Skip to content

Data & methodology

Where every number comes from

Everything on WellsMap traces to a source you can check. This page is the audit trail.

The well records

The dataset is 14.7 million well construction records across 45 states, collected from the state agencies that receive drillers' completion reports — geological surveys, departments of natural resources, health departments and water boards. Each state's records were normalized into one schema (location, total depth, static water level, yield, casing, contractor, completion date), coordinates were validated and reprojected where states publish local grids, and missing counties were reverse-geocoded against Census TIGER boundaries.

Known gaps, on purpose:

  • Five stub states — North Carolina, Louisiana, New Jersey, Nevada and North Dakota don't publish usable bulk records (NC's are largely handwritten GW-1 forms we are digitizing; others sit behind manual-request or legacy-database paths). Estimates there lean on neighboring data and honest "sparse data" flags.
  • Coordinate quality varies — a few states (MA, PA, TN, NY) publish incomplete coordinates; those wells are indexed but can't power map statistics.
  • City fields are ~50% filled because most wells are rural — that's correct behavior, not missing data.
  • Contractor names are as-filed: the same company can appear under several spellings, and some states omit names entirely.
  • TX / WA / AL / VA depth remap (2026-07-14) — these states always had coordinates (map dots) but depth was empty in the first load because depth lived in secondary tables (TX borehole intervals, WA CompletedDepth, VA TD, AL Construct join). Depth was re-harvested into wells.db from those sources; estimates now return real p10–p90 bands.

The depth estimate

Not machine-learning guesswork — spatial statistics on real logs:

  1. Find wells with recorded depth near your pin, widening the radius until there's enough signal.
  2. Compute a distance-weighted median plus a p10–p90 band.
  3. Blend gently with ~5 km grid and county priors when local samples are thin.
  4. Show the sample count and radius used, so you can judge the estimate's footing yourself.

AI (WellsMap AI) is optional and only explains the computed statistics — it never invents a depth, and it never runs without an explicit click.

The permit playbooks

All 50 states were researched and cross-verified in July 2026: a deep-research pass with adversarial verification (claims survived only if independent attempts to refute them failed), plus regional sweeps that fetched statutes and live agency portals. County rows carry a tier — verified portal, verified landing, or generic floor — and every row stores its sources and verification date. Fees carry effective dates because they change (several did during the research window).

What the permit pages are: a researched map of who issues what, for how much, filed how.

What they are not: legal advice, or a guarantee a rule hasn't changed since verification. The issuing authority's word always controls.

Corrections

Wrong fee? Moved portal? A record that identifies you and shouldn't? Email

team@fieldelevate.com with the page URL — corrections re-verify

against the agency source and update the row's verification date.