Skip to content

DAnO: where ADIRO overlaps it, and why ADIRO mints its own terms

A comparison of the ADIRO suite with the Drawing Analysis Ontology (DAnO), and the decision it settles: ADIRO does not import or extract from DAnO, and ships no DAnO artifact at all — no module, no import, no annotation-level crosswalk. ADIRO mints its own terms; the correspondence with DAnO is recorded here, in documentation, only.

Where the decision lives

Decision: #77, implemented in #76. The rule it applies is External ontology imports. The term survey that preceded it is Discussion #70. Revised 2026-09-25: an earlier revision of #76 shipped an optional aec_dano_alignment crosswalk module; it was withdrawn so ADIRO ships no DAnO artifact and the comparison is documentation-only.

Summary

DAnO is a well-shaped vocabulary that overlaps ADIRO on one module out of six. Outside that module the two do different jobs; inside it they are genuine alternatives, and there DAnO is currently the richer of the two.

ADIRO treats DAnO as evidence rather than as a dependency. It corroborated two ADIRO design decisions, exposed one real gap, and none of that required importing a single axiom. Where the two overlap, ADIRO mints its own terms and records the correspondence with DAnO here, in documentation — ADIRO publishes no crosswalk artifact.

This page is deliberately explicit about where DAnO is ahead, because the module where that is true (adiro_common_symbols) is the one due for work next, under UC-03 and UC-07.

1. What DAnO is

Drawing Analysis Ontology - https://w3id.org/dano, prefix dano:, CC BY 4.0, from RUB Informatik im Bauwesen. Its own scope statement: "concepts of technical drawings from the perspective of computer vision-based drawing analysis". 16 classes, 13 object properties, 7 datatype properties, no owl:imports, and no owl:disjointWith or owl:equivalentClass axioms.

DrawingElement -+- DisplayElement      "an element representing a real-world component"
                +- DescriptionElement  "an abstract element describing properties of display
                                        elements" (dimension lines, text)

DrawingElement --hasConfidence--> xsd:decimal
DrawingElement --hasMeta--> DrawingElementMeta -+- inferredAt   (xsd:date)
                                                +- inferredBy   (xsd:string)
                                                +- inferredFrom (xsd:string)
                                                +- inferredWith (xsd:string)

2. The overlap map

ADIRO is a suite, and the relationship with DAnO differs in each module. Stating this per module rather than in the aggregate is the only way to get it right.

ADIRO module DAnO counterpart Relationship
adiro_provenance DrawingElementMeta, inferredAt/By/From/With, hasConfidence Overlapping. ADIRO supersedes - see §3
adiro_drawing_metadata - sheets, layouts, revisions, projects, parties, title block TextField, TextElement only Largely complementary. DAnO models no document structure, no parties, no revisions
adiro_common_symbols - Dimension, Grid, ReferenceSymbol Dimension, DimensionChain, DimensionLine, AxisLine, Terminator, SectionSymbol, Symbol, EquipmentSymbol Directly competing, same layer. DAnO is currently richer - see §4
adiro_domain_common, adiro_facade_domain none No overlap. DAnO has no discipline or domain modelling

An earlier framing described the two as occupying different layers - DAnO for what was seen, ADIRO for what it means. That holds for the title-block vocabulary it was written about, and it does not generalise to the suite: adiro_common_symbols sits squarely in DAnO's layer and models the same objects.

The documented correspondence (§10) is the best measurement of the overlap: six provenance terms, and nothing else. Nothing corresponds to sheets, layouts, revisions or parties, because DAnO has no counterpart. Nothing corresponds to symbols, because that question is open rather than settled. Nothing corresponds to depiction either, because the linking property ORSD CQ 5.2 asks for is not minted - see §10.

3. Where ADIRO goes further: provenance

This is the substantive modelling difference, and it is requirement-driven rather than a preference for a heavier model.

DAnO attaches provenance to the element: one DrawingElement, one DrawingElementMeta, one confidence. ADIRO reifies the assertion: a FieldAssertion is a claim that some source states some value for some field kind, carrying its own assertedBy, hasInferenceMeta and hasConfidence.

That choice is what makes three ORSD competency questions answerable:

Competency question Why assertion-level is needed
CQ 4.4 - which extraction-service results have been aggregated, and from which service did each originate? Two services can assert different values for the same field. As separate FieldAssertions each keeps its own provenance and confidence; as one element with one meta object, the second write overwrites the first
CQ 2.4 - which detected errors are auto-resolvable and which require human intervention? Needs a validation status and an explicit conflict relation, both per-claim. DAnO has no human-in-the-loop concept
CQ 3.2 - which predicted individuals match their ground-truth counterparts, and which diverge? Divergence is a relation between two claims about the same thing, which presupposes that claims are individuals

In fairness, CQ 3.1 (what label, confidence score and geometry are recorded for a given model-prediction individual?) is satisfied by DAnO's model as it stands. The gap is specific, not general.

