The map does not draw a radius. It uses the Census Bureau's own boundaries: the ZIP Code Tabulation Area (ZCTA) that contains the address (ZCTAs approximate USPS ZIP codes; a few ZIPs — PO boxes, single buildings — have no ZCTA), every ZCTA that touches it (the "+ neighbours" tab), and the county. Resources are counted by whether their coordinates fall inside those polygons (point-in-polygon, even-odd rule), so a store one street over the ZCTA line counts for the neighbour, not for you — exactly as it would in real life.
| Item | Source (fetched live when you press Go) | Tag |
|---|---|---|
| ZCTA / county polygons | Census TIGERweb REST services (PUMA_TAD_TAZ_UGA_ZCTA, tigerWMS_Current); layer ids are looked up by name at run time | measured |
| Address → point | Census Geocoder (JSONP), falling back to the Esri World Geocoder, then OpenStreetMap Nominatim. The ZCTA and county are then found by asking TIGERweb which polygons contain that point. The address you type is sent to those geocoders; a 5-digit ZIP is only sent to TIGERweb. | measured |
| Population, households | ACS 5-year estimates (tables B01003, B11001) shipped in data/, built by tools/build_data.py; ZCTA↔county from the 2020 relationship file | measured |
| Food retailers | USDA FNS SNAP Retailer Location Data (live ArcGIS layer, ~250,000 stores). Only SNAP-authorized stores — nearly every supermarket, grocery and convenience store, plus chain drugstores and dollar stores (type "Other"); not restaurants or warehouse clubs that don't accept SNAP. | measured |
| Power plants, MW | EIA-860M plant inventory (period 2025-02); PrimSource = "nuclear" marks nuclear plants; Total_MW is nameplate capacity | measured |
| Electric substations | HIFLD electric substations (January 2025 copy hosted on ArcGIS Online; includes taps) | measured |
| Fuel terminals | EIA-815 terminal list (June 2023; copy hosted on ArcGIS Online) | measured |
| Hospitals | HIFLD hospitals (federal ArcGIS layer; open facilities only; bed counts last validated around 2016, missing beds shown as unknown) | measured |
| Nursing homes | CMS Nursing Home Provider Information (certified beds, average daily residents) | measured |
| Fire & EMS, police stations | USGS National Structures Dataset | measured |
| Dams | USACE National Inventory of Dams (purpose, hazard potential, height, storage) | measured |
| Cell sites | FCC licence locations for the original 850 MHz cellular band (federal ArcGIS layer). About 24,000 sites nationally, a subset of the ~140,000 towers, so treat counts as a floor. | measured |
| Pharmacies, fuel stations, water & sewage works | OpenStreetMap, from per-state snapshots in data/osm/ built by tools/build_osm.py (the snapshot date appears in the log). The public Overpass server refuses pages opened from a file, so live queries are only a fallback. Community-mapped: good for fuel stations, thin for pharmacies in many towns — the panel also counts chain drugstores from the SNAP list. Treat OSM counts as a floor. | measured (OSM) |
| Base map | Esri light-gray canvas (default), Esri streets, USGS National Map topo and imagery — switch in the map's lower-left corner |
All constants live in one block, GD.EST.K at the top of estimators.js. Edit them in any text editor and reload the page.
| Estimate | Formula | Constants (defaults) |
|---|---|---|
| Daily calorie need | population × kcal per person per day | 2,500 kcal/person/day |
| Stores here normally feed | Σ over stores of people each store type normally feeds. Compared with the residents it shows whether an area is a shopping hub (>1×) or shops elsewhere (<1×). | Super Store 8,500; Supermarket 5,000; Grocery Store 1,100; Other (drugstores, dollar stores) 280; Specialty 150; Convenience 180; Farmers and Markets 35; Restaurant Meals Program 0. Calibrated so the 2026 national SNAP store list feeds ≈ 335 million people. |
| Food on retail shelves | Σ over stores of (people fed × days of stock) × 2,500 kcal. "Days for residents only" divides by the people who live in the area, so a hub shows many days and a bedroom community few — the real stock is shared with everyone who shops there. | Days of stock: Super Store / Supermarket 7–14; Grocery Store 5–10; Other and Specialty 3–7; Convenience 1–3; Farmers and Markets 1–2. National average ≈ 6–12 days. Categories are the SNAP layer's own Store_Type values. |
| Food in local pipeline | daily need × pipeline days | 3–21 days (distribution centres, trucks in transit) |
| Cropland calorie potential (county tab, only if the NASS pull ran) | cropland acres × kcal per acre per year | 6.0–16.7 million kcal/acre/yr (upper bound = 180 bu/ac corn × 92,700 kcal/bu; lower reflects hay/pasture-heavy mixes). Grain calories: corn 92,700, soybeans 121,400, wheat 92,500 kcal/bushel. |
| Fuel in station tanks | stations × storage per station × fill fraction | 20,000–40,000 gal per station; 50 % full. 42 gal per barrel. |
| Capacity per person | Σ nameplate MW inside the area ÷ population | — (location of plants is not the same as supply to the area; the grid moves power hundreds of miles) |
| Beds per 1,000 | Σ beds ÷ population × 1,000 | only when the hospital layer carries a bed field |
These use a national snapshot built by tools/build_national.py: every county outline from Census TIGERweb and every point of the federal layers above plus the OpenStreetMap state snapshots, each assigned to the county whose boundary contains it. The per-county counts go through exactly the same formulas as the local tabs, so a county opened from the heat map shows the same numbers the map was shaded by (apart from any change in the live layers since the snapshot date). Heat-map colours are quantiles: each of the seven shades holds about one seventh of the counties, so the map shows rank, and the legend gives the value range of each shade. Counties with no population figure are grey. The "show every … nationwide" option draws the same snapshot as clustered dots; the food-retailer layer alone is about 250,000 points and can take a while on slower computers or phones.
Everything comes from public services that can change or go down; each layer fails independently and the panel says so. ArcGIS layers are queried by the bounding box of the area then filtered to the polygons in the browser, so very large counties can take a while. OpenStreetMap layers are a snapshot, not live; rebuild with tools/build_osm.py to refresh them. The SNAP list is a list of stores, not of inventory; the inventory-days constants are the single biggest assumption in the estimates and are labelled as such.
Nothing is stored or sent anywhere except the queries listed above (geocoding.geo.census.gov, tigerweb.geo.census.gov, arcgis.com feature services, geocode.arcgis.com and nominatim.openstreetmap.org only if the Census geocoder cannot match an address, the Overpass servers only if a state snapshot is missing, server.arcgisonline.com and basemap.nationalmap.gov map tiles). There is no analytics and no account.