Skip to content

Title-block vocabulary: review against the use cases and comparison with DAnO

Date: 2026-08-11 · Issue: RES-89 · Branch: res-89-aec-titleblock-tbox Reviews: the 73 proposed terms in Discussion #61 §11.1 Against: docs/specification/use-cases/ (UC-01, UC-03, UC-06, UC-07) and DANO

Why this document exists

The proposed title-block vocabulary was assessed for internal redundancy in Discussion #61 §11.1 (29 core, 19 useful, 20 marginal, 5 redundant). This document applies two external tests that assessment could not:

  1. Do the terms serve the use cases? The repo follows LOT — Use Case → Information Needs → Functional Requirements → Competency Questions → OWL Terms — and claims "forward traceability (every OWL term traces back to a use case need)" (docs/specification/use-cases/README.md §1).
  2. Has someone already built the drawing-analysis half of this? DANO is a published, CC BY 4.0 ontology for exactly computer-vision-based drawing analysis.

Both tests found things the internal review could not. The first finding below is the most consequential thing in this document and is not a redundancy question at all.


1. Headline findings

# Finding Consequence
1 UC-01 already implements title-block search, and it made the opposite architectural decision to RES-A-9 finding 2 — deliberately, and under review. The §1.1 reconciliation is not "reconcile with terms that drifted in". It is "choose between two reviewed architectures". Reframes pass 0 entirely
2 ~40 of the 73 terms have no competency question behind them, from any use case. Either the extraction work needs its own ORSD, or the vocabulary should be cut to what the use cases demand. Currently it violates the repo's own stated traceability rule
3 Discipline would duplicate dcommon:Discipline, a class hierarchy UC-01 already reuses — and a SKOS scheme would lose the sub-discipline reasoning UC-01 chose it for 🟢 Core → 🔴 Redundant
4 hasNorthPointOrientation duplicates UC-07's northArrowAngle, which is better specified 🟠 Marginal → 🔴 Redundant
5 scale duplicates dm:hasScale; the scale story is already two properties by design (string + numeric ratio) 🟢 Core → 🔴 Redundant (the field is core; the term is not)
6 DANO already models the extraction-provenance layer (§10) — including a timestamp ADIRO omits — and its DrawingElementMeta is a worked reification pattern Bears directly on the pass-5 blocker
7 DocumentType risks repeating a mistake UC-01 already corrected (its DrawingType → LayoutContentType deprecation) Keep, but state explicitly that it is a different axis

2. Part A — against the use cases

2.1 The architectural contradiction (finding 1)

UC-01 is "Titleblock-Based Drawing Search", priority XL, and marked "Done already" in the prioritisation matrix. Its ORSD contains this design note:

Design note — Titleblock as visual region vs titleblock data: The existing metadata:Titleblock class models the visual region on a sheet (for CV annotation). The information conventionally found inside it (drawingIdentifier, drawingTitle, …) is attached as datatype properties directly on DrawingSheet. The two layers coexist independently: a sheet has a drawing number whether or not its visual Titleblock region has been bounded in annotation.

That is the direct opposite of RES-A-9 finding 2 and Discussion #64 §7, which require "Domain is Titleblock, never Document".

This is not drift. UC-01 v0.3's changelog records the sheet-level placement as an alignment decision taken in review (@alelom's PR review, item G2: "Datatype properties attached directly to DrawingSheet — no intermediate entity"). It was proposed, reviewed, accepted and implemented. The ~12 dm: terms that Discussion

64 §1.1 treats as an accidental overlap are the UC-01 deliverable.

Both positions are defensible, for different jobs:

UC-01's sheet-level model RES-A-9's titleblock-level model
Purpose Search and filter sheets by metadata Extract assertions from a detected region
Multi-sheet conflict Not a concern — one value per sheet The central concern — two sheets may disagree
Provenance Not required Required per value
Query shape ?sheet metadata:drawingIdentifier "ST-101" ?tb tb:identificationNumber ?v . ?tb tb:assertsMetadataFor ?doc

UC-01's CQs never ask "which sheet asserts a conflicting value for this document", so it has no reason to carry the indirection. The extraction pipeline's entire justification for the split is that question.