DAnO's placement of hasConfidence on the element while the four inferred* properties sit on the meta object is worth reading as deliberate: inference metadata is usually shared across every element from one extraction run, whereas confidence is per-result. ADIRO reaches the same conclusion by a different route.

Secondary differences on the same terms, all pointing the same way: DAnO carries actor, tool and source as xsd:string, so two spellings of one model are unrelated strings, while ADIRO points at entities that can be described, versioned and queried. DAnO's inferredAt is xsd:date, so two extraction runs on one day are indistinguishable; ADIRO's is xsd:dateTime. ADIRO's provenance is aligned to PROV-O throughout, so PROV-aware tooling understands it without bespoke mapping.

4. Where DAnO goes further: drawing-mark decomposition

adiro_common_symbols currently declares three classes - Dimension, Grid, ReferenceSymbol. DAnO declares eight or so in the same territory, and decomposes a dimension into its parts: dano:Dimension is a composite of exactly one DimensionLine, two Terminators and one TextElement, with DimensionChain above it and AxisLine alongside.

On that ground DAnO is the more developed vocabulary, and any claim that ADIRO is simply more expressive would be false there. ADIRO's symbol coverage is thin because the use cases that need it have not been built yet, not because a decomposition was considered and rejected.

This matters now rather than later: UC-03 (Reference Symbol Cross-Sheet Linking) is the highest-priority open use case and touches SectionSymbol / Terminator / refersTo / isReferredToBy, and UC-07 (Wall Orientation and Measurement) needs exactly the dimension and axis-line territory. See §8.

5. What actually distinguishes ADIRO

Not expressivity in the abstract. ADIRO is built for the machine-learning loop, and that shows up as vocabulary DAnO has no counterpart for:

  • isComputerVisionLabel, exampleImage, extractionHint and related annotations - they tell an annotation tool and an extraction pipeline what to label and how, carried by the ontology itself.
  • expectedRange, mapsToFieldProperty - the bridge from a reified assertion to a validated, promoted statement.
  • A field-kind scheme whose membership is the extraction profile, with per-language synonyms.
  • Document structure, parties, revisions, projects and discipline taxonomies, so an extracted value can be placed in a real engineering context.

And one characteristic that is not vocabulary at all but shapes every modelling choice: ADIRO individuals are detections, not idealised drawing constructs. ADIRO targets historical drawings as well as modern ones, and a scanned historical sheet is routinely cut at the edge, torn, faded or partially legible. What ADIRO records is what a detector actually found, with confidence and with gaps. DAnO describes drawings as they are meant to be. That difference has teeth - see §8.

DAnO describes what is on a drawing. ADIRO describes what a drawing means, and how a model is trained, evaluated and held accountable for extracting it. That is a difference in purpose, not a score.

6. Maintenance and release discipline

ADIRO commits to per-module SemVer, version-pinned IRIs, changelogs, and a blocking DL reasoning gate on every pull request. DAnO's repository was created 2025-03-11 and last pushed 2025-07-15, with no tags or releases - though it does have a permanent w3id.org IRI, which is more than many peers offer.

The earlier title-block vocabulary review called it "borrowable, not dependable". That remains the right summary, and it is an argument about what to take a hard dependency on, not about the quality of the modelling.

7. The decision: mint our own terms, ship no DAnO artifact

ADIRO mints its own terms and records their correspondence to DAnO in this document only. There is no owl:imports, no extraction, no annotation-level crosswalk in any TTL, and no ADIRO module mentions DAnO at all. Even so it is worth setting down why a logical alignment (rdfs:subPropertyOf) was unavailable, and then why even a machine-readable annotation crosswalk was dropped.

The cheap-looking alternative is rdfs:subPropertyOf, so that DAnO-aware consumers understand us. That move is unavailable for three separate reasons, worth separating because only the first fails our own CI.

Type mismatch (fails locally). aprov:inferredBy, inferredWith and inferredFrom are object properties; DAnO's are datatype properties holding strings. OWL 2 DL has no sub-property axiom crossing that boundary, so the axiom is ill-typed and non-DL, and the blocking reasoning gate rejects it. Declaring the stub dishonestly as an object property would move the failure downstream onto anyone merging real DAnO.

