How NitrateSmart is built

The audit trail: where the records come from, how they are made comparable, where the picture is incomplete, what has been corrected, and what is kept private. Every displayed result should be something you can interpret and challenge from this page.

A note on authorship: the prose on these pages is drafted with the help of a large language model — the figures are not. Every number is computed by a deterministic pipeline from the utilities’ own compliance reporting, and the model never touches the data. Where the two meet, the pipeline wins.

On this page

Publisher
NitrateSmart
Scope
Community water systems, lower 48 states
Data build
9 August 2026
Result window
9 August 2023 – 9 August 2026 (rolling three years: each system’s most recent qualifying result within this one window)
Universe
38,762 classified of a 47,581-system active backbone
Methodology
v2 classification (adopted 2026-07-19); see change log below

What this page is for: checking us, not trusting us

NitrateSmart is a standing interpretation of public compliance data — the nitrate testing that community water systems already report to state and federal drinking-water programs, normalized to one unit, classified on a five-band scale, and sourced. It is not a laboratory, not a regulator, and not a substitute for testing your own tap. Private wells are outside its scope: they are federally unregulated, not routinely tested, and not reported in a national source.

Every figure traces to the utilities’ own compliance reporting

Four source families, each row carrying its system ID, source location, and retrieval date: EPA system records (SDWIS and UCMR 5); per-system sample results from state drinking-water portals; results transcribed from utility Consumer Confidence Reports where no machine-readable state data exist; and annual averages for the one state (Minnesota) that publishes only aggregated results. Original source files are preserved and never edited.

The universe, and “no current data”

The active backbone is 47,581 community water systems in the lower 48. 38,762 have a nitrate result inside the current three-year window and are classified. 65 more, whose latest result predates the window, are shown with that older result and its date — for reference, counted in no statistic. The remaining 8,754 have no usable in-window result and are shown as no current data — never estimated, filled, or counted as zero. So the classified figures describe the systems that reported, not the whole country.

The window, and what sets the band

Each system's band is set by its most recent qualifying result within the rolling three-year window — the current reading, not the highest it has ever hit — taking the high end where that result spans a range. The full three-year range is shown alongside, so a past spike or a low season is never hidden. The aim is to be current and comprehensive, not to pick a worst case.

Four things are standardized, and each one can change a number

Four corrections make systems comparable, in a fixed order — the categories and the reasons are public; the implementation that applies them at national scale is not.

  • One unit. Every result is converted to milligrams per liter as nitrogen. Values reported as nitrate ion are divided by 4.429. A federal-limit result is 10 mg/L as N, about 44 as nitrate ion; mixing the two is the most common error in nitrate data, and it can move a small number of systems across a band boundary near the edges.
  • The same analyte. Nitrate (EPA code 1040) is preferred; nitrate-plus-nitrite (1038) is a documented fallback, used only when it is not a mixed measurement.
  • Non-detects as evidence, not zero. A reported non-detect with a numeric limit uses one-half of that limit (about 14,278 systems) — the standard convention for censored environmental data — and the result is still labelled a non-detect.
  • One declared assumption. A system reporting a non-detect with no numeric limit is assigned 0.05 mg/L so it can be ranked (758 systems). That is an operator assumption, not a measurement, and it is distinct from the 2,042 systems that land at 0.05 because half their reported limit happens to equal it. Below 0.05, any analysis would be reading the method, not the water.

Where a system buys its water and reports no sampling of its own, it inherits its supplier's result and is labelled inherited (5,281 systems, 13.6%); the other 33,481 (86.4%) carry direct compliance samples.

Why a displayed result may differ from a utility's report

Usually the units (nitrate ion vs. nitrate as nitrogen, about 4.4× apart), the analyte (nitrate vs. nitrate-plus-nitrite), the window (most recent result plus three-year range vs. an annual average), or how a non-detect is represented. If a number looks off against a Consumer Confidence Report, the units and the analyte are the first things to check.

What confidence and the range profile mean

Every classification travels with two qualifiers. Confidence describes how much testing stands behind it — how many results and how direct — not how good or bad the water is. A range profile says whether the most recent reading sits entirely within its band (uniform), dips below the band floor (variable, often a blending signal), or has no reported low end (unknown). More than a third of systems carry a low-confidence record, and what each result rests on is shown alongside it rather than hidden.

142 systems are in the federal inventory and not on this site

Checking this count against SDWIS will produce a larger number. The difference is 142 systems, removed for four reasons:

  • Duplicate PWSIDs. A system listed under two identifiers after a merger or boundary change is carried once, under the identifier its most recent sample was reported against.
  • Inactive within the window. A system that went inactive during the three years from August 9, 2023 to August 9, 2026 has results but no population to attribute them to.
  • Wholesale-only systems. A system that sells to other utilities and serves no retail population would double-count people already counted under the systems it supplies.
  • No population reported. Carried in the source with no served population, so no figure here can be built from them.

The figure is counted rather than estimated, and each refresh records it in the change log below.

Where the picture is incomplete

Coverage is uneven, and the uneven parts are named rather than coloured in. Arkansas has no nitrate data in the inventory; New Hampshire and Rhode Island have too few reporting systems to characterise; Michigan is a known collection gap (its public layer does not publish nitrate in a joinable form); private wells are out of scope everywhere. Coverage is expected to grow as more state and utility sources are gathered.

Corrections — the change log

Corrections are not hidden. NitrateSmart is maintained and falsifiable, and the record of what changes after launch — to a record or to a method — is part of the audit trail.

  • at launchThis log begins when NitrateSmart is published. Every correction made afterward will be listed here, dated, so the interpretation stays accountable as the data and methods are refreshed.

Every figure here can be checked without the machinery behind it

Everything above is what a person needs to check a figure, or to show that one is wrong. The acquisition, the state-specific parsers, the reconciliation and the refresh operations are not published — that is how the dataset is kept current nationwide, not how any single number is justified.
The reasoning behind the boundaries — why 1, 3, 5, and 10 — is on What your number means, where each is cited in full.
NitrateSmart

Nitrate in community water systems, from the utilities’ own compliance reporting.

Results reaching back to July 2023 · current through August 9, 2026.

Data & methods

How this is built
Coverage and gaps

About

About
Privacy
Health data privacy
Contact