Ontology Semantic Versioning - Status

Unit ID: ontology-semantic-versioning Status: Slices 1-4 implemented, at the human’s explicit direction to implement the plan autonomously (2026-09-25). ADR-A86 itself remains Proposed — ratifying it is a separate human action this implementation does not take on its own behalf. Last updated: 2026-09-25 Trigger: human request, 2026-09-25 (“implement the semantic versioning plan autonomously”) Plan: ontology-semantic-versioning.md Sketch: ontology-semantic-versioning.md ADR: ADR-A86, Proposed

Current position

All four slices are implemented. The human’s instruction to implement is treated as confirmation of the ADR’s recommended defaults (the 0.2.0 baseline, owl:versionIRI as the sole authoritative signal), per the working pattern this repository has followed for other units executed the same way (vocabulary-temporal-binding): implementation proceeds on the documented recommendation, ADR ratification itself stays a distinct, later human action.

Slice 1 — governance records

Done in a prior session: ADR-A86 (Proposed), the sketch, this plan, this status record, and the ADR-index/INDEX.md entries.

Slice 2 — developer/agent guidance

• New docs/architecture/ontology-versioning-policy.md: the MAJOR/MINOR/PATCH table, the versioning-unit rule, the import-pinning cascade checklist, where the number physically lives per layer, and a worked example from this repository’s own history (vocabulary-temporal-binding’s MINOR addition).
• CONTRIBUTING.md “Where new content belongs” and docs/architecture/ontology-architecture.md §2 both cross-reference the new policy document.
• mise run topology:links was run after these additions: it reports the same pre-existing broken links this repository already carried (none of them in a file this unit touched) and zero new ones.

Slice 3 — baseline reset

Every document the sketch’s inventory named was reset to 0.2.0, and every owl:imports reference to one of the old IRIs anywhere in ontology/ was cascaded in the same change. Full before/after table: ontology-versioning-policy.md.

Deviation from the plan’s literal text, recorded here because it matters: the plan said to make this edit “in the README’s own turtle-spec block and regenerate via tools/literate_extract.py.” Running the tool first (in --check mode, before touching anything) showed it already fails for every one of the seven core layers today — Foundation’s, Vocabulary’s, and Party’s committed spec/*.ttl predate the extraction tool’s conventions entirely (different, hand-authored Turtle formatting, not a turtle-spec concatenation), and the remaining four layers’ README turtle-shapes blocks don’t agree with the tool’s --shapes argument contract. Running the tool in write mode would have overwritten large amounts of unrelated, already-drifted content with a full regeneration — well outside this unit’s scope and a real risk to content this unit has no mandate to touch. The version/import lines were therefore edited directly, by hand, in both the README (where one mirrors the header) and the generated file, keeping the two identical for that one block without invoking the extractor. This gap — extraction drift across the whole 7-layer literate-spec set — is now recorded as a discovered, separate problem in the policy document, not silently absorbed into this unit’s own scope.

Also found and fixed as a genuine importer, beyond the sketch’s original inventory: tools/surface/src/surface/namespaces.py’s SURFACE_ONTOLOGY constant, which the Surface compiler (compile.py) stamps into every generated surface’s owl:imports. Updated to the new .../lattice/surface/0.2.0 IRI in the same change.

Deliberately left alone: MORK’s own example/test fixtures (Meta_RML/GPTAttempt.ttl, Meta_RML/RMLMap.ttl, Zoo/Petstore.ttl, Zoo/UncertainMappings.ttl, test/data/Demo.ttl) that owl:imports MORK’s bare, unversioned ontology IRI. That IRI still exists and is still declared a owl:Ontology — adding a versionIRI to Mork.ttl did not invalidate it — and converting five example/test fixtures from “import whatever Mork currently is” to “pin an exact version” is a bigger behavioural change than a version-number reset, not something this unit’s cascade obligation covers. mork/spec/Executable.ttl, the one layer-level (non-example) importer of Mork, was likewise left importing the bare IRI, matching that same existing convention rather than introducing it as the sole exception.

Slice 4 — narrow enforcement tooling

tools/ontology_version_check.py + mise run check:ontology-versioning (wired into the aggregate check task). Validated two ways: a direct function-level mutation probe (flags “content changed, owl:versionIRI literal unchanged”; stays silent when both change together — both confirmed manually in this session) and a real run against this unit’s own working-tree diff, which reported zero unbumped files across all 29 in-scope ontology documents (every file this unit touched changed its version at the same time, and every file it left alone was left alone).

Decisions made during implementation (flagged for human review)

• The baseline number: 0.2.0. Applied as recommended; not separately re-confirmed by a human before this session’s implementation, per the explicit “implement autonomously” instruction.
• owl:versionInfo dropped, not kept, on applied/insurance/contract.ttl. owl:versionIRI is now that document’s sole authoritative version signal; its owl:priorVersion was updated to point at 0.1.1 (its immediate predecessor) rather than removed.
• applied/insurance/contract.ttl’s namespace family (neuro-semantic/insurance/..., outside lattice/) was left exactly as it was, per the plan’s explicit non-coverage — only its version number and its two disagreeing version signals were touched.
• MORK’s Mork.ttl and SPC’s spc.ttl first owl:versionIRI values use each ontology’s own existing namespace family (http://www.nebularis.org/ontologies/Mork/0.2.0 and http://example.org/spc/0.2.0 respectively), not the lattice/ tree — MORK is documented elsewhere as a pre-existing, independently-styled vocabulary, and SPC’s namespace harmonisation to nebularis.org (ADR-A62) is itself still Proposed, not Accepted; assigning SPC a lattice/-based versionIRI now would have pre-empted that still-open decision.
• Import-pinning strategy: kept exact, as recommended. No dependency-range mechanism was introduced.
• The Slice 4 check’s home: a new standalone check:ontology-versioning task, not folded into topology:links/topology:preflight — those two check documentation cross-references and the ADR-A77 migration manifest respectively, a different concern from ontology-file version hygiene.

Human validation gate

Not yet performed. Before treating this unit as fully closed:

1. Ratify ADR-A86 (or amend it) — its status remains Proposed.
2. Spot-check the baseline-reset table in ontology-versioning-policy.md against a few of the actual files, and confirm the applied/insurance namespace-family question’s disposition (left alone, as implemented).
3. Run mise run check:ontology-versioning in an environment with the repository’s full git history available (this session validated it in-sandbox against the working tree, which is the same code path mise uses).
4. Decide, separately, whether the newly-discovered tools/literate_extract.py drift across the seven core layers (recorded in the policy document) becomes its own remediation unit.

Blockers

None recorded against this unit’s own four slices — all are implemented. ADR-A86 ratification and the extraction-drift follow-up (both above) are open items for a human, not blockers to what this unit set out to do.