Description

Switches manifest, georeference, envited-x and ositrace to artifacts generated
from LinkML, and bumps them to manifest v6, georeference v6, envited-x v4 and
ositrace v7. The hand-written OWL, SHACL and contexts of these four are replaced;
the prior versions stay published under their w3id IRIs and are named by
owl:priorVersion. The importing domains move to the new versions in place.

The models follow what the hand-written files meant: each construct is traced to
its intent and modelled with the LinkML mechanism for it, and the generators are
the only thing that runs — no pre- or post-processing. What LinkML cannot state is
recorded in linkml/GAPS.md.

The generators are the reviewed ENVITED-X pipeline of the ASCS-eV LinkML fork,
pinned to the tag envited-x/2026.10.14 (536d6f194).

Commits

Based on v0.5.1.

1. fix(validators) — per-resource inference types blank-node objects by rdfs:range, as the default mode does.
2. fix(report) — the two hash-dependent parts of pySHACL's generic messages (the order of sh:in values; where a shared blank node's rendering is cut off) are rendered in a fixed form, so report snapshots are reproducible. The structured violations take the same normalization.
3. build(linkml) — linkml pinned to the tag envited-x/2026.10.14; one build script for every LinkML domain; openlabel-v2 rebuilt with it.
4. ci — the jobs that run just generate check out the pinned Gaia-X sources.
5. test(harness) — scripts/compare_artifacts.py measures a candidate against the committed artifacts.
6. feat(ontology) — the ontology-header profile (below).
7. feat(ontology)! — the switch-over and version bumps; the registry, catalog and properties overviews list the new versions, and the docs describe the generated domains.

Ontology headers: the ontology-header profile

The hand-written ontologies share one header convention: a permanent identifier,
the repository and release they come from, their version and prior version, their
shapes, related resources and their authors, with resources named by IRI. LinkML
gives a schema-level annotation a literal unless a metaclass the schema
instantiates declares its range — the route the LinkML maintainers recommend
(linkml/linkml#2892, #3379) and that the fork implements (ASCS-eV/linkml#33).

linkml/ontology-header/ is that metaclass as a schema profile, the pattern of
LinkML's implements/instantiates guide and of linkml-microschema-profile, which
downstream schemas import:

• OntologyHeader declares owl:priorVersion, prov:wasDerivedFrom,
  dcterms:references and dcterms:publisher: range nodeidentifier,
  implements: [owl:AnnotationProperty]; the release and the publisher are required.
• Each domain imports the profile and instantiates: [ontology-header:OntologyHeader].
  Its header values are IRIs on declared annotation properties (OWL 2 requires the
  declaration in the import closure), and its OWL states no logical axioms about them.
• The domains build with --no-mergeimports / --exclude-imports, so the metaclass
  stays in the profile: a domain's OWL has owl:imports of the profile and types
  the ontology ontology-header:OntologyHeader; no domain class, shape or context
  term is added.
• The profile's shapes check those headers: every generated ontology conforms, a
  literal publisher or a missing release does not (tests and fixtures included).
• https://w3id.org/ascs-ev/envited-x/ontology-header/v1 resolves through the
  existing generic w3id rules.

owl:versionIRI is not stated. Every hand-written ontology used its version IRI as
its ontology IRI, and the schema id is that IRI, so the triple only repeated it;
OWL 2 makes the version IRI ontology-header structure, not an annotation, and the
reviewed generator rejects it as one. The registry, docs build and class pages use
the ontology IRI as version IRI, so no published version IRI changes. OWL 2 §3.3,
OBO FP-004 and FOOPS VER1 prefer a permanent ontology IRI with a distinct version
IRI; LinkML has no field for that (GAPS G22, linkml/linkml#768).

Generator features this needs

All in the ASCS-eV LinkML fork, each its own PR, combined on the pipeline branch (ASCS-eV/linkml#14) after review:

gen-shacl --inlined-as-node was rejected in review and is not used: what a nested
value must satisfy is stated with range_expression on the slot that holds it.

