Data dictionary · provenance + exported fields

Provenance model

Provenance is a side table, not a footnote. For each exported field in a record there is a corresponding provenance row recording the source identifier, source URL, source class, and effective value state. Values without commercial-use cleared provenance are excluded from the paid export path.

{
  "field":         "field_name",
  "value":         "...",
  "source":        "cms-sdud",
  "source_class":  "government | registry | manufacturer | aggregator | derived",
  "source_url":    "https://data.medicaid.gov/dataset/158a1baa-5506-400a-8ec3-97756f0b0536",
  "effective":     true,
  "superseded_at": null
}

Export shape

The primary entity is drugs. The export combines the canonical entity table with active provenance from 13 sources. JSON responses and the full JSON download include nested provenance. CSV, Parquet, and SQLite are explicitly curated flat extracts and do not carry nested or separate provenance tables.

Key fields

These fields are selected from the field catalog as the main dimensions for filtering, grouping, matching, and API usage:

• -idRxAtlas canonical drug-product id.Record identity
• -primary_sourceSource code for the first source that created the canonical record.Record identity
• -ndc_listDistinct NDCs observed across source records for this canonical product.Record identity
• -source_idFirst-seen source identifier used to seed the canonical record, such as an NDC, RxCUI, SPL set id, or application number.Record identity
• -source_record_urlURL for the latest source record that contributed to the canonical row.Record identity
• -rxcuiRxNorm concept identifier when the product has an unambiguous redistributable mapped RxCUI.Classification
• -ingredient_identifiersUNII ingredient associations with openFDA UNII substance-name synonyms and independent NDC/UNII source versions.Product form
• -shortagesPackage-NDC-level Drug Shortages observations, including source status, availability, presentation, dates, company, related operational information, and reason fields.Shortage signals
• -brand_nameMarketed brand, label title, or branded RxNorm concept name where supplied.Product names
• -generic_nameGeneric, ingredient, or normalized RxNorm clinical concept name.Product names
• -manufacturer_nameLabeler, applicant, or manufacturer name from product-level source records.Product names
• -dosage_formDosage form for the marketed product.Product form
• -routeAdministration route, normalized as source text.Product form
• -strengthProduct strength, usually built from source ingredient strengths.Product form
• -application_numberFDA application number, such as an NDA, ANDA, or BLA. Used as a soft cross-link because one approval can cover many marketed products.Regulatory status
• -marketing_statusMarketing category or approval status supplied by FDA product sources.Regulatory status

Field catalog

The catalog below contains 28 documented fields from the same field model rendered on this page. Field types and descriptions come from the export contract and source registry. Nullable means unknown or not applicable; it does not mean zero, false, or inferred.

NULL policy

A missing field on a record is a missing field: not a zero, not a negative assertion, and not an imputed value. Fields that do not apply to a record are explicitly NULL. Coverage gaps are visible by design.

Audit trail

Every provenance row carries the observed value, source code, source record URL, effective state, and ingest context. Snapshot manifests and source licence notes document which upstream records were cleared for commercial use.