What this means for pass 0. Discussion #64 §1.1's three options need restating:

  • Option A/B as written (reuse classes, parallel assertion layer) is still viable, but it is not "avoiding an accident" — it is deliberately running two vocabularies for one concept, one of which is a shipped, CQ-validated deliverable. That is a much higher bar to clear, and it needs UC-01's owner in the room.
  • Option C (migrate to titleblock-level) would overturn a reviewed decision and break UC-01's CQ SPARQL. Already rejected for annotation-tooling reasons; this is a second, independent reason.
  • A fourth option not currently on the table: keep sheet-level as the validated layer (UC-01, unchanged) and introduce the titleblock-level layer only for unvalidated extraction output, with promotion sheet-ward on validation. That is close to option B but framed as a pipeline stage rather than a parallel vocabulary — which is arguably what RES-A-9 finding 2 actually described ("promote to the Document after validation") and would make the duplication temporary-by-design rather than permanent.

Recommendation: put this in front of whoever owns UC-01 before pass 1. It is the single highest-value conversation in the whole plan.

2.2 What the use cases actually demand

Consolidating the four ORSD traceability matrices, the terms with a CQ behind them are:

Use case Status Terms it needs (all already in adiro_drawing_metadata / adiro_domain_common / adiro_common_symbols)
UC-01 Titleblock search ✅ implemented DrawingSheet, Layout, LayoutContentType, DrawingRevision, Project, Person, StatusCode, DrawingPackage, dcommon:Discipline; drawingIdentifier, drawingTitle, hasScale, sheetSize, revisionCode, issueDate, personName, projectName, projectNumber, packageName, statusLabel; contains, hasProperty, hasDiscipline, hasRevision, isRevisionOf, belongsToProject, belongsToPackage, isAuthoredBy, isCheckedBy, isApprovedBy, hasStatusCode
UC-03 Cross-sheet linking ✅ implemented csymbol:ReferenceSymbol, hasReferenceSymbol, referencesLayout, layoutIdentifier
UC-06 Content aggregation ⚠️ draft Material, HazardClassification, depictsMaterial, isDepictedOn, hasHazardClassification, materialName, materialSymbol, hazardLabel — none of them title-block terms
UC-07 Wall orientation ⚠️ draft BuildingElement, Wall, FacingDirection, northArrowAngle, scaleRatio, wall dimension properties — one overlap: the north arrow

Every title-block field a current use case needs is already implemented. UC-01 covers identity, project, revision, the three person roles, discipline, content type, status and package. Nothing in the proposed 73-term vocabulary is required to answer any existing CQ.

