Summary
The create-ecc skill produces a single, evidence-backed characteristic and
validates it. This issue captures the larger direction it's a first step toward.
Vision
Define each CC4K diagnosis term as a computable function of characteristics
(and their values), so samples are classified from their characteristics rather
than curated by hand. When knowledge changes (e.g., WHO CNS4 → CNS5), we update
the definitions and re-run classification: samples re-map automatically and we
get a diff of what moved — instead of an expert re-reading and manually
re-curating every sample.
What an ECC likely needs to get there
- Logical definitions: each term as AND/OR/NOT over characteristics,
distinguishing essential (necessary) vs desirable (supporting) criteria à la
WHO CNS5; ideally OWL equivalent-class axioms so a reasoner classifies.
- Computable values: structured tokens plus assay/method, units, threshold,
evidence level, and explicit missing/indeterminate semantics.
- Typed cross-ontology mappings (not bare codes): SSSOM-style predicate +
justification + confidence — NCIt/MONDO/ICD-O/SNOMED for terms, HGNC/Sequence
Ontology/HGVS for molecular characteristics.
- Provenance & versioning: source edition (e.g., "WHO CNS5, 2021"),
curator/date, and supersession links (replaces / deprecated-by / split-into) so
updates migrate mappings instead of breaking them.
- Interoperability: persistent URIs for characteristics and values; emit
OWL/RDF/JSON-LD so other ontologies can consume ECC.
Open design question
WHO diagnostic logic is non-monotonic (grading, NOS fallbacks, hierarchical
tie-breaks). We need to decide how much lives in ECC definitions vs. a separate
classification ruleset.
Current status
The create-ecc skill covers authoring + validation of individual
characteristics — the substrate. The mapping/definition/classification layer is
future work tracked here.
Summary
The
create-eccskill produces a single, evidence-backed characteristic andvalidates it. This issue captures the larger direction it's a first step toward.
Vision
Define each CC4K diagnosis term as a computable function of characteristics
(and their values), so samples are classified from their characteristics rather
than curated by hand. When knowledge changes (e.g., WHO CNS4 → CNS5), we update
the definitions and re-run classification: samples re-map automatically and we
get a diff of what moved — instead of an expert re-reading and manually
re-curating every sample.
What an ECC likely needs to get there
distinguishing essential (necessary) vs desirable (supporting) criteria à la
WHO CNS5; ideally OWL equivalent-class axioms so a reasoner classifies.
evidence level, and explicit missing/indeterminate semantics.
justification + confidence — NCIt/MONDO/ICD-O/SNOMED for terms, HGNC/Sequence
Ontology/HGVS for molecular characteristics.
curator/date, and supersession links (replaces / deprecated-by / split-into) so
updates migrate mappings instead of breaking them.
OWL/RDF/JSON-LD so other ontologies can consume ECC.
Open design question
WHO diagnostic logic is non-monotonic (grading, NOS fallbacks, hierarchical
tie-breaks). We need to decide how much lives in ECC definitions vs. a separate
classification ruleset.
Current status
The
create-eccskill covers authoring + validation of individualcharacteristics — the substrate. The mapping/definition/classification layer is
future work tracked here.