Intentional Arrangement ABox WorkbenchINSTANCES BY DESIGNKnowledge Graph Academy · Module 5 · Grounding Concepts in FactsABox Workbench
✓ Saved
TBoxCBoxFoundation is what you already built. Your ontology (TBox) gives the types: classes, and the properties each class can have. Your SKOS taxonomy (CBox) gives controlled values: concepts you pick from a list. Both stay read-only here; the facts you add in Build attach to them.
Load your model
…or drop .ttl files here
Or paste Turtle
Two files (your ontology and your SKOS taxonomy) or one file with both. Instances already in the file come in as a starting ABox.
Nothing loaded yet.
Settings for the data
Where new instances get their IRIs. The Studio uses the ontology namespace; many teams keep data in its own namespace so the ontology can be versioned on its own.
Added to every rdfs:label you write, as "Ada Lovelace"@en. Leave blank for untagged labels.
Classes & properties
Load your ontology to see its classes and properties here.
Concept schemes
Load your SKOS taxonomy to see its concepts here.
CQQuestions are what your data must answer. Fill in the blanks: which class the question is about, and what must be true about it. The answer appears live from your facts, and the same question appears as the SPARQL query you hand in for Activity 3.
The instances in the answer are lit. A dashed step is the taxonomy climb through skos:broader.
instancecontrolled valueclassclick an instance to open it in Build
ABoxBuild adds facts about real things (Activity 2). Pick a class, name the instance, then fill in its properties: a value you type, a link to another instance, or a controlled value picked from your taxonomy. Each fact is checked against your ontology and drawn on the right as it is written: in words, as Turtle, and as a graph.
Create an instance
Your instances
Pick a class to create an instance, or select one from the list.
Your facts, in words
✓ Everything below is encoded and will export. Confirmed statements flash green.
instancecontrolled valueclassvaluedrag a node · scroll to zoom · click to edit
✓Review checks your facts against your ontology and your questions, lists what is missing or wrong, and exports the one Turtle file to hand in.
Coverage do the questions get answers?
Warrant does every instance answer a question?
Findings statements checked against the model
Export
Add facts to your knowledge graph
Your ontology gives you the types. Your taxonomy gives you controlled values. The ABox is where you add the facts, one at a time.
TBox · types
From your ontology
Classes and properties. Person, Event, Venue; name, plays instrument, takes place at. A class is a type: it has instances.
+
CBox · values
From your taxonomy
SKOS concepts. Violin, Musician, London. You pick them from a list as the value of a property. A concept is a value, never a type.
→
ABox · facts
What you add here
Real things, each typed by a class and described with properties. Paul McCartney is a Person. He plays instrument Bass guitar.
The steps: load your two files → create an instance → fill in its facts → ask a question → export the Turtle.
Module 5: Activity 1 (rework CBox and TBox) happens in your files: decide what is a class and what is a controlled value; Foundation flags names used for both. Activity 2 (write instances) is Build. Activity 3 (query your knowledge graph) is Questions, whose SPARQL you paste into the SPARQL Playground with the Everything export.
Eight rules for a judicious ABox
Each rule is checked somewhere in the workbench. The Review tab reports the ones your data breaks.
1Every instance answers a question
Create data because a competency question needs it. An instance no question reaches is maintenance with no return. Review lists the instances that answer nothing.
2One thing, one instance
Paul McCartney is one instance however many events he performs at. Two instances with the same label in the same class are usually one thing entered twice.
3A concept is a controlled value, never a class
Type with a class: ex:paul-mccartney a ex:person. Tag with a concept through a property: ex:paul-mccartney ex:plays-instrument ex:bass-guitar. SKOS is SKOS: a concept stays in the taxonomy and is picked as a value. Writing ex:paul-mccartney a ex:musician turns a value into a class and breaks both layers.
4Domain and range are promises, and a reasoner believes them
ex:performs-at has domain Person and range Event. Assert it between a Venue and a Person and nothing fails; the reasoner infers that the Venue is a Person and the Person is an Event. A mismatch is an inference you did not intend. A property that takes controlled values has the range skos:Concept.
5Things you will ask about are instances; values you only read are literals
If a question asks which events share a venue, Venue is a class with instances. If no question ever asks about a capacity beyond reading it, the capacity is an integer literal.
6Pick the most specific value that applies
The hierarchy gives the broader matches for free: a venue located in London answers a question about the United Kingdom through skos:broader, and Bass guitar answers a question about String instruments. Picking the broad term loses the detail and gains nothing.
7Labels are for people, IRIs are for machines
An IRI is permanent and opaque; a label can change language, spelling or case without touching a single link. Never rename an IRI to match a new label.
8Stop when the questions are answered
Enough data to answer every question with a believable result. Extra rows do not improve the model; they make it slower to check and easier to get wrong.