How WaterScore works
How the data is sourced, how grades are assigned, what the numbers mean, and where the limits are.
What is WaterScore?
WaterScore translates federal drinking-water data into plain-language A–F grades for any address in the US. It reads the same databases regulators use, and presents them so you don't need a PWSID, a regulatory glossary, or three cross-referenced tables to understand what's in your water.
Grades are assigned per contaminant category, never combined into a single composite score for municipal systems. A system can earn an A for violations and a C for chromium-6 in the same report, and you see both.
The platform is free, requires no email, and sells nothing on the results page. That's a deliberate position, not a temporary one.
Where does WaterScore get its data?
Every grade traces to a named public source. WaterScore does not collect its own samples or generate original measurements; it structures, grades, and contextualizes data that utilities and federal agencies already publish.
Consumer Confidence Reports (CCRs)
Every community water system in the US is required to publish an annual Consumer Confidence Report listing its detected contaminants and their concentrations. These reports are the richest public record of what's actually in a given system's water, and they are scattered across thousands of utility websites, state portals, and PDFs in no consistent format.
WaterScore compiles them into a single structured national dataset, computed live from extraction records. See the current population and system counts on the live coverage page. A standardized, machine-readable national CCR dataset does not otherwise exist; building and maintaining it is the core of the project.
EPA SDWIS
The Safe Drinking Water Information System provides health violations, compliance status, and system metadata. WaterScore uses SDWIS facilities and violation-enforcement records to grade violations and to describe each system's sources and treatment.
EPA UCMR5
The fifth Unregulated Contaminant Monitoring Rule provides testing results for PFAS, chromium-6, lithium, and haloacetic acids, for systems serving more than 3,300 people (2023–2025). These are contaminants not yet (or not fully) regulated at the federal level, the kind of detections standard compliance reporting often misses.
EPA LCRR
Lead and Copper Rule Revisions service-line inventories track lead, galvanized, and unknown-material lines as submitted by utilities.
USGS machine-learning rasters
For private wells, where no utility reporting exists, WaterScore uses USGS probability models at 1 km² resolution: PFAS (Tokranov et al. 2024), arsenic (Lombard et al. 2021), and nitrate (Ransom et al. 2021).
EPA/USGS Water Quality Portal (WQP)
waterqualitydata.us aggregates measured water-quality samples and well-station metadata from EPA, USGS, and state programs. WaterScore queries it for measured contaminant concentrations and well depths near a given coordinate on private-well pages.
Texas Water Development Board (TWDB)
twdb.texas.gov/groundwater/data publishes Texas's groundwater database, including well water-quality samples and well-depth records. WaterScore uses TWDB for Texas private-well pages alongside WQP.
How does WaterScore grade water?
Grades are assigned per category against a specific, published threshold. Each contaminant or risk gets its own grade. Nothing is blended into a composite for municipal systems. Below is every grading rule on the site, one per category.
PFAS
UCMR5 testing results. The B/C boundary is set by max(PFOA, PFOS), the two compounds with finalized MCLs.
- A
- tested, no PFAS detected
- B
- any PFAS detected, max(PFOA, PFOS) <1 ng/L
- C
- max(PFOA, PFOS) 1–4 ng/L
- F
- any compound above its 4 ng/L MCL
- N/T
- system not in UCMR5
Chromium-6
UCMR5 testing results. No federal MCL exists. California's 10 µg/L notification level is the reference point.
- A
- not detected
- C
- detected, below 10 µg/L
- F
- above 10 µg/L
Lithium
UCMR5 testing results. No federal limit. WHO drinking-water reference value is 10 µg/L.
- A
- not detected
- B
- detected, below 10 µg/L
- C
- at or above 10 µg/L
Lead
Based on the utility's LCRR service-line inventory. There is no F tier because LCRR doesn't draw one. In pregnancy mode, medium becomes C and unknown becomes D.
- A
- no lead lines, or low-risk inventory
- B
- some lead lines, medium risk
- ?
