Source capability evidence review¶
The reproducible review checks every cell of the pinned source capability contract against the materialization audit and current evidence bytes. The report retains 11 source families × 17 dimensions = 187 rows with each row's support state, materialization state, evidence paths, coverage metrics, reason codes, and human review flag.
The matrix contains 68 supported, 58 partial, and 61 unsupported capability cells. Separately, 73 cells have complete materialization, four partial, 49 missing, and 61 not applicable. The report hashes all 74 distinct present evidence files and explicitly lists 30 distinct missing paths. Human review remains required for 126 rows. A present file proves only that the declared artifact exists at its recorded digest; it does not promote a source dimension or establish scientific admission.
The producer refuses incomplete or duplicated source–dimension rows, a difference between declared support and audit support, changed evidence presence, invalid count metrics, and paths escaping the repository. Source support and evidence presence remain independent. Country coverage, transformation lineage, publication language, and source-specific scientific review are governed by their separate source and country ledgers; this review does not replace them.
The producer marks a taxon_identity cell declared unsupported by its source contract with not_provided_by_source. This reason describes source-native taxon availability; broad land-use categories do not become taxa. A supported or partially supported source with missing repository evidence retains its missing materialization state instead. Null source taxa, unmapped concepts, support states, coverage counts, and human-review requirements retain their separate meanings.
Reproduce¶
From the repository root:
PYTHONDONTWRITEBYTECODE=1 .venv/bin/python -m \
bijux_pollenomics_dev.source_completion.capability_evidence \
--repository . \
--output artifacts/source-capability-evidence/replay.json
Compare artifacts/source-capability-evidence/replay.json byte-for-byte with report.json. The focused tests exercise the support/materialization split and refusal cases.