2.3 New redundancies the internal review missed

Term Rated in #61 Revised Why
Discipline (SKOS scheme) 🟢 Core 🔴 Redundant dcommon:Discipline is a full class hierarchy (Structural, MEP > Mechanical / Electrical > Lighting / Plumbing, Architectural > Facade / FireLifeSafety, Masterplan, + Civil). UC-01 FR 5 reuses it and explicitly dropped a disciplineCode string (change G3) because rdf:type checks against the hierarchy give automatic sub-discipline rollup — querying dcommon:MEP matches Mechanical, Electrical, Plumbing, Lighting via rdfs:subClassOf. A flat SKOS scheme loses exactly that. Reuse dcommon:Discipline + dcommon:hasDiscipline
hasNorthPointOrientation 🟠 Marginal 🔴 Redundant UC-07 specifies northArrowAngle (xsd:decimal, degrees clockwise from page-up) on the drawing, with a stated datum convention, and explicitly rejects modelling the arrow as an entity. ADIRO's version is the same measurement with a vaguer name, a different datatype and no datum. It is also a drawing-level datum, not a title-block field
scale 🟢 Core 🔴 Redundant as a term dm:hasScale (string) exists from UC-01 FR 1, and UC-07 adds scaleRatio (decimal) for arithmetic. The scale story is already two properties by design. A third string property is duplication. The field remains core to extraction — it just binds to dm:hasScale
dimensionUnits 🟡 Useful ~~🟡 Useful — confirmed genuinely new~~ → 🔴 Withdrawn 2026-09-14, on evidence Nothing in any use case or module covers the German 1:50 – m,cm convention, and that finding still stands — this is not a redundancy call. It was withdrawn because it appeared in none of the four surveyed projects and so fails the <40% rule. The corpus is UK/US with no German project in it, so treat this as unevidenced here, not not a real field — see titleblock-field-survey-2026-09.md §7.3
DocumentType 🟢 Core 🟢 Core, with a warning UC-01 v0.2 introduced a DrawingType (Plan / Section / Elevation / Detail / Schedule) and v0.3 deleted it as a duplicate of metadata:LayoutContentType (change G4). ADIRO's DocumentType is DIN Planart + ISO 19650 type codes — design stage and document kind, a genuinely different axis from view type. Keep it, but say so in the rdfs:comment, or it will be deleted for the same reason DrawingType was
SuitabilityStatus, DocumentLifecycleStatus 🟡 Useful 🟡 Useful, but no CQ demands the split UC-01 satisfies CQ 7.1/7.2 with a single StatusCode + statusLabel. RES-A-9 finding 5's three-way split is justified by standards fidelity, not by any current query. That is a legitimate reason — but it should be stated as such rather than presented as a requirement
ContainerState / hasContainerState 🟠 Marginal 🟠 Marginal — confirmed, and now doubly so Not printed on a sheet (internal review) and no CQ in any of the four use cases references CDE location

2.4 The traceability gap (finding 2)

Cross-referencing all four ORSDs' CQs against the 73 terms, roughly 40 have no competency question behind them — including every one of these:

Organisation · hasClient · hasLegalOwner · hasOriginator · hasResponsibleDepartment · hasTechnicalReference · ConfidentialityClassification · hasConfidentialityClassification · SuitabilityStatus · ContainerState · ClassificationCode · hasClassification · keyword · Language · hasLanguage · planKey · paperSize · hasProjectionMethod · purpose · intendedUse · validFrom · validUntil · organisationIdentifier · sequentialNumber · volumeSystemCode · levelLocationCode · supplementaryTitle · numberOfSheets · numberOfPages · pageNumber · revisionDescription · supersedes · hasRole · the whole §10 provenance layer

Some of these are obviously valuable — hasClient is on nearly every AEC title block, and planKey buys free supervision. The problem is not that they are bad terms. It is that the repo's stated methodology cannot justify them, because the demand comes from somewhere the LOT pipeline does not model: the RES-A-13 extraction experiment, whose driver is "capture whatever the title block prints", not a competency question.

Two honest ways to close this:

  • (a) Write an ORSD for the extraction use case. "As a data engineer, I want every field printed in a title block captured with provenance and confidence, so that…" — with real CQs ("which extracted values on sheet X have confidence below 0.8?", "which sheets assert conflicting clients for one document?"). This legitimises the provenance layer, conflictsWith, and the assertion-level modelling in one stroke, and it is the missing use case that RES-A-9 finding 2 is really serving. Recommended.
  • (b) Cut the vocabulary to use-case demand. Defensible, but it would delete the extraction experiment's reason to exist.

Doing neither leaves ~40 terms that the repo's own README says should not exist.

2.5 Gaps — what the use cases need that the title-block vocabulary lacks

Minor, but worth noting since the review runs both ways:

  • DrawingPackage / packageName (UC-01 FR 8) — volume/package grouping. Title blocks do often print a volume or package reference, and the proposed vocabulary has no term for it. dm: covers it; the extraction profile should include it.
  • layoutIdentifier (UC-03, pending UC-01 v0.4) — a layout number within a sheet. Printed near views rather than in the title block, so probably correctly absent, but it interacts with sheetNumber.
  • No gap on the 17 core extraction fields — all are covered by the proposal or by dm:.

3. Part B — against DANO

Partly superseded

The findings in this part stand. The recommendations drawn from them in §3.2, §4 and §5 do not, and §3.4's "complementary, not competing" is narrower than it reads. What replaced them, and why, is in the DAnO comparison.