Modelling

• Nested values. A class range gives sh:class; the range class's shape reaches
  a nested value only when the value is typed (in the data, or by rdfs:range under
  the validator's RDFS inference). What a link, its file metadata or its dimensions
  must satisfy regardless of type is a range_expression. The JSON-LD artifact
  condition is a class expression on ArtifactLink and on the artifact properties;
  the license, manifest-reference and ENVITED-X vocabulary narrowings sit on the
  owning properties.
• openlabel-v2. Built by the same script with --no-use-native-uris: its OWL
  declares the schema.org and Commons IRIs (schema:QuantitativeValue,
  schema:minValue, …) its SHACL and context already use. Rule 17 ("DrivableAreaEdge
  includes EdgeNone") is has_member; a bare equals_string on a multivalued slot
  means every value and the reviewed converter skips it. Rules on a class also
  constrain its subclasses' shapes.
• Nested OWL restrictions name slots by their slot_uri (skos:note,
  sh:conformsTo, ositrace:identifier), the IRIs the ontologies declare.

Data-contract changes

envited-x v4

• Every artifact and referenced link uses the ENVITED-X access roles and categories. A referenced artifact stays optional.
• The simulation data / documentation / metadata / media links are required of a simulation asset's manifest (envited-x:SimulationManifest) only, no longer of software, service and code assets. An external simulation manifest declares the type to be checked; the fixtures do.
• A license link no longer needs an IRI. The license is gx:license on the resource description — a list, as in Gaia-X.
• hasFormat is optional on a domain specification, and description on the software, service and code bases, as their shapes, the leaf domains and Gaia-X state. leakage-test, service and survey define no format.
• hasSoftwareResource, hasServiceOffering and hasCodeArtifact stay sub-properties of hasResourceDescription, whose range now covers all four kinds of description.

manifest v6

• ArtifactLink, LicenseLink, ManifestLink: the JSON-LD artifact condition applies to the artifact properties, so the self-reference (hasManifestReference) is exempt by structure.
• The artifact properties have the range manifest:Link plus their conditions, so membership in ArtifactLink no longer follows from the range in OWL.
• Manifest is defined by subClassOf; the equivalentClass axioms (one with minCardinality 0) made every individual a Manifest.
• manifest:isMiscellaneous is a MiscellaneousCategory (it was typed with an undefined class).
• An empty CID is rejected; the pattern no longer needs a lookahead, which SHACL's regex dialect lacks.
• width, height and depth are unbounded: the metamodel has no exclusive bound (GAPS G8).

Nested values (all four) — without RDFS inference, an untyped nested value fails its sh:class check; OMB's default validation infers the type from rdfs:range and accepts it.

All four headers — owl:imports of the ontology-header profile and rdf:type ontology-header:OntologyHeader; owl:priorVersion, prov:wasDerivedFrom (releases/tag/v0.6.0), dcterms:references and dcterms:publisher are IRIs; no owl:versionIRI. dcterms:identifier (DCMI range: literal), dcterms:license and dcterms:conformsTo are literals. Contributors are GitHub account IRIs (GAPS G16).

🏷️ Versioning & Compatibility

• Major (Breaking change that makes existing Self-Descriptions invalid)

manifest v5→v6, envited-x v3→v4, georeference v5→v6, ositrace v6→v7; ontology-header v1 is new. A self-description against the old versions keeps validating against them. The importing domains are repointed without bumps of their own, as when hdmap moved to v6.

🏗️ Type of change

• New type (non-breaking change which adds a type)
• Breaking change (Change that would cause existing Self Descriptions not to be accepted anymore by a Federated Catalogue)

🧪 How Has This Been Tested?

just generate                        # all 18 generated artifacts reproduce byte for byte (against the pinned Gaia-X sources)
just validate                        # all checks pass: valid data of 22 domains conforms; 52 negative fixtures behave as recorded
pytest tests/                        # 683 passed, 1 skipped, 1 xfailed
mkdocs build --strict                # after properties_updater and class_page_generator, as cd-docs runs it: no warnings

