Ontology design patterns

The MatWerk KG is built from a spreadsheet whose tabs are not arbitrary: each one encodes a design pattern. A row is an entity — an individual carrying identity and type — surrounded by value nodes that hold its values, and object properties pointing at entities in other tabs.

That pattern has until now been implicit, spread across a ROBOT directive row and whatever a curator typed. These pages make it explicit: for every tab, the terms it uses (TBox) and one real row from the live sheet (ABox), drawn, explained, and checked by a reasoner.

Generated, not written

Every page here is produced by scripts/gen_patterns.py from the live workbook and the live ontology. Re-run it after a sheet edit or an ontology upgrade and the documentation follows; it cannot drift from the data it claims to describe. Do not edit these pages by hand.

python scripts/gen_patterns.py --fetch \
    --mwo mwo-full.ttl --mwo-version 3.0.1 --out docs/patterns

Ontology version: MWO 3.0.1 (http://purls.helmholtz-metadaten.de/mwo/mwo.owl/3.0.1). Each pattern.ttl records the version it was generated against in its header.

The shape every tab shares

      ┌─────────────── entity ───────────────┐
      │  msekg:<epoch_ms><counter>           │   ← the row's identity
      │  a  <the tab's class>                │   ← the TYPE column
      │  rdfs:label "…"                      │
      └───┬──────────────────────────┬───────┘
   denoted by (IAO_0000235)       other object properties
          │                           │
   ┌──────▼───────┐          ┌────────▼─────────────┐
   │  value node  │          │  an entity in another│
   │  title / url │          │  tab (organisation,  │
   │  description │          │  person, process …)  │
   └──────────────┘          └──────────────────────┘

Two consequences are worth stating plainly, because both cause silent damage:

A column marked I holds the label of another instance, not a value. ROBOT resolves it at build time. A typo therefore produces a missing relation, not an error — 301 of 7 955 label references in the workbook currently resolve to nothing.
Every value node needs a unique label, or the reference is ambiguous.

Patterns

Core building blocks

Pattern	What a row is
value nodes	the shared value nodes — titles, descriptions, URLs, names — that every other tab points at
agent · role · process	agents, the roles they bear, and the processes those roles are realised in
temporal · city	time regions and places

Who and where

Pattern	What a row is
people	a person, with names, e-mail and ORCID
organization	an institution, with ROR, city and parent
collaboration	a collaboration between agents

What is produced

Pattern	What a row is
dataset	a dataset
publication	a publication
software	a software product
ontologies	an ontology
materials	a material designation
fdos	a FAIR digital object

Where it lives

Pattern	What a row is
dataportal	a data portal / repository
sparql_endpoints	a queryable endpoint and the dataset behind it
metadata	a metadata standard
service	a service offered
instrument · largescalefacility	instruments and the facilities housing them

NFDI-MatWerk structure

Pattern	What a row is
matwerkta	a Task Area
matwerkiuc	an Infrastructure Use Case
matwerkpp	a Participating Project
event	an event

Where the value nodes went

req_1 and req_2 are not patterns. They are the workbook's shared pool of value nodes — titles, names, websites, roles, identifiers — that every other tab's I columns point into. req_2 alone holds 4 871 rows across 70 distinct types.

Documenting them as two pages would say nothing; documenting them as seventy pages would divorce each value node from the entity it exists to describe. So they are dissolved: every value type is attributed to the entity tab that actually references it, and appears there under Components (value nodes) — where it is already present in the example, because value nodes are resolved.

Value node	Documented in	References
contact point role	process	316
publishing process · document title · author list · DOI	publication	231 · 230 · 150 · 139
software title · software repository · version number	software	209 · 124 · 85
written name	agent · ontologies	105 · 61
event frequency datum · collection	event	100 · 92
email address	people · role	76 · 56

A value type that no entity tab references keeps a page of its own — currently only nfdi matwerk consortium.

Consistency

Each pattern.ttl is merged with MWO and checked with

robot merge -i mwo-full.ttl -i pattern.ttl \
      explain --reasoner hermit -M inconsistency --explanation report.md

— the same reasoner process_spreadsheets runs over every generated module, so a pattern that contradicts the ontology fails here rather than in production.

The check is not cosmetic. It is what catches the class of error that produced the Graph Metadata bug: a description node typed academic event instead of textual description — two IRIs one digit apart, both real, both undeprecated — which made every graph carrying it inconsistent while every test still passed.

Shapes — SHACL constraints covering what a reasoner structurally cannot (missing values, deprecated terms, which node was at fault).
Harvested RDF modelling — the same treatment for graphs harvested from GitHub and Zenodo.
Pipeline — the DAGs that turn these tabs into the KG.