Evidence: data/review-20260914.cvXJSA/REVIEW.md, snapshot data/merged/20260914/merged-kg.tar.gz, SHA-256 0fb80d5d865d0da503d9c759286c91216b3dab5f5cc5df0a8e759cd4e233ec81 (3,353,182 nodes / 15,295,064 edges). Counts describe the immutable reviewed archive, not a rebuilt output.
Full prefix review flags GOP24 nodes, GOREL2, ORCID49, UBERON_CORE69, WIKIDATA47 (time is a separate reviewer bug). ORCID contributor identifiers are emitted as OntologyClass nodes, indicating metadata/ontology-boundary leakage. Registration alone must not turn contributors into biological classes.
Trace each prefix to raw ontology declarations and incident assertions; distinguish namespace aliases/referenced classes from annotation-only metadata. Filter only evidence-backed annotation entities, normalize verified aliases through the canonical registry, and record exclusions. Preserve real class axioms and provenance; no blanket prefix drop.
Related modeling work already exists: #644 ontology class/category tension (621,573 subclass predicate-side violations), #645 RHEA enables/enabled_by, #642 assay Procedure input/output, #643 organism phenotype allowance. Additional source-specific cases include 11,335 ChEBI has_attribute chemical targets and FOODON in_taxon. These need semantic adjudication, not a universal OntologyClass addition or related_to replacement.
Acceptance: immutable raw fixtures showing metadata vs genuine nodes, registry/normalization tests, exact before/after evidence, and documented disposition of modeling cases without silently weakening validation.
Evidence:
data/review-20260914.cvXJSA/REVIEW.md, snapshotdata/merged/20260914/merged-kg.tar.gz, SHA-2560fb80d5d865d0da503d9c759286c91216b3dab5f5cc5df0a8e759cd4e233ec81(3,353,182 nodes / 15,295,064 edges). Counts describe the immutable reviewed archive, not a rebuilt output.Full prefix review flags GOP24 nodes, GOREL2, ORCID49, UBERON_CORE69, WIKIDATA47 (time is a separate reviewer bug). ORCID contributor identifiers are emitted as OntologyClass nodes, indicating metadata/ontology-boundary leakage. Registration alone must not turn contributors into biological classes.
Trace each prefix to raw ontology declarations and incident assertions; distinguish namespace aliases/referenced classes from annotation-only metadata. Filter only evidence-backed annotation entities, normalize verified aliases through the canonical registry, and record exclusions. Preserve real class axioms and provenance; no blanket prefix drop.
Related modeling work already exists: #644 ontology class/category tension (621,573 subclass predicate-side violations), #645 RHEA enables/enabled_by, #642 assay Procedure input/output, #643 organism phenotype allowance. Additional source-specific cases include 11,335 ChEBI has_attribute chemical targets and FOODON in_taxon. These need semantic adjudication, not a universal OntologyClass addition or related_to replacement.
Acceptance: immutable raw fixtures showing metadata vs genuine nodes, registry/normalization tests, exact before/after evidence, and documented disposition of modeling cases without silently weakening validation.