• Every valid fixture of every domain conforms, including the 12 repointed importing domains and the new ontology-header domain.
• Every negative snapshot still fails on its root violation (the missing hasContent, the wrong CID, MIME type, file size or name, the bad lat/lon and bounding box, the missing topic, the manifest-vocabulary role or category, the literal publisher, the missing release). Without --inlined-as-node's sh:node, a holding node reports no "does not conform to shape" violation of its own; the ENVITED-X vocabulary violations report sh:in; manifest fail_04 reports the sh:or instead of the rule's sentence (GAPS G21).
• The negative fixtures of manifest, envited-x, ositrace, hdmap and openlabel-v2 match their snapshots under different PYTHONHASHSEED values.
• tests/integration/test_linkml_models.py checks every header against the profile (imports, type, no redeclared metaclass, conformance), the header terms (IRIs on declared annotation properties, no owl:versionIRI), the JSON-LD artifact condition, license categories, MCAP channels, geodetic alternatives and the documented dimension gap with real SHACL validation.
• The fork features: their PRs' CI and the pipeline branch's suite.

Docs and registry

docs/registry.json, docs/ontologies/catalog.md and docs/ontologies/properties.md are committed with the change, as for every ontology change: the documentation site is built from them, and with the previous registry mkdocs build --strict stops on 59 links to the replaced versions. The LinkML page (formerly under Future) states which domains are generated and how they are built; the contribute guide sends a change to a generated domain to its schema; the Gaia-X pages name envited-x v4.

Next steps

Before v0.6.0 — these change manifest v6 / envited-x v4, which are not released yet:

• G8 — width, height and depth greater than 0 again: minimum_value: 0 plus a class none_of that excludes maximum_value: 0. A probe accepts 1.5 and 1e-9 and rejects 0, -1 and NaN in SHACL; the OWL stays weaker (G19).
• R5 — decide whether manifest and envited-x keep separate isPublic, isLicense, isManifest and isMiscellaneous individuals.
• D5 — decide which URI-valued properties are IRI nodes and which are xsd:anyURI literals.

One fork round and one re-pin, for generator bugs visible in these artifacts:

• G18 — the ontology-header context maps nodeidentifier to @id.
• G17 — gen-owl leaves foreign predicates undescribed.
• G21 — anonymous shapes get an sh:message.
• G19 — OWL anonymous class expressions keep their presence conditions.

In this repository, not necessarily in this PR:

• R2 — --per-resource validation loads fixtures and the ontology's named individuals; needed before the comparison harness measures the next domain.
• cd-release — its push of the registry to main is declined by the branch rules.
• Mark this PR ready for review and request reviewers.

At release

• The release must be v0.6.0: the generated headers name releases/tag/v0.6.0 as prov:wasDerivedFrom.
• The v0.6.0 version-bump PR regenerates the registry for v0.6.0 before the tag (docs/ecosystem/release-process.md); until then the registry names v0.5.1.
• After the docs deploy, https://w3id.org/ascs-ev/envited-x/ontology-header/v1 resolves through the generic w3id rules.

Known gaps after merge

Recorded in linkml/GAPS.md; none of them blocks the switch-over:

• D5 — which URI-valued properties are resources (IRI nodes) and which are data (xsd:anyURI literals).
• D11 — the renamed envited-x classes lose their bare context terms.
• G8 — no exclusive numeric bounds (manifest dimensions).
• G16 — contributors by IRI only; native license / conforms_to are literals although DCMI recommends URIs (gen-owl emits rdfs:isDefinedBy and dcterms:license as string literals instead of URIs linkml/linkml#3546); non-linguistic annotations carry @en.
• G17 — gen-owl describes a reused foreign predicate (label, comment, skos:inScheme).
• G18 — the ontology-header context coerces its terms to shex:nonLiteral instead of @id.
• G19 — OWL omits presence conditions inside anonymous class expressions.
• G20 — OpenLABEL minValue ≤ maxValue is not enforced.
• G21 — anonymous shapes carry no sh:message.
• G22 — no version IRI in the metamodel: no permanent ontology IRI with a version IRI per release.
• R2 — --per-resource validation merges neither fixtures nor ontology individuals.
• R5 — manifest and envited-x each define isPublic, isLicense, isManifest, isMiscellaneous.
• R6 — the generated ontologies state no owl:versionIRI; the registry, docs build and class pages use the ontology IRI, and registry_updater logs a warning for each.

✅ Checklist

• My code follows the modelling guidelines of this project
• I have performed a self-review of my own changes
• I have updated the version number in the ontology file (if applicable)
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes