[jh-RLI]

Summary of the discussion

Covers a municipal heat plan, its target scenario and one final energy consumption value: deep enough to test the LinkML-authors-SHACL decision against real MHPO and OEO terms, small enough to finish.

The decision holds. gen-shacl puts class_uri straight into sh:targetClass, so the generated shapes constrain MHPO and OEO IRIs directly rather than a LinkML-shaped shadow of them. MHPO's role reification did not bite here, because OEO's quantity value already reifies the same way LinkML would — but that is a deferral, not a clearance: the areas-and-roles subtree is three quarters of MHPO and this slice does not enter it.

Enums are the pattern to reuse. OEO models units, carriers and sectors as classes while a statistic is about the carrier type, so pointing at a class IRI is unavoidable punning. An enum whose every meaning: is an OEO IRI generates sh:in, a real value-set constraint instead of sh:nodeKind sh:IRI. oekg/shapes/oekg_shapes.ttl already does this by hand for ~120 IRIs.

Four things that will shape later tickets:

• The IRI policy cannot be generated. SHACL constrains a node's own IRI with sh:pattern at node level and LinkML cannot emit that; a type-level pattern is dropped (silently with uri: set, otherwise logged as ERROR and still exit 0), and on an identifier slot it lands on an invented property that no triple uses, so it can never fire. Hence the hand-written companion file — the first artifact here that is not generated.
• Enum membership cannot be checked by anything in this repository. OEO's energy carrier has four asserted subclasses, all abstract; natural gas qualifies only by inference. Confirming membership needs a reasoner and the toolchain deliberately has none.
• MH-02 §6 and MHPO collide over the municipality. The policy reuses municipality/AGS_06611000 for the commissioning body, but the only class for it is municipality area — a spatial region cannot commission a plan, and the two IRI shapes cannot both be satisfied. Needs a human.
• pyshacl accepts repeated -s and silently uses only the last one. The first validation run reported Conforms: True against both shape files while consulting neither, and still passed with a required label deleted. validate.py merges the shapes into one graph for that reason.

Also recorded: closed shapes reject provenance triples; German labels must be plain literals because @de is rdf:langString and fails sh:datatype xsd:string; and Turtle cannot abbreviate these IRIs at all, since a prefixed name may not contain an unescaped slash.

Five term requests are left as commented blocks at the position they would occupy, so the holes are visible in the artifact. No term is minted here and gen-owl is never run.

Verified with linkml 1.11.1, pyshacl 0.40.1, rdflib 7.6.0 against MHPO pin 34776b6 and OEO 2.13.0. validate.py asserts all 14 negative-control violations are caught and exits 1 when the IRI shapes are removed.

Type of change (CHANGELOG.md)

Added

• mhpkg/schema/: first cut of the MHPKG data shape — a LinkML schema for one slice, the SHACL
  generated from it, a hand-written companion enforcing the IRI policy, and a passing plus a
  deliberately failing example (#54)

Updated

• Nothing. This PR only adds; pyproject.toml and uv.lock are deliberately untouched, since
  MH-04's schema and graph dependency groups already carried everything needed.

Removed

• Nothing.

Workflow checklist

Automation

Closes #55

PR-Assignee

• 🐙 Follow the workflow in CONTRIBUTING.md
• 📝 Update the CHANGELOG.md
• 📙 Update the documentation

Reviewer

• 🐙 Follow the Reviewer Guidelines
• 🐙 Provided feedback and show sufficient appreciation for the work done