homeobox

Finalize harmonized Lance tables in a data package — assign automatic columns (uid, dataset_uid, derived fields), connect tables by populating RegistryKeyFields, and validate every table against its target homeobox schema.

Describes the creation of a data package, which is a collection of datasets organized in a directory as a prerequisite for agent-based data standardization and alignment. Provides guidelines for tagging files and a reference for using the Collection API.

Use this skill to harmonize raw collection- and dataset-level tables in a data package to a homeobox schema. Covers the resolution procedure for adding new columns, replacing values, and using ontology and database resolution tools.

Write an ingestion script that adds a finalized collection of harmonized datasets to a homeobox atlas. The script wires per-feature-space loaders to polycomb.ingestion.ingest_collection; the only real work is turning a dataset's raw DATA files into a homeobox Reader.

Use after create-data-package when a coalesced data package and homeobox schema YAML IR are ready. Stages per-dataset OBS and VAR, the per-dataset DatasetSchema scaffold, and collection-level LIBRARY tables into Lance.

Use when designing or writing a homeobox atlas schema. Produces a declarative schema YAML (the homeobox schema IR) that codegens into a schema.py — covering the obs/dataset/feature-registry/fk-registry/other table sections, enums, PointerField markers with feature_registry_schema, all informational field markers (stable_uid, registry_key, polymorphic_registry_key, ontology_aligned, cross_reference, summary), declarative constraints (require_any, equal_length), computed fields, and presence flags. Validates the YAML with scripts/validate_schema_ir.py.

Use after prepare-package-for-resolution on multimodal datasets. Reconciles per-modality cell barcodes across feature-space obs tables and writes a canonical multimodal_barcode column via audited curation.