0010 Versioned Bioloupe Disease Taxonomy

Versioned Bioloupe Disease Taxonomy Owns Runtime Publication

Status: accepted

The default disease publication path reads one checked-in, reviewed Bioloupe taxonomy release. Each release owns its NCIt-anchored nodes, aliases, and product-DAG edges. Release identifiers are content-immutable, and one explicit active-release pointer determines what canonical.disease, canonical.disease_hierarchy, and the q_diseases family publish.

The first release recovered useful product intent from legacy Data Gov’s simplified=true set and required parent closure, then became a repository-owned artifact. Legacy state is no longer read by the default graph and cannot silently expand or alter the active taxonomy.

Considered Options

• Read the mutable legacy simplified set on every run.
• Let every exact source mention or NCIt result expand the published disease universe.
• Import the full NCIt hierarchy as the product rollup graph.
• Publish an immutable, reviewed Bioloupe release and keep legacy/authority differences as reconciliation or review demand.

Consequences

• Runtime publication is reproducible without Data Gov access.
• Reusing a release id with different bytes fails.
• Source mentions can resolve to active diseases or create review demand, but cannot mint taxonomy rows by themselves.
• NCIt anchors identity and may suggest changes; the reviewed Bioloupe DAG owns product rollups.
• migration.legacy_disease_taxonomy_seed remains disabled-by-default reconciliation evidence and never feeds the normal canonical graph.
• A taxonomy change requires a new reviewed release, tests for uniqueness/closure/acyclicity, and an explicit active-release change.