How the map is built

What it is not

• Not about people. Every layer describes a business or a building. No layer uses, infers, or stores anything about residents: no census person data, no voter data, no demographics, no one's health status.
• Not a referral. A listing means a public registry says the business exists under a relevant category. Reviewed providers live in the directory, where the trust criteria are published.
• Not a legal opinion. Compounding, telehealth, and controlled-substance rules differ by state. The pharmacy layer asserts licensure or FDA registration only, never that a particular prescription is lawful in a particular state.
• Not for housing, lending, or insurance. The index must not be used in, or marketed toward, any decision about a person's housing, credit, or insurability. The data license (CC BY-NC 4.0) and this page say so.

Coverage

• National overview · TRT, GLP-1, pharmacy · H3 r4–r5 · scored against the national distribution.
• Every state · the registry layers, plus gyms inside metros · H3 r6–r7 · scored nationally.
• 20 metros · all four layers · H3 r7–r9 · scored against the metro's own distribution, with the national score alongside.

Metros: Dallas–Fort Worth, Houston, Austin, San Antonio, Phoenix, Atlanta, Miami–Fort Lauderdale, Tampa Bay, Las Vegas, Nashville, Los Angeles, New York, San Francisco Bay Area, Seattle, Boston, Washington, DC, Denver, Salt Lake City, Chicago, Minneapolis–St. Paul. The gym layer and the OpenStreetMap pharmacy candidates exist only inside these boxes; outside them the index uses the three registry layers with their weights renormalised, and the hex says so.

Sources (public, business-level)

Zips resolve to Census ZCTA centroids (2023 Gazetteer), with GeoNames filling names and non-ZCTA zips. Street geocoding uses the Census Bureau batch geocoder; rows it cannot match stay at their zip centroid and are labelled as such wherever they appear. Inside metros, Nominatim fills some gaps under its usage policy. Distances are great-circle miles from the zip centroid.

Confidence

• 0.85 and up · licensed or verified: a TSBP/FDA license, or a specialty match plus a name keyword.
• 0.6 to 0.85 · specialty match: the registry taxonomy alone.
• 0.3 to 0.6 · keyword match: a generalist whose name says hormones or weight loss, or an unlicensed pharmacy candidate.

Rows under 0.3 are not published. Confidence weights the index and breaks distance ties. It is never moved by money.

The index

For each H3 hexagon and each layer that covers it:

raw_k  = Σ confidence inside the hex + 0.5 × Σ confidence in the 6 neighbours
comp_k = min(1, ln(1 + raw_k) / ln(1 + P95_k))     P95 over the scoring region
MII    = 100 × Σ w_k·comp_k / Σ w_k                 w: TRT clinic 0.30, GLP-1 prescriber 0.25, Compounding pharmacy 0.20, Gym 0.25

P95 is taken over the whole country for national and state layers and over the metro itself for metro layers; every hex also carries the nationally scored value. The log compresses the top so one medical tower does not flatten the rest of the map. The neighbour term smooths single-listing noise. Empty hexes are not drawn. Weights are a published judgment call, revisited each quarter.

Programmatic pages

A city gets a TRT page only with five or more TRT listings and a GLP-1 page only with eight or more GLP-1 listings. Below that it is reachable through the lookup and the state map but has no page of its own: a page with two rows is noise for readers and search engines alike. Current build: 259 TRT city pages, 7 GLP-1 city pages.

Prices

A listing shows a member price range only once three or more approved reports exist, as the interquartile range so one outlier cannot move it. Below three, the published range for the category is shown and attributed. Reports never include names, emails, or phone numbers; the API rejects them.

Submissions and moderation

Members can report a price, a correction, a closure, or a missing business. Every report is created as pending and reaches the site only after a moderator approves it and the dataset is rebuilt. A report stores an account id and the form fields. IP addresses and user agents are never read or stored. Share your experience, don't prescribe to others.

Snapshots

The dataset is rebuilt and frozen once a quarter. Openings, closings, counts per city, and median reported prices are appended to a time series. A business id is a hash of its name, address, and zip, so a move or rename reads as a closure plus an opening.

Money

There is no paid placement on the map. If sponsored pins or verified-pricing badges ever ship, they will be labelled, kept out of the ranking and the index, and described here first. Any paid relationship follows the directory rule: disclosed at the link, never changes a rank, a score, or a grade.

Limits

• Registries lag; a clinic can close months before NPPES notices.
• Keyword match is exactly that. A wellness clinic's name is not a workup.
• Open Payments is a prescribing proxy, not a prescribing record.
• State board exports are manual, so 503A coverage grows state by state; until a state's file exists, its licensed compounders appear only as FDA 503B facilities or OpenStreetMap candidates.
• The gym layer is metro-only. A state hex's score outside a metro says nothing about gyms.
• About one registry row in ten sits at its zip centroid rather than its street; those rows say so.