedure.md). owl:real is now a first-class dense value space, OWL 2 Syntax section 4.1 keeps the xsd:float value space outside it, and a NegativeDataPropertyAssertion removes the one remaining admissible value. type-inconsistency is 126 pass, 1 fail (out of 127).

WebOnt-Thing-005 disposition (2026-07-17, #299) — outcome (a): harness categorization bug, FIXED, not an engine gap. Catalog evidence, quoted verbatim from third_party/testing/owl/type-inconsistency.rdf (TestCase WebOnt-Thing-2D005):

<test:description>The extension of OWL Thing may not be a singleton in OWL Full.</test:description>
<test:status rdf:resource="&test;Proposed" />
<test:semantics rdf:resource="&test;RDF-BASED" />
<test:species rdf:resource="&test;FULL" />
<test:species rdf:resource="&test;DL" />

plus a same-subject owl:NegativePropertyAssertion immediately after the TestCase block:

<owl:sourceIndividual rdf:resource=".../TestCase-3AWebOnt-2DThing-2D005" />
<owl:assertionProperty rdf:resource="&test;semantics" />
<owl:targetIndividual rdf:resource="&test;DIRECT" />

i.e. the catalog declares test:semantics = RDF-BASED and explicitly DENIES test:semantics = DIRECT for this TestCase — an OWL Full domain-size/comprehension test ("owl:Thing's own extension may not be a singleton"), not a Direct-Semantics DL test, despite also carrying test:species DL alongside FULL (species and semantics are orthogonal facets in this catalog — species says which syntactic profile the document parses as, semantics says which model theory the inconsistency claim is made under).

Census (2026-07-17 sweep, all 128 InconsistencyTest cases in type-inconsistency.rdf, script-verified): 127 carry BOTH test:semantics DIRECT and RDF-BASED; exactly 1 — WebOnt-Thing-005 — carries RDF-BASED alone (and is the only one with a semantics-NegativePropertyAssertion). Not a systemic issue elsewhere in this catalog — a single named case.

bin/owl-runner/owl_runner.ml's run_inconsistency_test (both RL and DL regime — this catalog is always run --regime dl per generate-report.sh) implements Direct Semantics only (RL closure -> Tableau.Refute -> RL model theory; no RDF-Based domain-size reasoning over owl:Thing's own extension), and — before this fix — never consulted test:semantics at all for InconsistencyTest scoring (unlike run_negative_entailment, which already gates its closure mode on owl_semantics_mode_for). Scoring Thing-005 as a DL fail was therefore a harness categorization bug: fixed by gating run_inconsistency_test on owl_semantics_mode_for info = RDF_Graph_Executable.owl_semantics_rdf_based and returning a new Skip_semantics_rdf_based_only outcome, counted outside the pass/fail denominator with its own labelled reason line — same pattern as the pre-existing Skip_functional_syntax_only. RDF-Based (OWL Full) domain-size/comprehension reasoning remains a genuine, undispatched engine gap; the fix is scope correction, not new capability — if the engine later grows RDF-Based semantics support, Thing-005 is the test to re-enable it against.

Scores (bin/linux-x86_64/owl_runner, this worktree, 2026-07-17):

• tinc before: 124 pass, 4 fail (out of 128), 0 skipped.
• tinc after: 124 pass, 3 fail (out of 127 scored), 1 skipped (semantics-scope: RDF-Based only) — flip: WebOnt-Thing-005 FAIL/unexpected-consistency -> SKIP/semantics-rdf-based-only; all other 127 verdicts unchanged.
• type-consistency.rdf --regime dl (tcon, unaffected by construction — the fix only touches run_inconsistency_test, tcon uses run_consistency_test): 352 pass, 0 fail (out of 352), zero unexpected-inconsistency — gate held.
• floor rdf-turtle: 313 pass, 0 fail (out of 313) — gate held.

dl-502 nominal-DPLL wave (2026-07-17, #299/#209) — analysis-only; root cause re-diagnosed#

🧭 Decision for the next wave: dl-502 is blocked upstream of the refuter, not inside it. This wave set out to extend the Tableau.Refute DPLL machinery with nominal identify-branching so the refuter could search the ~2^9 boolean assignments of WebOnt- description-logic-502 ("the classic 3 SAT problem"). Investigation found that no amount of search extension can flip this test today, because the SAT instance is never loaded into the tableau. Landed as analysis-only (no engine change) per the zero-movement-scaffold ban — adding a nominal-branching tier that fires on zero currently-loadable tests, and adds wall-clock on the DPLL hot path, would be scaffold.

