Skip to content

Graph Metadata

Every named graph the pipeline publishes carries a small block of triples in which the graph describes itself — what it is, what it is about, who loaded it, when, and how big it turned out to be. It is the same shape in every graph, which makes it both the easiest thing to get uniformly right and the easiest to get uniformly wrong.

It was uniformly wrong. The description node was typed academic event instead of textual description — two IRIs one digit apart — and because a description sits on the right-hand side of denoted by, which requires information rather than an event, every graph carrying one was logically inconsistent. Nothing caught it: the IRI was real, it was not deprecated, and the tests asserted the same wrong term the code wrote.

Visualization

Four kinds of thing appear here, and telling them apart is the whole model:

Node Kind Why it matters
the graph data collection, void:Dataset a continuant — it persists, it does not happen
the description textual description also a continuant: information, not an event
the load process the only occurrent, and so the only thing that may occupy a stretch of time
the instants temporal instant zero-dimensional; the begin and end of the load

The rule that does the work

occupies temporal region has domain process or process boundary, and is functional. It may not be written on the graph, or on the description, or on a person — only on the load. If you want to say when a continuant was valid, that is a different property and a different pattern; see Role Bearer.

What goes wrong

Diff this against the block above: the description is typed NFDI_0000018, and the instants are typed with the parent class. The first is fatal, the second is silent, and the pair is a fair summary of what a reasoner can and cannot do for you.

Fatal — the reasoner catches it. Turn reasoning on and the graph is inconsistent. Every axiom in the chain belongs to BFO, IAO or nfdicore; only the one asserted type is ours:
denoted by range information content entity
information content entity ⊑ generically dependent continuant ⊑ continuant
academic event ⊑ event ⊑ process ⊑ occurrent
continuant owl:disjointWith occurrent

An inconsistent graph is not a graph with one bad triple in it. Under OWL semantics everything follows from a contradiction, so every query against that graph is entitled to return anything at all — which is why the check belongs before publication, not after.

Silent — only the shape catches it. The instants are typed BFO_0000148 zero-dimensional temporal region rather than BFO_0000203 temporal instant. BFO_0000203 is a subclass of BFO_0000148, so the looser type contradicts nothing and no reasoner will ever mention it. It simply discards the precision the range axiom exists to provide. The SHACL shape on this page notices; HermiT never will.

The same holds for NFDI_0000009, which is deprecated in favour of IAO_0001000 data collection. Deprecation is advice, not an axiom — it cannot make anything inconsistent, so only a shape will raise it.

Declare your properties, or none of this happens

Worth stating on its own, because it invalidates the obvious way of checking any of the above.

An OWL parser must decide what kind of property a statement uses. If the file being read does not declare it, the fallback is to treat the statement as an annotation — a note, carrying no logical meaning. A reasoner does not evaluate annotations; it does not see them. So a reasoner run over a bare data export returns consistent however broken the data is, and the run looks exactly like a passing one.

Both .ttl files on this page begin with declarations for that reason:

obo:IAO_0000235 a owl:ObjectProperty .   # denoted by
obo:BFO_0000199 a owl:ObjectProperty .   # occupies temporal region

Delete those two lines from broken-data.ttl and the inconsistency disappears — the data is just as wrong, and the reasoner has been handed nothing to check. Before trusting any green reasoner result over pipeline output, confirm the check was live: inject a known-bad triple and make sure it is still caught.

SHACL Constraints

Target Constraint
void:Dataset has a non-empty rdfs:label
void:Dataset denoted by a NFDI_0001018 textual description carrying a non-empty NFDI_0001007 value
void:Dataset output of exactly one BFO_0000015 load process
void:Dataset not typed with a deprecated class
load process at least one RO_0000057 participant
load process exactly one BFO_0000199 region, a BFO_0000038 with first and last instants
BFO_0000203 exactly one time:inXSDDateTimeStamp typed xsd:dateTimeStamp

Terms

Term Name
IAO_0001000 data collection — replaces the deprecated NFDI_0000009
IAO_0000235 denoted by
NFDI_0001018 textual description
NFDI_0001007 has value
RO_0002353 output of
BFO_0000015 process
RO_0000057 has participant
BFO_0000199 occupies temporal region — functional, processes only
BFO_0000038 one-dimensional temporal region
BFO_0000222 / BFO_0000224 has first / last instant — range is BFO_0000203
BFO_0000203 temporal instant
NFDI_0000004 person