Known issues

A completeness claim that conceals deferred work or defects is a provenance lie. This is the dated register of every known caveat in the schema-derived ontology, each cross-referenced to the ODR that governs it. The important distinction throughout: none of these is a defect in OPDA's own materialised graph. They are consumer-side traps, by-design outcomes, a report-side miscount, and one item already resolved by a rename.

The register (dated 2026-06-14)

Why the EPC cross-trip is not a model defect

This is the issue most likely to be misread, so it is worth stating plainly. If you load the OPDA ontology into a generic reasoner running full RDFS or OWL — one that applies rdfs:domain entailment — and you assert an opda:EPCCertificate that carries currentEnergyRating, that reasoner will infer the certificate is an opda:Property, because currentEnergyRating is declared with rdfs:domain opda:Property. That inference is wrong: an EPC certificate is a document about a property, not the property itself.

OPDA does not make that inference. The Safe-Group closure adopted in ODR-0025 (§R7) deliberately excludes domain and range rules. The seven frozen rules cover subclass, subproperty, inverse, symmetric and transitive entailment — the relations whose closure the author actually wants — and nothing that would manufacture a spurious class membership from a property's declared domain. The round-trip exemplars are validated against this closure, and they pass. So the cross-trip is a consumer-side trap (and a modelling smell worth knowing about), not a defect in OPDA's graph, and it does not block draft → accepted. The provenance for this ruling is on the Decision provenance page.

Why thin profiles are by design

"A form IS its SHACL overlay" (ODR-0010). But a PDTF form is mostly questions whose answers are A/B/C/D/E/F treatments — carried by dct:subject and a JSON-pointer dct:source — not Category-G substantive attributes that bind as per-leaf sh:path shapes. A profile that binds only its Category-G leaves and GAP-registers the rest is therefore the ratified outcome, not an omission (session-034). A "thin" profile that binds zero per-leaf shapes (oc1, llc1, tf, and others) is correct: the leaves it would otherwise bind are register extracts or free-text, and the GAP register records exactly which leaves are unbound and why. The live profile counts and the bound/thin split are on the overview and the bake-off's custom-layer rendering.

Why the ruleset is not an OWL 2 RL reasoner

An earlier framing risked implying that config/opda-rdfs-plus.rules was an OWL 2 RL profile reasoner. It is not, and ODR-0029 R5 resolved this by renaming the artefact and giving it an honest header. The ruleset is a sound but RL-incomplete fragment: it implements seven of the OWL-RL rules and deliberately omits domain/range, equality (owl:sameAs), functional-property and equivalence reasoning. Soundness means everything it derives is valid; RL-incompleteness means it does not derive everything OWL 2 RL would — which is the point, because the omitted rules are exactly the ones that would produce the EPC cross-trip and spurious owl:sameAs collapses. The rule logic is now frozen.

Related: Decision provenance (the governing records), Consume the ontology (the advanced=True validation recipe), and the Tool bake-off (the custom layer that emits the GAP register).