Range conflict (fails on merge). aprov:inferredAt has range xsd:dateTime; dano:inferredAt has range xsd:date. The two datatypes are disjoint, so an inherited range makes every recorded timestamp inconsistent.

Domain leakage (silent). dano:hasConfidence carries rdfs:domain dano:DrawingElement. In OWL a domain is an inference rule, not a constraint: from x dano:hasConfidence 0.87 a reasoner concludes x is a dano:DrawingElement. Under a sub-property axiom, every FieldAssertion ADIRO produces would be entailed to be a drawing mark - filing a claim about a mark under the class of marks. Nothing errors. The same applies to dano:depicts (rdfs:domain dano:DisplayElement) and, through owl:inverseOf, to dano:isDepictedBy.

Why this is easy to miss

A local stub carries none of the source's axioms, so the second and third failures pass the ADIRO gate silently. They fire the first time someone loads ADIRO alongside DAnO itself - which is exactly the audience an external IRI exists to serve. An alignment that works only for consumers who ignore it is not an alignment.

Any machine-readable link would therefore have to be an annotation, not a logical axiom: a reasoner must conclude nothing from it, and no domain, range or type may be inherited. ADIRO went one step further and dropped the machine-readable crosswalk altogether. The correspondence is documented here (§10) rather than shipped, so there is nothing in any ADIRO file for a reasoner to misread and nothing to keep in sync with DAnO.

Why the correspondence is documentation, not a shipped crosswalk

An earlier revision of #76 shipped a machine-readable crosswalk — first with skos:closeMatch, then with a minted ADIRO annotation property in an optional aec_dano_alignment module. Both were withdrawn on 2026-09-25: a crosswalk embedded in the ontology was judged more likely to confuse consumers than to help them, and it carried maintenance cost for no current consumer.

skos:closeMatch was in any case unusable for the job. It inherits rdfs:domain skos:Concept through skos:semanticRelation, so a consumer loading ADIRO and SKOS would derive that every mapped term — ADIRO's own properties and classes, and DAnO's — is a skos:Concept. Nothing errors (OWL 2 DL permits punning), but the statements are false: these are properties and classes, not concepts. That is the same silent domain leakage this page rejects dano:hasConfidence for (§7), landing on exactly the consumers a crosswalk exists to serve — anyone deliberately loading a crosswalk is likely to have SKOS in their graph, and adiro_drawing_metadata is itself a SKOS vocabulary.

The general lesson outlives the decision: a property is annotation-level because it is declared one, not because it is used as one. Every mapping vocabulary worth reaching for — SKOS, and the OBO oboInOwl:has*Xref family — carries axioms of its own, and those axioms arrive with the vocabulary whether or not the person writing the mapping expected them. A minted local owl:AnnotationProperty avoids that (no domain to inherit, no logical force by construction); declaring skos:closeMatch itself locally as an annotation property does not, since it contradicts SKOS's own owl:ObjectProperty declaration and produces illegal punning on merge. With no crosswalk shipped at all, none of this reaches a consumer — the correspondence lives in §10 for a human or a crosswalk author to read.

Option 1 is really two moves

The published rule folds both under "local stub + alignment", but they differ in which IRI is the subject. Move A - reuse the external IRI as a stub (prov:Agent in adiro_provenance): the external term stays the subject and ADIRO's axioms point at it. Move B - mint an ADIRO term and relate it (everything on this page): the ADIRO IRI is the subject and the external term appears only as an annotation value, or — as for DAnO — only as a documented correspondence. Move B is the right one whenever the external term's own axioms are incompatible with ADIRO's.

8. Open: adiro_common_symbols, and what this strategy does and does not constrain

adiro_common_symbols is the one module where ADIRO and DAnO compete, and it is next in line for work. Nothing on this page decides it; the decision is tracked in #79. What follows is the procedure it inherits.

When may a core module reference an external vocabulary directly? The test is not "external or not" - adiro_provenance references PROV-O in the core and should. The test is:

Condition PROV-O DAnO
Source is stable, versioned and maintained W3C Recommendation since 2013 No releases, quiet since 2025-07
Alignment is logically sound inferredBy genuinely is a prov:wasAttributedTo Type, range and domain clashes throughout
The entailments it carries are wanted Yes - PROV-awareness is the point No - the mappings are for humans and crosswalk tools
Verdict Core Documentation-only comparison

So the rule the suite now follows: an external term may be referenced from a core module only when the source is stable and the alignment is both sound and wanted. Otherwise ADIRO mints its own terms and does not reference the external vocabulary from a core module at all; a machine-readable crosswalk, if ever wanted, lives in an optional layer that carries no logical force and consumers opt in. For DAnO even that was judged unnecessary — the correspondence is documentation-only.