The fixture (worked on paper). TorF is a class defined by TEN owl:oneOf enumerations: {T,F} (with T owl:differentFrom F) and {plus_k, minus_k} for k=1..9. Two owl:oneOf on one class force their enumerations to denote the SAME set, so {plus_k,minus_k} = {T,F}; with T ≠ F this is a 2-element set, hence plus_k ≠ minus_k and each of plus_k/minus_k is T or F — the truth value of boolean variable k (plus_k = T ⇔ var k true). Then T is asserted rdf:type of 45 anonymous owl:oneOf {la,lb,lc} classes — each a 3-literal clause T = la ∨ T = lb ∨ T = lc, satisfied iff one literal equals T. The 9-var/45-clause 3-SAT instance is UNSAT, so the premise is inconsistent.

Measured current behaviour (bin/linux-x86_64/owl_runner type-inconsistency.rdf --regime dl, this worktree, 2026-07-17, freshly rebuilt from source): dl-502 = FAIL/unexpected- consistency; tinc run completes in 46.26s with ZERO refuter cap-trip / owl_closure_timeout messages. The refuter reaches quiescence and reports a model — it is NOT exhausting the ~5s / 20000-fuel budget. This is the decisive datum: dl-502 is not budget-limited.

Root cause (two layers, both verified by source inspection):

1. Tableau.fst:303 — parse_class_expr reads a class's enumeration via find_first_object g s owl_oneOf, i.e. only the FIRST owl:oneOf. TorF's other nine enumerations are dropped. And no individual is asserted rdf:type TorF, so even the first enumeration is inert.
2. OWL.Closure.fst has NO owl:oneOf rule at all (grep: zero matches). The RL closure never materialises the set-equality between two enumerations of one class.

Consequently the {plus_k,minus_k} = {T,F} variable constraints and the plus_k ≠ minus_k distinctness never reach the tableau. T's 45 clause-labels reference plus_*/minus_* individuals that carry no other constraint, so T is freely identifiable with any of them → a model always exists → correct-given-what-was-loaded "consistent" verdict → InconsistencyTest FAIL. The 3-SAT structure is invisible to any search that operates on the loaded state.

Design for the flip (future wave — three parts, in order):

• F1 — multiple-owl:oneOf set-equality materialisation (the blocker; a loader/closure change, foundational). For a class C with ≥2 owl:oneOf enumerations L1, L2: CEXT(C) = set(L1) = set(L2), so soundly inject, for each m ∈ L2, m rdf:type oneOf(L1) (and symmetrically) — fully general, no cardinality needed. ⚠️ Soundness-subtle addendum: the plus_k ≠ minus_k distinctness that makes each pair a genuine boolean is a CARDINALITY consequence (|{plus_k,minus_k}| = |{T,F}| = 2 because T ≠ F), NOT of the membership injection. Without it the search admits the degenerate plus_k = minus_k = T model in which every literal equals T, so every clause is trivially satisfied and the instance looks consistent — the refutation would be unsound-incomplete. F1 must emit the distinctness (from a 2-element base whose two members are provably distinct), which makes it a narrow, pattern-specific rule of exactly the kind CLAUDE.md's "known sound-but-narrow rewrites" section cautions about.
• F2 — nominal identify-branching in Tableau.Refute (the O-rule as a BRANCHING rule, not only the existing clash rule). Extend find_identify_nodes / a new find_nominal_branch to offer, for a node labelled CE_OneOf members, the identifications i = m for each m ∈ members; identify_branch already implements the AND-semantics (TClash iff every member's identification closes) and the threaded-budget measure. Model-theoretic soundness: i ∈ {m1..mn} entails i = mj for some j in every model, so refuting every j refutes the branch. This piece is sound and self-contained but moves zero tests without F1, so it must NOT land alone.
• F3 — DPLL convergence on the loaded 9-var/45-clause instance within budget. Unit propagation (a plus_k/minus_k forced by a decided sibling) and clash-first pruning already exist for the union encoding (find_union_nodes_scan, section 8b) and would need the nominal analogue. ⚠️ Wall-clock risk: this is the tinc hot path (measured 46s for 128 tests); a nominal search that reaches the ~5s per-test cap on dl-502 (and its sibling dl-504) alone adds ~+11%, close to the +15% gate — F3 must be measured on the full manifest, not just dl-502, before landing.

