Data sources & vintages
- Gas stations — OpenStreetMap
amenity=fuel (Overpass snapshot, June 2026); brand/name/operator tags normalized to station formats. Brand reach is labeled separately as national/multi-state, regional, or local/unknown.
- Traffic — federal HPMS 2020 (all functional classes) ∪ WSDOT current AADT (WA state highways; preferred on overlap).
- Local catchments — Census 2024 ACS 5-year block-group population and employed residents, located at TIGER representative points.
- Site capacity — mapped OpenStreetMap fuel-site footprints for selected top-ranked OSM-way stations; missing sites remain unknown.
- State gasoline anchor — EIA SEDS (MGTCP) through 2024.
- EV chargers — AFDC station locator (Esri Living Atlas live mirror, June 2026).
- Supply infrastructure — EPA FRS state facility files, cross-checked with
EIA-820 refineries, linked state environmental programs, targeted OpenStreetMap
terminal/tank/rail/marine evidence, and the local HPMS/WSDOT highway screen.
Exact NPMS pipeline geometry is not reproduced.
- Underground tank corroboration (WA/TX pilot) — Washington Ecology and Texas TCEQ
facility/tank registries. Records are classified as retail, probable retail, ambiguous,
or non-retail before matching. Registered capacity is storage evidence, not sales or throughput.
- Vehicle registrations — CA DMV fuel-type-by-ZIP (2018–2024) · WA DOL by county (2017–2024).
Demand model (v1)
Top-down disaggregation: each state-year's SEDS gasoline total is split across its counties by a
fuel-intensity-weighted fleet (gasoline 1.0 · hybrid 0.5 · plug-in hybrid 0.3 · diesel/BEV 0).
County sums match the state anchor exactly.
Station volume
Screening estimate — brand-derived site format and road AADT distribute each state’s aggregate
demand; block-group population/employed-resident catchments, public road-access fields, and available
mapped site footprints add local context before the result is re-raked to the state anchor.
Warehouse clubs, grocery fuel centers, travel centers, larger convenience formats, branded stations, and
independents use separate public-data priors. The displayed range spans alternate public model
allocations and is widened when catchment, access, or capacity inputs are sparse. It is a sensitivity range, not measured
throughput or a statistical confidence interval.
Inventory review flags highlight possible closures, duplicates, brand conflicts, weak road
matches, source freshness, and tank-registry status or classification conflicts. They are prompts
for verification—not automatic corrections.
The current station model estimate uses a stabilized consensus of five independently state-raked
public-data structures. Its sensitivity range and agreement evidence describe how consistently those
structures allocate demand; they are not measured throughput or a confidence interval.
Validation & confidence
EIA state and prime-supplier series can calibrate aggregate fuel demand, but public data does not
provide dependable station-level throughput for direct backtesting. Treat rankings as comparative
screening evidence. Traffic-grounding confidence is shown per station; Texas forecast trends are
lower confidence because they begin with a short observed registration window.
Demand forecast (2025–2035)
Per-county EV-share logistic S-curve anchored on the observed 2024 share and 2018–24 trend; every
county trends toward the same ceiling C = 0.85 (EV-friendly metros sooner, rural later — never
frozen). The county path now also applies Census population/employment trends, FHWA VMT per capita,
EPA remaining-fleet efficiency assumptions, Census commuting/seasonal-housing change, and BTS border
traffic. Commuting, tourism, and border factors are normalized within each state so they refine county
allocation without counting statewide travel twice. Higher/Base/Lower vary activity, efficiency,
and EV adoption together. Every 2035 driver contribution, source, vintage, evidence grade, and missing-data
fallback is shown in the county and target details. Projected charger
counts grow proportionally to each county's EV fleet (chargers-per-EV held at the 2024 ratio); locations
are unknown, so they are never drawn on the map.
Station score (Buy lens)
Weighted blend (default Vol 40 / Disrupt 30 / Comp 30) of three 0–100 components:
Vol volume percentile within state · Comp lower distance-decayed competitor capacity within
radius R, using speed/access evidence as a drive-time proxy · Disr county demand retention through
2035. AADT is already inside modeled volume and is not scored a second time. Each station also shows
a score range across gallon/competition uncertainty and all three demand scenarios. Scores are
state-relative; equal scores in different states do not imply equal absolute economics.
The road model is not
turn-by-turn routing or observed parcel ingress.
Network opportunities
Network opportunities (hollow teal diamonds — never real stations): points every 0.75 mi
along corridors carrying ≥ 15k AADT, one per ¾-mi cell. Components: Traf corridor AADT percentile ·
Under less existing station volume within radius R = higher · Disr as above ·
Grow county population CAGR percentile (Census PEP 2020–2024) · Feas public land-use,
access, interchange, and cannibalization screen. Default weights Traf 25 / Under 25 / Disr 15 /
Grow 10 / Feas 25. Each result is classified as an undersupplied market, highway/corridor
gap, interchange opportunity, competitor white space, or evidence-supported site candidate,
with the reason shown in its result and detail view. Site-candidate status additionally requires
suitable adjacent public land use, parcel/zoning or a usable nearby site outline, and full or partial
routed public-road evidence without a same-side restriction; it is still not parcel
ownership, driveway approval, or construction approval.
Classification order:
the site-evidence gate takes precedence; interchange opportunities require Traf ≥60 plus a
≤5-minute routed ramp or a ≤1-mile ramp with interchange score ≥75; competitor white space
requires Under ≥75 plus sparse/distant competition; undersupplied markets require Under ≥65
plus Growth or mid-case retention ≥60; remaining points are highway/corridor gaps.
The displayed opportunity score then adapts the five visible base weights to that type:
interchange fit emphasizes traffic/access, white-space fit underserved/competition,
market-gap fit underserved/growth/retention, corridor fit traffic, and site-evidence fit
feasibility. The detail drawer shows the effective emphasis and an unchanged-weight baseline.
Supply infrastructure
NAICS 424710 and SIC 5171 cover both true product terminals and smaller bulk
facilities, so they never establish rack status alone. The map defaults to confirmed,
probable, and EIA-supported refinery records. Unconfirmed storage is opt-in. Rail,
marine, mapped tanks, state-program links, source date, highway proximity, and the
evidence grade are shown per facility. Highway travel time is a public-road screening
estimate, not observed truck routing. Public pipeline points appear only as context;
exact NPMS geometry is restricted to the NPMS viewer/request process.