Data sources

Public sources, visible caveats, and no unsupported claims.

The site is designed to be useful because every listing is tied back to a source. The current directory includes 21,000+ source-backed listings, and these source notes explain what those records can and cannot prove. They help identify public resources to verify; they do not prove individual GLP-1 coverage outcomes.

Two kinds of source, kept apart

There are two different things on this site being sourced, and they are not interchangeable. The first is the directory: rows of providers, pharmacies, health centers, and counseling programs that came out of a public dataset. The second is the guidance: the statements about what a Part D plan may require, what the Medicare GLP-1 Bridge is, and what each medicine is approved to treat. Directory rows come from the datasets listed first below. Guidance comes from CMS, Medicare.gov, and FDA pages listed second. A dataset never backs a claim about coverage, and a CMS page never puts a listing in the directory.

How the sources are joined into one record

Each dataset carries its own identifier for an organization, so records cannot simply be stacked on top of one another. Every row keeps the dataset it came from and that dataset’s own id for the row, stored as a source name and source id pair. That pair is what makes a record traceable: it is what lets the same row be found again in the source on a later refresh, and it is what stops two different datasets from overwriting each other’s version of the same place.

Where two datasets describe the same provider, the NPI is the common thread, because it is the identifier that follows a provider across datasets. Where they describe the same physical location without sharing an identifier, matching falls back to the address. That address match is where duplicates come from: two rows can describe one clinic at one address, and both are correct. Rather than deleting one, the site records that the rows share an address and points the duplicate’s canonical at the primary row, so there is one address to index instead of two competing for it.

What stays visible on every listing

Provenance is not a footnote here; it is carried on the record itself. Every listing keeps the name of the dataset it came from, that dataset’s id for the row, a link to the dataset, and the date the row was last verified against it. A record also carries a confidence value, which reflects how directly the source supports the category the row was filed under, and it carries the caveat shown with its source. When a source page is still live, you can open it and check the row yourself instead of taking the listing on faith.

What these sources do not contain

None of these datasets records whether a given clinician prescribes a GLP-1, whether a pharmacy has a particular dose in stock today, which drug list a plan uses, or what a plan will approve for a specific person. None of them records prices. Those things are not missing by accident: they are not published in these sources at all, and a site that showed them would be guessing. That is why every page tells you what to confirm rather than what is true for you, and why nothing here states that a medicine is covered.

The datasets also carry ordinary lag. A practice moves and the address is still listed. A phone number changes. A health center site closes. The row is accurate as of the source it came from and the date shown with it, which is why the date is shown.

Keeping sources current

Sources are refreshed rather than hand-edited, and pages that read from the directory are revalidated on a schedule rather than on every request. When a refresh no longer returns a row, the record is marked stale rather than deleted, and stale rows are filtered out of every listing, count, and sitemap entry. Keeping the row is deliberate: if a source returns it again, the record’s history is still intact, and the id can be matched back.

Reading a listing back to its source

If you want to satisfy yourself that a listing is real before you call it, the route is short. Open the listing, read the source name and source id attached to it, and open the dataset it names. Search that dataset for the id, and compare the fields one by one: name, address, phone, and the taxonomy or program type the record was filed under. If the dataset is the NPI Registry, the id is the NPI and the record is public. If it is HRSA or Care Compare, the id is that program’s own site or clinician number.

What you should expect to find is that the directory agrees with the source on those fields and says nothing the source does not. What you should not expect to find is a claim about GLP-1 coverage, because none of these datasets makes one. A listing that agrees with its source is working as intended even if the clinic no longer prescribes the medicine you want — that is the kind of thing only a phone call settles, which is why every page ends with what to confirm rather than what is true.

How these sources are listed

Where the coverage guidance comes from

These pages back what the site says about plan rules, the Bridge, and drug approvals. They are public, and you can read them without an account.