Gate evidence for this analysis-only landing (no engine change → all gates hold by construction; measured on the fresh worktree binary): tinc 124 pass, 3 fail (out of 127), 1 skip — 46.26s; tcon 352 pass, 0 fail (out of 352), ZERO unexpected-inconsistency — 389.3s; floors rdf-turtle 313/0, w3c_runner --rdf12 29/0.

OWL 2 DL positive entailment (PE) — 193 pass, 11 fail, 2 skip (out of 204) [re-measured 2026-07-29]#

Linked: #298, #209. Every one of the 11 residuals is named, with a per-test reason and disposition, on the public conformance page docs/web/conformance/owl2.md (/web/conformance/owl2/) — that page is the source of truth for this row, and its disposition table (10 by-design, 2 planned-family, 3 disputed-fixture across all OWL catalogs) is current. The per-item lists below are a 2026-07-28 snapshot kept for the reasoning they record; where they disagree with the conformance page, the page wins. Items struck from them on 2026-07-29: WebOnt-I5.24-002/-003/-004 and WebOnt-SymmetricProperty-002 now PASS (rdfs:range/rdfs:domain metamodeling + extensional owl:SymmetricProperty, OWL.Closure.fsti Groups E(g)/E(h) and comp-w6/w7/w8).

• 20 by-design. 17 of them carry test:species FULL plus an explicit owl:NegativePropertyAssertion in the catalog denying test:species DL, i.e. the W3C suite states they are not OWL DL entailments: WebOnt-Class-001/002/003, WebOnt-FunctionalProperty-003/004, WebOnt-InverseFunctionalProperty-003/004, WebOnt-complementOf-001, WebOnt-equivalentProperty-005, WebOnt-I5.3-014/015, WebOnt-I5.8-017, WebOnt-extra-credit-002/003/004, rdfbased-sem-restrict-maxqcr-inst-obj-one. Plus WebOnt-I5.8-004 (test:status test;Extracredit) and WebOnt-I5.5-005 / WebOnt-I5.26-010 (OWL 1 Full comprehension, scope.md). ⚠️ The owner approved a flag-gated OWL Full engine mode on 2026-07-28 ("If we can do it - eg under a flag or engine mode or api if not directly alongside other OWL variants, then sure"); when that mode exists, the 17 species-denied tests become planned work rather than by-design.
• 10 planned-family. WebOnt-I5.21-002 (property-characteristic propagation); WebOnt-cardinality-001/003 (owl:cardinality shorthand); WebOnt-I5.8-010 (datatype value-space intersection); WebOnt-oneOf-004 (owl:dataRange over literals); WebOnt-unionOf-003/004 (extensional unionOf).
• 1 dependency-blocked. WebOnt-equivalentProperty-004 — premise property-inclusion outside the refuter's role hierarchy (#298).

The former grouped breakdown against 69 fails is superseded: the tableau-classification wave (5eed66e) and the closure-rule wave (0e9406b) closed the whole DL98 stress group (description-logic-201/202/206/208/661/662/901/903, Consistent-but-all-unsat), the AllDifferent/oneOf/unionOf comprehension rules, the vocabulary axiom family, and rdfsext-domain/range-subprop (previously listed here as dependency-blocked — it now passes).

2 skips: Qualified-cardinality-boolean, Qualified-cardinality-restricted-int — by-design(functional-syntax-only fixtures; the parser targets RDF/XML premises/conclusions, not OWL functional syntax).

OWL 2 DL negative entailment (NE) — 22 pass, 1 fail (out of 23)#

WebOnt-imports-002 (FAIL/unexpected-entailment) — planned-family(import-resolution edge case; same family as the OWL 2 row's syntax-dl species misses on imports-004/012/014). Confirmed unchanged across every OWL wave this session traced (#298/#299 comments both say "NE 22/1 unchanged").

OWL 2 RL positive entailment — 28 pass, 2 fail (out of 30)#

WebOnt-I5.5-005, WebOnt-I5.26-010 — by-design(documented in docs/designissues/2026-07-03-owl-rl-pe-fails-fix-sketch.md as OWL 1 Full comprehension entailments deliberately outside the OWL 2 RL closure). The suite moved from 27/2/1 (that doc's "end state") to 28/2/0 — the 1 skip it left open has since resolved to a pass; the 2 fails are unchanged.

