Skip to content

Ontology and SHACL Pipeline

4. Ontology and shape-generation pipeline

4.1 End-to-end source pipeline

The demonstrated source pipeline is:

flowchart LR
    A["GS1 Navigator JSON"] --> B["gdsn-json-to-tsv"]
    B --> C["GDSN TSV module"]
    B --> D["GPC TSV module"]
    C --> E["uml2semantics-python"]
    D --> E
    E --> F["gdsn.ttl"]
    E --> G["gpc.ttl"]
    F --> H["gs1-gdsn-holon"]
    G --> H
    H --> I["Brick-scoped SHACL"]
    H --> J["Selected GDSN SHACL"]

4.2 JSON-to-TSV findings

The source investigation identified an undocumented numeric type field across GDSN UML classes. Five inferred categories were routed to different TSV outputs:

  • XML Schema primitive-like datatypes;
  • UML business classes;
  • GDSN named datatypes such as GTIN or GLN;
  • GS1 code-list classes;
  • UML enumeration classes.

Code-valued classes were excluded from Classes.tsv to prevent double representation as both classes and enumerations.

4.3 Malformed source constraints

Fifteen of approximately 2,274 attribute multiplicities were reported as malformed, with examples such as .., 1.. and 0...1. Several limits were also malformed. The transformation policy was:

  • do not guess;
  • leave generated minimum and maximum blank;
  • preserve the original expression;
  • emit gdsn:conversionWarning or equivalent diagnostics.

This is a material architecture principle because downstream SHACL and ontology constraints must not give false precision to defective source metadata.

4.4 Extended attributes

The source mapping documentation did not provide a reliable natural owner class for all extended attributes. The reference transformation therefore placed them under synthetic containers such as gdsn:GDSNAVP and gdsn:GDSNExtendedAttribute, optionally grouped by groupName.

This is an explicit modelling decision and should remain traceable in generated annotations and mapping records.

4.5 OWL generation capabilities

The reported uml2semantics-python capabilities include:

  • classes and properties;
  • datatype restrictions;
  • enumeration patterns;
  • owl:unionOf and disjointness for exclusive choices;
  • property chains supplied through a dedicated TSV;
  • source annotations and IRI construction.

The presence of property-chain support confirms the intended boundary: OWL can compose object-property paths, but conditional branching on data values remains outside pure OWL DL.

4.6 Independent convergence of the SHACL implementation

The gs1-gdsn-holon CLI was developed independently of the initial speculative examples but implemented the same central pattern:

  • generate a Brick-scoped shape from gpc.ttl;
  • optionally add a selected GDSN shape with --include-gdsn;
  • require a Brick code or GTIN-to-Brick mapping because the ontology contains no product instances;
  • run pySHACL checks by default;
  • include parse, empty-graph, conforming-instance and deliberately broken-instance tests.

This convergence is supporting evidence for the architecture. It is not proof that the reference implementation is a complete GS1 solution.