S01.1: Complete the semantic baseline

Ownership and readiness

Parent session: #148. Epic: #147. Status: Complete through PR #221; independently reviewed and qualified on merged main. Findings: CM-01, CM-15. Maintenance #209 and roadmap refresh #217 are complete; this is the first remaining architecture task.

Delivered baseline

#141 is closed and PR #145 is merged. v1.1.2 includes the sequence correction and singleton response normalization. Reuse test/unit/compiler-compositor-occurs.test.ts, the sequence-occurrence-wrappers.wsdl conformance fixture, test/integration/sequence-occurrence-transport.test.ts, test/helpers/occurrenceTransport.mjs and test/conformance/registry.ts.

These artifacts are starting evidence. Their presence does not complete the independent reference validation or broader semantic characterization required here. v1.1.3 and v1.1.4 are now published; their publication is separate from the original v1.1.2 fix.

Bounded case manifest

Read repository instructions, #141/#145/#210 and current compiler/client code. Pin the tested main revision; the October 9 planning observation is b633cd1561ccb2689134a7289cf48bb097b0d1f1. Link or extend the sanitized original reproduction instead of creating duplicate coverage.

Commit the minimal manifest before expanding the investigation. Cover these case families, reusing a fixture across families when it gives independent assertions:

• Repeated and optional wrapping sequences, including the exact original enclosing-sequence reproducer.
• Simple choices and their ordering boundaries.
• Wildcards and the limits of the current projection.
• Zero-occurrence particles with unbounded descendants.
• Required choices with one selection and with multiple selections.
• Empty alternatives and their observable value representation.
• Finite count gaps, including the 2-or-4 case.

For each valid schema, include an accepted instance and a targeted rejected instance where that family permits one. If a proposed schema is itself illegal, record the schema rejection and add a legal equivalent for the intended semantic question when possible. Do not confuse invalid schema construction with an implementation rejecting a valid instance.

Each manifest entry records a stable case ID, CM finding, fixture/provenance, schema legality, reference instance results, actual catalog/TypeScript/OpenAPI results, applicable SOAP/gateway evidence, intended semantics, and disposition with a downstream owning issue. Record unsupported projections explicitly instead of manufacturing passing expected output.

Reference runner and evidence boundaries

Select and pin an independent XSD 1.0 validator; document setup, exact version, offline inputs and a committed runner command. The runner reports schema compilation and instance validation separately. It must not fetch schemas from the network during a validation run.

Provide a small deterministic reference subset in a required CI lane and a full-manifest qualification command using the existing scripts/workflows. Record which command owns each lane and how its tool is provisioned. A missing validator or fixture fails a required reference lane; running ordinary characterization tests without the tool is not a passing reference result.

Keep current-output characterization and intended semantic assertions distinct. Separate contributor-reported production acceptance from independently reproduced evidence. Reference disagreements and implementation bugs get a reproducer and a triage record; they do not automatically expand this task into later architecture implementation.

Acceptance and tests

• The bounded manifest covers every agreed family and links the exact report to reusable fixtures.
• Schema legality and instance validity have separate reproducible reference outcomes.
• Current output, intended semantics and unsupported behavior have distinct recorded expectations.
• Pinned tool setup, offline inputs and required CI/full qualification commands are documented and executed.
• Missing reference tooling fails the required reference lane instead of silently skipping it.
• Existing CI passes; tracked limitations are not presented as fixed coverage.
• Every remaining gap has fixture provenance, CM traceability and a downstream owner.

Close when the agreed evidence is reproducible and gaps are assigned. Fixing all observed semantic differences is not an acceptance requirement; any necessary focused evidence fix needs an explicit bounded disposition in #148.

Code quality and maintainability

Follow the code-quality contract. Keep the reference adapter test-only and small, reuse the existing registry and transport harness, and document the manifest format once. Avoid duplicating the public support matrix or introducing a generic validation framework.

Delivery

Provide commit/PR, manifest and fixture paths, commands/results/unrun checks, observed consumer impact and inputs for #167. Update #148's handoff. Public model redesign and release delivery belong to later work.

Delivered acceptance, October 9, 2026

PR #221 merged at b86c653dfa0672231ca9b7cbc56d15ef6ae18bae; its tree bf3d8e7caa18a12ac204b061ad0ccc6e8563c6fb matches tested head 94340c89b384831b411edce61c1e9994119b51d6. Source baseline: b633cd1561ccb2689134a7289cf48bb097b0d1f1, published v1.1.4. The manifest was committed before broad investigation at 26cb21e.

The manifest is test/conformance/semantic-baseline.json; registry-linked fixtures and instances are in test/conformance/fixtures/xsd/compositors/ and the reused sequence fixture. Runner: scripts/semantic-reference.mjs and test/conformance/reference/validate.py. Setup and reproduction: docs/content-model-baseline.md.

npm run reference:setup provisions lxml 6.0.2/libxml2 2.14.6, xmlschema 4.2.0 and elementpath 5.0.4. npm run test:reference passed six cases/15 instances; npm run test:reference:full passed 14 cases/40 primary instance checks, with 13 schema-accepted cases, one schema rejection and one explicit secondary oracle disagreement. Full reference was also executed on clean merged main; its report identifies the exact merged revision.

npm run ci passed on the clean PR head with Node 24.19.0, npm 11.9.0, TypeScript 6.0.3 and Vitest 5.0.3: 670 tests/55 files, all conformance stages, required reference subset, docs, pipeline smoke and installed-package consumers covering 73 TypeScript artifacts. Focused characterization/missing-tool tests and conformance scoped typechecking passed. Independent read-only review approved after findings were resolved. Hosted Node 24/26 and CodeQL passed on the final head; merged-main CI passed.

Known limits remain explicit: ancestor choice/all/disabled/wildcard boundaries, ordered group relationships, repeated choices, empty alternatives and count gaps. Manifest dispositions assign analysis, legality, matching, projection, encoding and HTTP validation to #178/#179/#180/#182/#188/#198. libxml2's disabled-row acceptance is independently contradicted by xmlschema; #181 owns CM-15 oracle follow-up. No production repair or support/default activation was introduced.

No required local checks were skipped. IDE inspections were unavailable; repository-local checks were used. Historical commits, private vendor production validation, package publication/release preflight and S02 feasibility probes were not rerun. #167 consumes this delivered evidence; #148 remains open until that follow-up and its combined gate pass.

Operational pin follow-up

PR #223 corrected the test-only lxml pin to 6.1.0 after CVE-2026-41066 surfaced, preserving libxml2 2.14.6, all case inputs and literal semantic expectations. Fresh final full CI/reference qualification and merged-main checks pass at 86089d73d02f39b04c0ae2eac8522efc2b3d47c8. Use the current setup documented in docs/content-model-baseline.md; the 6.0.2 results above remain the exact historical #166 delivery evidence.