Skip to content

The entity and query model

The knowledge base stores more than a searchable copy of prose. It preserves typed entities and relationships so a resolver can answer bounded questions deterministically.

PS3.6 data-element records include the tag, name, keyword, VR, VM, retirement state, edition, and source reference.

A range tag remains distinct from a concrete tag, and tag input is normalized before lookup.

UID registry records include the UID value, name, keyword, type, referenced part, retirement state, edition, and source reference.

Transfer Syntax lookups enrich applicable UID records with PS3.5 details.

The graph includes:

  • IODs;
  • modules;
  • macros;
  • IOD-to-module uses;
  • functional-group uses;
  • attribute uses;
  • macro include rows;
  • conditions;
  • SOP Classes;
  • SOP-Class-to-IOD links.

This graph supports questions about composition and contextual use. It avoids treating every occurrence of an attribute as one global requirement.

Enumerated values and defined terms share a storage family but retain distinct term kinds.

A term can link to:

  • a PS3.6 data element;
  • a PS3.3 attribute-use row;
  • a source section or variable list;
  • contextual module, macro, IOD, or SOP Class paths.

The resolver preserves multiple contexts rather than unioning them automatically.

A VR record can preserve name, class, length notes, padding behavior, character-repertoire notes, and binary/text classification.

A detail record can preserve explicit-VR behavior, endian behavior, encapsulation, compression family, and bounded notes linked to its UID.

DICOM media-type records carry service context, transfer-syntax constraints, directions, and source references.

DICOMweb transaction records carry transaction name, resource category, HTTP method, route template, request and response constraints, status codes, and media-type references.

Template metadata and ordered rows preserve TID, extensibility, relationship type, value type, concept name, cardinality, condition, and include-TID relationships.

Context-group metadata and ordered rows preserve CID, extensibility, version, coding scheme, code value, meaning, and include-CID relationships.

Coded concepts are locally derived from parsed context-group rows to support lookup and joins. They are not a separately distributed terminology product.

Generic DocBook records preserve sections, anchors, tables, cross-references, and text. SQLite full-text search helps locate a section, while retrieval returns a bounded excerpt.

Text records complement typed entities; they do not replace them.

Natural-language labels and reused identifiers can map to more than one record. The resolver returns candidate information when the stored graph cannot select one result deterministically.

This is a feature of the model: ambiguity remains observable instead of being hidden by a fuzzy guess.