How the dataset is built
1. Source first
We start from official publishers — CMS provider data, Medicare payment files, the NPI registry, FDA and NIH sources, and hospital price-transparency files. A source enters the registry with its official URL, its reuse terms, its update rhythm and its authority class before a single record is pulled.
2. Provenance is kept, not summarized away
Every record keeps its source name, source record id, source URL, the date it was retrieved, its license and a confidence score. If a number on this site cannot be traced to its source in one click, it does not ship.
3. Normalization without invention
Records are mapped to shared categories — hospitals, prices, quality, providers, drugs, devices, conditions, research, trials, public health, regulations — and linked by real identifiers (CMS certification numbers, NPIs, billing codes). Missing fields stay missing and are flagged; nothing is filled in by guesswork.
4. Price kinds are never blended
A list charge is not a cash price is not a negotiated rate is not a Medicare payment. The dataset stores each amount with its kind, exactly as published, and every screen and API response labels it. Averages are only ever computed inside one kind.
5. Deduplication and versioning
Records are deduplicated on source + source record id, with a content hash per record. When a source republishes, the record's version moves and the retrieval date updates — the history is the audit trail.
The limits, stated plainly
- Published prices are what hospitals and Medicare report — not a quote for your care. Your actual bill depends on your plan, your deductible and the exact services you receive.
- Medicare average payments are averages across many cases; they are the standard benchmark, not a prediction.
- Sources update on their own schedules. Each record shows its dates so you can judge freshness yourself.
- Coverage grows source by source. Where the dataset has no record, we say so instead of estimating.