Dataset review checklist

Use this before proposing a Store recipe or sharing an importable package. Record the outcome beside the dataset recipe. A structural validation pass is not scientific validation or permission to redistribute data.

Source and permissions

• Identify the study, exact input products, retrieval date and checksums.
• Record access/use conditions, attribution and separate redistribution status.
• Check annotation and terminology sources as well as the primary data.
• Exclude credentials, private access details and restricted records from Git.

Mapping and linkage

• Distinguish patients, visits, specimens and technical/biological replicates.
• Account for every source column, including exclusions and unresolved meanings.
• Review ontology terms, units, negative findings and missing-value conventions.
• Keep longitudinal clinical snapshots distinct; preserve explicit sample links.
• Remove converter placeholders that are not observations; do not invent dates.
• Declare each assay's feature namespace, value scale and upstream processing.
• If ClarIDs are included, preserve original IDs and review their source fields.

Package and checks

• Record converter/mapping versions, preparation recipe and validator revision.
• Run the package validator; review warnings and separate BFF/terminology checks.
• Include the dataset manifest, clinical records and any declared assay files.
• For ZIP import, include resource.json and exactly one dataset-root folder.
• Record file/archive checksums; exclude DuckDB caches, accounts and user state.
• Confirm expected patient, visit, sample and feature counts after a clean import.
• Inspect clinical fields and assay availability in the app; document limitations.
• Use a new resource version for changed published contents; do not replace archives.

Review outcome

Record the reviewer, date, tested COHORTome version, aggregate counts and open issues. State separately whether the recipe can be shared, whether the data archive can be redistributed, and which analyses are appropriate. A compatible format does not make studies scientifically comparable.