OWL 2 DL syntax-dl species — 319 pass, 2 fail, 2 skip (out of 323)#

⚠️ This ledger's existing OWL 2 row headline ("302 pass, 2 fail (out of 304 scored), 19 skipped") is stale — the committed formal/fstar/ocaml-output/owl_syntax_dl_results.log confirms 319/2/2 today (17 of the 19 former functional-syntax-only skips now run and pass; likely the "OWL profiles" work the task brief flagged as concurrent). The 2 fails are unchanged from the ledger's existing prose:

• FS2RDF-literals-ar — disputed-fixture(RDF/XML premise our parser rejects; existing ledger text).
• WebOnt-I5.5-005 — disputed-fixture(header-less conclusion; the graph-identical I5.5-006 sibling is judged FULL by the corpus, so no graph-level function separates them; existing ledger text).

OWL 2 profile EL / QL — 118 pass, 2 fail (of 120) / 85 pass, 2 fail (of 87) [re-measured 2026-07-28]#

🧭 Both rows were new in docs/test-results/latest.json (commit 0f4090d5, ancestor-confirmed) with no matching row in this ledger and no GitHub issue found referencing either suite name at the time. Since then, an OWL.Closure.fsti fix (2026-07-28) flipped 10 EL tests + 2 QL tests that latest.json had not yet picked up (stale dashboard, no code change needed) — regenerated via formal/fstar/generate-report.sh; numbers above re-confirmed by a fresh catalog re-run 2026-07-28 (EL 118 pass, 2 fail out of 120, 1 skipped; QL 85 pass, 2 fail out of 87).

✅ Triaged 2026-07-28 (#308). Remaining 2 fails on each catalog are WebOnt-I5.26-010 + WebOnt-I5.5-005, both by-design(OWL 1 Full comprehension-principle entailments, deliberately outside the OWL 2 RL/EL/QL closure) — the disposition docs/claude-rules/scope.md's "OWL 1 Full comprehension-principle entailments" section already records, now applied per-catalog. Public statement with per-test reasons: docs/web/conformance/owl2.md. The EL/QL 1-skip difference is WebOnt-Thing-005 (test:semantics RDF-BASED only, DIRECT explicitly denied), present in the EL catalog and absent from QL.

QUDT integrity — 0 pass, 0 fail, 29 skip (out of 29)#

Re-run fresh this session (./bin/linux-x86_64/qudt_runner, matches the committed log exactly): all 29 skips are accounted for and the suite is not silently dead — it parses 131,169 triples + runs 9-for-9 on the sibling qudt_user_shapes suite in the same process. Not a 🔴 finding.

• 26 — environment(per-shape wall-clock budget: the 420s cap trips before wide-target shapes finish; docs/designissues/2026-07-10-qudt-scoping.md names this a perf-program work item — indexed SPARQL-constraint evaluation, not an engine defect).
• 2 (Concept-label-IRI-Shape, UnitsHaveRdfClassQudtUnitTriple) — dependency-blocked(sh:SPARQLTarget, #181 — the one acknowledged SHACL gap).
• 1 (UniqueSymbolTypeRestrictedPropertyConstraint) — by-design (sh:deactivated true, excluded by SHACL semantics itself).

RIF Core — 46 pass, 1 fail, 3 skip (out of 50)#

All named in .github/test-suites/rif.yaml's own header, re-verified against the committed log:

• RDF_Combination_Constant_Equivalence_4 — disputed-fixture(a malformed xsd:string datatype IRI present in both the official Core_v1.22.zip and the archived W3C wiki source; not fixable without breaking correct datatype-IRI semantics).
• Builtins_List, NestedListsAreNotFlatLists — dependency-blocked(RIF List terms are not modelled in RIF.Core.Syntax.fst).
• Builtins_Time — dependency-blocked(the ~60-builtin date/time/ duration family; only the EBusiness_Contract slice is implemented).

rml_io — 17 pass, 1 fail, 55 skip (out of 73)#

Re-run fresh this session (./bin/linux-x86_64/rml_runner --io -v, matches the committed log's 17/1/55 exactly):

