Where these numbers come from, and how they are processed
Every source file, every join, every judgement call — and the errors we have found in our own work.
The four source files
Everything here is built from four public datasets published by the U.S. Centers for Medicare & Medicaid Services. Open Payments carries what drug and device manufacturers report paying clinicians under the Physician Payments Sunshine Act. The Medicare Physician & Other Practitioners file carries what clinicians billed Original Medicare Part B. The Medicare Part D Prescribers file carries what they prescribed and what those prescriptions cost. NPPES, the National Plan and Provider Enumeration System, supplies the identity: name, credential, specialty, city and state.
All four are public-domain federal records. None of them is scraped, purchased, estimated or modelled. Where a figure appears on this site it came out of one of those four files, and the page it appears on names which one.
Why a doctor is a number, not a name
Clinicians are identified by National Provider Identifier, never by name. Thousands of NPIs share a common name, and joining on names would merge unrelated people into a single profile — the most damaging error this kind of site can make. Every join here is on NPI.
Open Payments contains a category of record that has no NPI at all: a manufacturer reported a payment against an internal profile identifier without one. Those get their own page keyed by that profile ID and are never merged into a same-named NPI, even when the match looks obvious.
The three programs are never added together
Open Payments reports four payment types, which resolve to three programs. General payments are meals, travel, gifts, consulting fees, speaking fees and royalties. Research payments — reported both directly and as 'associated research', where funding flows through an institution against a named principal investigator — are study funding, and the named physician frequently never personally receives the money. Ownership is an equity stake held by the clinician or a family member.
Adding them produces a number that describes nothing, and it is the single most common way this data is misreported. A trial investigator with several million dollars of associated research funding has not been paid several million dollars for promoting anything.
We got this wrong. Until August 2026 our ingest matched three type names and quietly folded anything else — including 'Associated Research' and 'Ownership/investment' — into the general-payments total. Measured across the lifetime summary file, that was 62.5% of all dollars landing under the one heading it must never land under, and it meant researchers were displayed as having received marketing money. It is fixed, unknown payment types are now excluded and reported rather than absorbed, and this paragraph stays here because a methodology page that only lists successes is not a methodology page.
What we deliberately do not publish
No street addresses, ever. City and state are enough to tell two clinicians apart and enough for search; a meaningful number of solo practitioners list a home address in NPPES, and republishing it creates a category of harm that the data does not justify.
No malpractice history and no state-board discipline records. Both are a different reliability class from federal payment data, and sitting them beside money figures invites a reader to draw a connection that neither dataset supports.
No ratings, no rankings of suspicion, no scores that purport to judge a clinician. The site computes a richness score to decide which pages to build first; it is a measure of how much data exists, not an opinion about a person.
Refresh cadence and stability
CMS publishes Open Payments annually, typically in June, and the Medicare files on their own schedule. We rebuild from the source files when a new vintage lands, and every page states the program years it covers rather than implying it is current.
URLs are stable. A clinician whose NPI is deactivated keeps their page, marked inactive, rather than having it deleted — a dead link helps nobody and a disappeared record looks like concealment.