What that permits for adiro_common_symbols:

  • Minting ADIRO terms that parallel DAnO's decomposition - DimensionLine, Terminator, AxisLine and so on - and documenting their correspondence to DAnO here (a machine-readable crosswalk remains an option but is not currently shipped). Fully compatible with this strategy, and the likely answer, because ADIRO symbol classes are also labelling targets: they carry the suite's own labelling annotations, which downstream annotation tooling reads. Annotating an external IRI with ADIRO's own annotation properties would couple the core to DAnO through the back door.
  • A pinned SLME STAR extract, if a future use case genuinely needs to reason with DAnO's axioms. The published rule's options 2 and 3 remain open. Such an extract is a different artefact from any crosswalk - it holds real external axioms, must be version-pinned and justified, and whatever needs it imports it explicitly. Adding one would not contradict this page; it would need its own decision under the same test above.

A further argument for minting, specific to ADIRO's data. dano:Dimension is not just a label, it is a composite definition with exact qualified cardinalities:

dano:Dimension ⊑ dano:Composite
  ⊓ =1 contains.DimensionLine
  ⊓ =2 contains.Terminator
  ⊓ =1 contains.TextElement

On a clipped, torn or partially legible historical sheet - ADIRO's routine input - a detected dimension may be missing a terminator, its text, or the line itself. Under OWL's open-world assumption those axioms do not reject such a dimension. They infer the missing parts into existence. Ask a reasoner afterwards which dimensions are incomplete and it answers "none", having already supplied what the detector never saw.

That directly undermines a competency question ADIRO already holds: CQ 2.1 asks whether an extracted graph lacks any mandatory functional sub-components required to form a semantically complete assembly. Adopting an exact-cardinality axiom makes missing sub-components unfindable by reasoning. Completeness is a validation question - a SHACL shape over a finished extraction, under closed-world semantics - not an entailment over detections. Tracked in #84, with a worked note in UC-07 §7.5.

The axioms would not fail the reasoning gate. They would quietly give wrong answers, which is worse, and it is the same failure mode as the domain leakage in §7: an adopted axiom that entails something ADIRO does not mean.

What it rules out: declaring DAnO IRIs inside a core module and building axioms on them (move A for DAnO), because DAnO fails the first two conditions of the test.

What is not yet known: whether UC-03 or UC-07 need DAnO's decomposition at all. UC-03 already ships csymbol:ReferenceSymbol and its own linking properties; UC-07 has no ontology work yet. The question to answer when they are picked up is whether DAnO's published decomposition is better than what ADIRO would design unaided - and if it is, to adopt the shape and mint the terms, which is what ADIRO has already done twice with DAnO (see §9).

With no crosswalk shipped there is no dangling-annotation hazard and nothing to stale-check: a documented correspondence is maintained with the page. The staleness-check idea (#81) is therefore moot unless a machine-readable crosswalk is added later.

9. What DAnO contributed anyway

None of this required importing anything, and all of it changed ADIRO:

  • It corroborated the reification choice. ADIRO was holding open RDF-star versus a meta-object for per-value provenance. DAnO had already chosen the meta-object. A peer group meeting the same problem the same way is real evidence, and aprov:InferenceMeta follows it.
  • It exposed a genuine gap. ADIRO proposed no extraction timestamp anywhere. dano:inferredAt made the omission obvious; aprov:inferredAt exists because of it.
  • It corroborated two over-modelling suspicions. DAnO carries source and model as plain strings, which supported the independent finding that SourceFile and MLModel as classes were heavier than the requirements justified.

This is the pattern worth keeping: adopt the shape, mint the terms, map the names.