• RMLSTC0009a — expects an empty dataset (error=true) but the runner produces 3 triples — planned-family(no dedicated issue found; same shape as an unhandled error-detection case in the rml-io source-tests section, needs its own triage — flagged, not fabricated a root cause).
• 41 — by-design(RMLTTC0* logical-target/output-serialization tests, out of scope per the 2026-07-05 rml program plan Stage 9).
• 7 — dependency-blocked(XPath/XML reference formulation not implemented).
• 4 — dependency-blocked(compressed source, needs decompression).
• 1 — dependency-blocked(relational SQL2008 source, needs an RDBMS).
• 1 — dependency-blocked(non-UTF-8/UTF-16 source encoding not supported).
• 1 — dependency-blocked(SPARQL-over-RDF iterator not implemented).

ShEx validation — 1182 pass, 0 fail (out of 1182)#

start2RefS2 — disputed-fixture, linked shexSpec/shexTest#43 (closed 2023, unreconciled upstream — start2RefS2.json/.ttl use predicate p1, start2RefS2.shex uses p2; per tests/shex-shexj-twins/README.md, our ShExJ-first runner scores against the JSON and carries the mismatch as a fail rather than patching a vendored fixture).

tests_unit — dashboard 35 pass, 6 fail (of 41); fresh run 45 pass, 0 fail (of 45)#

environment(the dashboard capture ran against a checkout with gaps this session reproduced the failure signature for: this worktree's third_party/testing/w3c/ submodule was NOT populated until tools/ensure-test-env.sh was re-run against the actual worktree path — an initial run against a sibling checkout's root had falsely reported "ok". With the submodule missing, 4 tests (owl_direct_pipeline_timing, owl_rl_bisect, owl_rl_closure_diff, owl_rl_sequenced) fail on Sys_error("...simple.ttl: No such file or directory") — the same ENOENT-from-missing-submodule shape workflow-gotchas-debugging hazard #15 describes, though this session's reproduction hit 41/4 not the dashboard's 35/6, so the dashboard's exact capture environment was likely worse-broken still, e.g. opam switch not activated per iron rule #12). After tools/ensure-test-env.sh + eval $(opam env --switch=fstar --set-switch), the suite is clean: 45 pass, 0 fail (out of 45). The dashboard number needs republishing from a hazard-15-clean checkout — flagged for the publication step, not fixed here (docs-only commit, no rebuild triggered).

xml_conformance — 1414 pass, 0 fail, 1171 skip (out of 2585)#

The runner's own "HONEST BREAKDOWN" (re-verified against the committed log), five categories, all by-design(explicit Stage-A scope limits, not silent gaps):

• 386 — valid docs rejected at a Stage-A DTD boundary (markup entity / external subset / DTD construct beyond the WF slice).
• 332 — not-wf violation is in the DTD internal subset (parsed-but-not-validated — Stage-A scope limit).
• 416 — other (DOCTYPE-external-on-valid, encoding-name edge cases, invalid/error-by-design, external-entity exemption, file-not-found).
• 37 — out-of-profile (not-wf holds only under XML 1.1 or Namespaces in XML; this parser is XML 1.0 non-namespace).
• 0 — vacuous (retired by Stage-A DTD support; the runner asserts this stays 0).

XSLT 1.0 — 79 pass, 9 fail (out of 88)#

Linked: #302. Landed via xslt-burndown-1 (fd5fc372, ancestor-confirmed): "XSLT 79 pass, 9 fail (of 88), was 75/13" — this ledger's dashboard-adjacent headline is stale at 75/13. Named residuals, from #302's 2026-07-16 comment:

• copy-0601, match-045, namespace-4101, namespace-4501, namespace-4801, node-1601 (comment counts this family as "5" but names 6 — quoted as written per the steers-quoting discipline rather than silently corrected) — 🟡 in flight: branch xslt-namespace-nodes, dispatched per the owner's 2026-07-16 XSLT-polish priority; not yet merged into this worktree's HEAD (branch not found in git branch -a as of this measurement).
• copy-2601, select-1001 — planned-family(#302, document-node prolog/epilog comment handling).
• id-016 — planned-family(#302, DTD-ID id() function).

All nine are confirmed in-scope (none PSVI/XSLT-2.0-schema-aware; #302 explicitly excludes that boundary, tracked separately at #190).