3.1 What DANO is

Drawing Analysis Ontology — https://w3id.org/dano, prefix dano:, CC BY 4.0, from RUB Informatik im Bauwesen (the same group as exdoc). Scope, in its own words: "concepts of technical drawings from the perspective of computer vision-based drawing analysis". 16 classes, 12 object properties, 7 datatype properties.

Maintenance: repo created 2025-03-11, last pushed 2025-07-15 — quiet for about 13 months, 5 stars, no tags or releases. Better than DiCon (dormant since 2022) and it has a permanent w3id.org IRI rather than a version-pinned path, but there is no release discipline. Treat as borrowable, not dependable.

3.2 The provenance overlap — DANO covers §10 substantially

This is the real find. ADIRO §10 proposes six provenance terms; DANO already has most of that ground:

ADIRO §10 proposal DANO equivalent Assessment
extractionConfidence (xsd:float) dano:hasConfidence (xsd:decimal, on DrawingElement) ~~Direct equivalent. Align or reuse~~ → superseded — minted as aprov:hasConfidence; see the DAnO comparison
extractedFrom → SourceFile (object) dano:inferredFrom (xsd:string, "the origin file") DANO uses a string. Corroborates that SourceFile-as-a-class may be over-modelled for current needs
extractedByModel → MLModel (object) dano:inferredBy (xsd:string, actor/company/software) + dano:inferredWith (xsd:string, algorithm) DANO splits actor from algorithm, which is a better decomposition than ADIRO's single property — and both are strings, independently corroborating the #61 verdict that MLModel is over-modelled
(nothing proposed) dano:inferredAt (xsd:date) 🔴 GAP IN ADIRO. No extraction timestamp is proposed anywhere. Re-extraction comparison is impossible without one. Add it
hasBoundingBox → BoundingBox dano:hasGeometry, dano:defaultGeometry DANO links to a geometry rather than a bespoke bbox class — more general and reusable. Worth considering instead of minting BoundingBox
hasValidationStatus (none) ADIRO-unique. DANO has no human-in-the-loop concept. Keep
conflictsWith (none) ADIRO-unique. Keep
(nothing proposed) dano:hasIfcRepresentation (xsd:string) Cheap alternative to the §9 IFC alignment axioms for the common case

3.3 DrawingElementMeta bears on the pass-5 blocker

Pass 5 is blocked on RDF-star vs reification for per-value provenance (RES-A-9 next step 5). DANO has already made that choice:

DrawingElement --hasMeta--> DrawingElementMeta
                              ├── inferredAt   (date)
                              ├── inferredBy   (actor/software)
                              ├── inferredFrom (origin file)
                              └── inferredWith (algorithm)

It is reification via a meta-object — the option RES-A-9 §10 describes as "uglier but far better tool support". A peer research group hitting the same problem chose the same way, which is real evidence for that branch of a decision currently held open.

Note DANO's asymmetry: hasConfidence sits on the element, not on the meta-object, while the four inferred* properties sit on the meta. That is worth copying deliberately or rejecting deliberately — confidence is per-assertion, whereas the inference metadata is often shared across every element from one run, so putting them in different places is arguably correct rather than sloppy.

Recommendation: use DANO as evidence in the pass-5 decision. (The companion recommendation to align tb: provenance to dano: was withdrawn — see the DAnO comparison.) Do not import DANO — it would drag its 16 drawing-element classes into a suite with a blocking DL reasoner for the sake of seven datatype properties.

3.4 DANO is relevant beyond the title block — flag to other use-case owners

Out of scope for RES-89, but this review found it and it should not be lost:

DANO terms Relevant to Note
SectionSymbol, Terminator, refersTo, isReferredToBy UC-03 cross-sheet linking UC-03 mints csymbol:ReferenceSymbol and referencesLayout (no inverse — see #21). DANO has a published equivalent pattern. UC-03 is already implemented, so this is an alignment question, not a rebuild — but worth knowing before UC-03 v0.3
Dimension, DimensionChain, DimensionLine, AxisLine, hasGeometry UC-07 wall measurement UC-07 needs on-drawing lengths and angles. DANO models dimension lines and geometry natively
depicts, isDepictedBy UC-06 materials UC-06 mints depictsMaterial / isDepictedOn. DANO's generic depicts / isDepictedBy could be the parent property
TextField, TextElement, isText The title block itself A title block is a dano:TextField (a composite of min 2 TextElements). This is complementary, not competing: DANO gives the CV layer (here is a text region, here is its string), tb: gives the semantic layer (this string means "checked by"). The clean division is dano: for what was seen, tb: for what it means
contains, isContainedIn dm:contains Already covered internally

