types

Versioned cache records.

Each sub-package is one logical cache table (<name>/) containing one package per schema version (v1/, v2/, ...). A version package vertically integrates the SQLAlchemy ORM (schema.py), the Pydantic record (model.py), the CRUD helpers (store.py), and its inbound migration (migrations.py), and re-exports canonical aliases from its __init__.py. Discovery is driven by bitfount.cache.registry.

The ORM (schema.py) is the single authoring source of truth for a version's table. migrations.py holds an Alembic op-API upgrade(op) / downgrade(op) pair taking v(N-1) -> vN, autogenerated from the ORM diff by python -m bitfount.cache.tools.gen_migration <name> <N> — never hand-authored from scratch. v1 has no predecessor, so its migrations.py defines neither.

Mental model — who owns what (this diverges from canonical Alembic on purpose):

• The ORM owns table existence. A missing table is created directly from its target-version ORM by ensure_types (implementations/sqlite.py), not by replaying a v1 create_table migration. The additive-only invariant is what makes creating straight at the target safe: every later version is a superset, so lower-version consumers still read/write the table correctly.
• Migrations own the incremental delta between two existing versions (vN.upgrade = v(N-1) -> vN) — the ALTERs (add/rename/drop/retype) autogen emits. There is no v1 create_table / drop_table: v1 has no prior version, a v1 create would merely duplicate the ORM, and the driver never reaches it (_apply_upgrades walks current+1 .. target with current >= 1).

Consequence: a fresh cache open is one table.create() per type from the ORM — never a full v1..vN chain replay.

Module

Submodules

• bitfount.cache.types.cst_calculation - cst_calculation cache type.
• bitfount.cache.types.ehr_data - ehr_data cache record — per-patient EHR demographics and code lists.
• bitfount.cache.types.ehr_patient_ids - ehr_patient_ids cache record — patient IDs from the EHR listing endpoint.
• bitfount.cache.types.file_metadata - file_metadata cache record — filesystem metadata for datasource files.
• bitfount.cache.types.flow_specs - flow_specs cache record — persisted replay material for background runs.
• bitfount.cache.types.fluid_calculation - fluid_calculation cache record — per-scan fluid volume metrics.
• bitfount.cache.types.ga_calculation - ga_calculation cache type.
• bitfount.cache.types.ga_calculation_without_fovea - ga_calculation_without_fovea cache type.
• bitfount.cache.types.gcc_calculation - gcc_calculation cache type.
• bitfount.cache.types.lesion_calculation - lesion_calculation cache record — per-scan en-face area metrics.
• bitfount.cache.types.model_inferences - model_inferences cache record — raw model inference results.
• bitfount.cache.types.patient_eligibility - patient_eligibility cache record — rolled-up per-patient verdicts.
• bitfount.cache.types.patient_enrichment - patient_enrichment cache record — externally supplied patient data.
• bitfount.cache.types.patient_level_eligibility - patient_level_eligibility cache record — per-patient EHR eligibility verdicts.
• bitfount.cache.types.patients - patients cache record — patient identity rows.
• bitfount.cache.types.runs - runs cache record — lifecycle tracking for flow executions.
• bitfount.cache.types.scan_eligibility - scan_eligibility cache record — per-scan eligibility verdicts.
• bitfount.cache.types.scan_metadata - scan_metadata cache record — vendor-normalized per-scan header metadata.
• bitfount.cache.types.schema - Shared columns for every versioned cache record.
• bitfount.cache.types.schema_versions - schema_versions cache record — pipeline schema version snapshots.
• bitfount.cache.types.trials_published_data_pointer - trials_published_data_pointer — per-project published task-hash pointer.

---