Method
Seven steps. The first sources are public. The website is accepted only when the page corroborates the registry identity. Unknown is allowed at every step.
01
NPPES publishes a weekly file of clinician identifiers. CMS publishes Medicare clinician data monthly. Open places data adds location candidates that can share a phone and a ZIP.1
These sources describe clinicians and places. They do not, by themselves, describe a practice website, a published email, or who owns the office.
02
A clinician, a legal organization, an operating brand, and a location are different things.2 The pipeline keeps them separate and joins them when the evidence agrees: a shared site of practice, a consistent brand, a provider listed at that address.
A shared phone number or a shared building is a candidate relationship, not a merge.3 Conflicting suites, brands, or specialties stay in review.
03
Places that share a phone, and search when that is not enough, propose candidate domains. A directory listing is a clue. It is never accepted as the practice's own website.4
A domain guessed from the brand name stays a hypothesis until the page is fetched and read. A successful HTTP response proves that a server answered, nothing more.
04
A site is attached to a location only when the page itself corroborates the registry identity. The conservative rule is a matching phone plus name or address, or the provider's name plus street and locality, in the same page context.5
A name on one page and a city in a national footer does not count. If the page contradicts the target, a different state, a provider who has left, a conflicting suite, the site is not accepted.
05
From a matched site, the pipeline reads published services, team names, locations, hours, booking links, contact forms, and technology fingerprints such as a patient portal or a scheduling tool.6
Emails are kept only when they appear on that page. They are not invented from a name and a domain. A vendor or developer address that is not the practice's channel is left out.
06
Each field can carry a status. Confirmed means the evidence met the identity rule or came straight from the registry field it cites. Seen on website means the practice published it and the quote was stored. Possible match means a candidate that has not cleared the rule.7
Unknown stays empty. A missing website is "not resolved", not "they do not have one".
07
Weekly registry diffs and later site checks emit events: provider added, moved, or deactivated; phone changed; website found; group size changed.8 You can subscribe instead of re-pulling the country.
A quiet week means we did not observe a change. It is not proof that nothing changed.
Limits
Get a key, read a record, and widen the query when the shape is right.