10. Per-term verdicts

DAnO term Verdict Reason
dano:inferredBy Superseded by aprov:inferredBy (documented correspondence) Object/datatype type mismatch; ADIRO's resolves an agent
dano:inferredWith Superseded by aprov:inferredWith (documented correspondence) As above
dano:inferredFrom Superseded by aprov:inferredFrom (documented correspondence) As above; ADIRO's is PROV-aligned
dano:inferredAt Superseded by aprov:inferredAt (documented correspondence) xsd:date disjoint from ADIRO's xsd:dateTime
dano:hasConfidence Superseded by aprov:hasConfidence (documented correspondence) Domain leakage; ADIRO's is per-assertion
dano:DrawingElementMeta Superseded by aprov:InferenceMeta (documented correspondence) Same pattern, PROV-aligned
dano:depicts Not reused, and no ADIRO counterpart minted yet Domain dano:DisplayElement is wrong for a class that also covers dimensions and grids. The ADIRO term is unminted because its consumers specify their own - #80
dano:isDepictedBy Not reused Same as dano:depicts - #80
dano:Dimension, DimensionChain, DimensionLine, AxisLine, Terminator, SectionSymbol Open, but reuse of the IRIs is disfavoured adiro_common_symbols / UC-03 / UC-07 - see §8, #79. dano:Dimension's exact cardinalities would infer undetected parts into existence on clipped drawings - #84
dano:DisplayElement vs DescriptionElement Open The split may be worth minting as ADIRO's own; ADIRO's DrawingElement currently spans both - #83
dano:hasGeometry, defaultGeometry Deferred Gated on the GeoSPARQL decision, #36
dano:hasIfcRepresentation Not assessed Possible cheap alternative to IFC alignment axioms

ORSD CQ 5.2 and its open issue OI-1 remain open. An earlier revision of this work minted a generic metadata:depicts here; it was withdrawn, because the use cases that need depiction have each already specified their own - UC-06's depictsMaterial and UC-07's depictsElement, both with domain Drawing rather than the metadata:DrawingElement that CQ 5.2 frames. Minting a parent ahead of them would have pre-empted a decision belonging to those use cases, in a PR scoped to UC-01 and UC-03. Tracked in #80.

11. Precedent: neither ontology imports the other's source

DAnO has no owl:imports. It declares the GeoSPARQL terms it needs locally, with rdf:type, rdfs:label, rdfs:comment, rdfs:isDefinedBy and a version-pinned owl:versionIRI annotation:

geo:hasGeometry
  rdf:type owl:ObjectProperty ;
  rdfs:comment "A spatial representation for a given feature." ;
  rdfs:isDefinedBy <http://www.opengis.net/ont/geosparql> ;
  rdfs:label "hasGeometry"@en ;
.

That is ADIRO's option 1, move A, arrived at independently by another group - external corroboration of the import rule, and a reason to read ADIRO's treatment of DAnO as the same courtesy rather than a criticism.

ADIRO does not ship a crosswalk file at all; the correspondence is documentation-only (§10). Had a machine-readable crosswalk been wanted, the lifecycle argument would have placed it in a separate optional file — the shape GeoSPARQL itself uses (core plus alignments.ttl) — rather than a core module: an unreleased third-party vocabulary should not force a version bump on a core module that downstream consumers pin, and crosswalks to DiCon, ifcOWL, BOT and GeoSPARQL would otherwise pile up inside two core files. For DAnO even a separate file was judged unnecessary. It is worth being precise that annotation-only mappings would not have endangered an SLME extraction if left inline — locality is defined over logical axioms and skos:closeMatch is not one — so this is not the merged-alignment-graph hazard the import rule warns about.

It also retires an argument made earlier in ADIRO's own review, that importing DAnO would "drag its 16 drawing-element classes" into the suite. True as far as it goes, but DAnO imports nothing, so bloat was never the operative risk. The operative risk is type and domain incompatibility, which is sharper and applies even to the lightest alignment.

12. Method and limits

Verified against primary sources (2026-09-21): the raw dano.ttl fetched from https://w3id.org/dano (HTTP 200, resolving to rub-informatik-im-bauwesen.github.io/dano/dano.ttl). Every domain, range and property type quoted here is read from that file, not from DAnO's generated specification. The absence of owl:imports, owl:disjointWith and owl:equivalentClass was confirmed by direct search. ADIRO class and annotation-property counts are read from src/*.ttl on the branch implementing this decision.

Supersedes: the earlier title-block vocabulary review recommended aligning ADIRO's provenance terms to dano: rather than minting parallel ones. That review read DAnO from its generated specification and explicitly flagged that domains and ranges should be confirmed against the source before any alignment axiom was written. They now have been, and the recommendation does not survive the check.

Corrects: an earlier framing of the two ontologies as occupying different layers. That holds for the title-block vocabulary it was written about; it does not hold for adiro_common_symbols.

Not claimed: that DAnO's modelling is wrong for DAnO's purpose. Every objection here concerns what ADIRO can safely depend on, given a blocking DL gate and a different set of competency questions.

Not disjoint: DAnO declares no owl:disjointWith axioms, so the domain-leakage objections describe wrong entailments rather than inconsistencies that would fail a reasoner today. That would change if DAnO declared DisplayElement and DescriptionElement disjoint, an obvious future addition.