Documentation

NPI Data Product Reference

Understand the M2 release contract, artifacts, review controls, interpretation boundaries, and staged path to future programmatic access.

M2 published
1,578,155 reviewed rows
Manual publication
On this page

Overview

NPI Data Hub is a staged product layer over official public provider data. The current public surface reports a manually published catalog release and its evidence; it does not expose provider rows and is not a live commercial provider API.

Source

CMS NPPES

Storage

Versioned objects

Promotion

Operator only

Release contract

A product definition states entity type, taxonomy rule, geography rule, row grain, included fields, source version, and known limitations. Counts belong to a release—not to the calendar date.

Representative product definition
{
  "entity_type": "Type 1",
  "primary_taxonomy": "Dentist",
  "practice_state": "CA",
  "release_state": "published",
  "commercial_download": false
}

Artifacts

The hosted worker prepares five candidate artifact classes: normalized Parquet partitions, a quality report, a 100-row sample, a delivery ZIP, and a manifest that binds the evidence together.

Immutable object layout
releases/{release_id}/
  normalized/
  quality/report.json
  samples/provider-sample.csv
  delivery/provider-list.zip
  manifest.json

Publication

The Railway worker can claim a run, heartbeat, acquire sources, build artifacts, freeze evidence, and submit a candidate. It cannot publish or roll back.

An authorized operator reviews the frozen candidate and changes the public pointer atomically. Every decision is appended to the audit history.

Interpretation limitations

An NPI is an identifier. It does not by itself prove current practice, licensure, credentialing, specialty, availability, purchase intent, or willingness to be contacted. NPPES fields are source-reported.

Planned API

Public lookup, search, bulk enrichment, API keys, quotas, and webhooks are later product phases. No live API endpoint or paid API entitlement is advertised in M2.

Release semantics come first

Future endpoints must return release identity, source date, provenance, and warnings instead of creating a second source of truth.