4. Revised rating deltas

Changes to Discussion #61 §11.1 implied by this review. #61 has not yet been updated — apply these before it is used for pass-0 review.

Term #61 Revised Driver
Discipline 🟢 Core 🔴 Redundant dcommon:Discipline (§2.3)
scale 🟢 Core 🔴 Redundant dm:hasScale + UC-07 scaleRatio (§2.3)
hasNorthPointOrientation 🟠 Marginal 🔴 Redundant UC-07 northArrowAngle (§2.3)
MLModel 🟠 Over-modelled 🟠 Confirmed — drop to a literal DANO uses strings (§3.2)
SourceFile 🟢 Core 🟡 Useful — consider a literal dano:inferredFrom is a string (§3.2)
BoundingBox 🟢 Core 🟡 Useful — consider a geometry dano:hasGeometry (§3.2)
extractionConfidence 🟡 Needed, blocked ~~🟡 Align to dano:hasConfidence~~ → 🟢 Minted as aprov:hasConfidence §3.2; see the DAnO comparison
extractedByModel 🟡 Useful 🟡 Split into actor + algorithm dano:inferredBy / inferredWith (§3.2)
NEW — extraction timestamp (absent) 🟢 Core — add it dano:inferredAt; re-extraction comparison needs it (§3.2)
DocumentType 🟢 Core 🟢 Core + comment required Must state it is not LayoutContentType (§2.3)
SuitabilityStatus, DocumentLifecycleStatus 🟡 Useful 🟡 Useful — standards-driven, not CQ-driven §2.3

Net: three more terms drop to redundant (8 total), one new core term appears, and four provenance terms change shape. The mint shrinks from ~48 to roughly 44, and gains a timestamp.


5. Decisions this review asks for

# Decision Owner Blocks
1 Sheet-level (UC-01) vs titleblock-level (RES-A-9) placement — including the "pipeline stage" fourth option in §2.1 UC-01 owner + Alessio pass 1. Supersedes the §1.1 framing
2 Write an extraction ORSD, or cut to use-case demand (§2.4) Alessio the legitimacy of ~40 terms
3 Confirm dcommon:Discipline reuse and drop the SKOS Discipline — pass 3
4 Drop scale, hasNorthPointOrientation; bind to dm:hasScale / UC-07 northArrowAngle — pass 2
5 ~~Align provenance to dano: rather than minting parallel terms~~; add the timestamp — Resolved. Timestamp added as aprov:inferredAt; the alignment half is reversed — see the DAnO comparison
6 Use DANO's DrawingElementMeta as evidence in the RDF-star vs reification decision — pass 5
7 Flag DANO to the UC-03 / UC-06 / UC-07 owners (§3.4) — nothing here

6. Method and limits

Read in full: UC-01 ORSD v0.3 (402 lines), UC-06 ORSD v0.1, docs/specification/use-cases/README.md; UC-07 and UC-03 entity/traceability sections; the DANO specification pages; the DANO repository's activity.

Not read: UC-07 and UC-03 in full (CQ sections skimmed via their traceability matrices); the raw DANO TTL — DANO's terms come from its generated specification, so exact domains/ranges should be confirmed against the source before any alignment axiom is written. UC-02, UC-04 and UC-05 have no ORSD, so "no use case needs this" means no use case that has been written down — UC-05 (contractor design comparison) in particular could plausibly need conflictsWith and the assertion-level model.

The "~40 terms have no CQ" count is a cross-reference of written traceability matrices, not a judgement that those terms are useless — §2.4 says so explicitly. Three of the seven use cases being unwritten is the larger gap here, and it means this review can under-state demand but not over-state it.