- unknown materials, or no inventory submitted
- D
- high proportion of confirmed lead service lines
Violations
Severity-weighted, not count-based. A single open health violation outweighs a stack of resolved administrative paperwork.
- A
- no violations
- B
- resolved health violations only
- C
- recent enforcement (within 2 years), or 10+ historical health violations in 10 years
- D
- 1 open administrative violation
- F
- 2+ open administrative violations, or any open health-based violation regardless of count
Arsenic, nitrate, PFAS (private wells, modeled)
All three modeled-well grades share one probability ladder, applied to the USGS raster value at the well coordinates. The probability is what the model estimates for the surrounding 1 km² area, not your well specifically.
- A
- probability <20%
- B
- probability 20–40%
- C
- probability 40–60%
- D
- probability 60–80%
- F
- probability ≥80%
Nearby measured wells (private wells)
If the EPA/USGS Water Quality Portal or TWDB reports a tested well within the search radius, WaterScore adds a measured grade per contaminant alongside the modeled grade, computed from the actual reported concentration. Modeled and measured grades are shown side by side. Neither is collapsed into the other.
How accurate is the data?
Accuracy depends on how the source was published. EPA compliance data is pulled directly from the source. Well data is pulled directly from the Water Quality Portal and the Texas Water Development Board data portals.
CCR data accuracy varies. Some states publish structured data, which is easy to extract and upload to WaterScore. For states that don't publish structured data, we use LLMs to extract data from utility-published PDF reports. LLM extractions can be incorrect because of complex PDF tables. A QC script flags extreme outliers (values that exceed published MCLs by implausible ratios, suggesting unit or extraction errors), and we maintain audit reports under /audits in the repo.
WaterScore prefers structured state sources wherever a state publishes one, and treats PDF extraction as the fallback rather than the default. It's also a concrete argument for states to publish their water data in structured formats.
Every system page carries a “Does this look wrong?” control next to the data. Reader corrections are part of how the dataset is kept honest at scale.
How much of the US does WaterScore cover?
SDWIS violations are available for 48,278 water systems covering all 50 states, Puerto Rico, and tribal nations. CCR coverage is computed live and published on the coverage page, broken out by state and by population served. Currently we have CCR coverage for 74% of the US population.
How does WaterScore recommend filters?
Filter recommendations are based on third-party certification, not manufacturer marketing. WaterScore prioritizes products certified by NSF, WQA, or IAPMO for the specific contaminant in question; independent lab data (e.g. SGS) is accepted as a secondary tier, with that lower confidence stated plainly.
We recommend filters for the specific contaminant data WaterScore has for your address. Recommended filters are certified to reduce your specific contaminant levels to levels within either federal limits, WHO limits, or research-backed pregnancy thresholds. Filter research is painful. If you have any questions about the filters recommended on WaterScore, Wirecutter, Consumer Reports, or anywhere else, don't hesitate to reach out and ask marissa@know-your-water.com.
Recommendations are presented without the anxiety overlay common to the category. No “X times above health guidelines” framing. Several recommended products come from companies whose own marketing WaterScore wouldn't endorse; the products are judged on their certifications and datasheets, not their funnels.
Why does pregnancy mode use different thresholds?
Pregnancy mode applies stricter, evidence-backed thresholds for contaminants with documented associations to pregnancy outcomes. It currently covers eight contaminants, each tied to a specific peer-reviewed source with a live DOI: lead, nitrate, arsenic, manganese, PFOS/PFOA, fluoride, and lithium.
Two contaminants that appear on other “pregnancy-risk” lists (TTHM and HAA5) are deliberately excluded, because the evidence for their pregnancy-outcome effects is characterized as uncertain (Jauniaux et al. 2024, BMJ Open).
What WaterScore doesn't do
WaterScore reports what the data shows and is explicit about what it can't. The most important limits:
“Not tested” is not “tested clean.”
A contaminant absent from a report wasn't necessarily measured. WaterScore distinguishes a confirmed non-detection from a gap in the data rather than letting silence read as safety.
Lead can come from your plumbing.
A system with no lead service lines can still deliver lead through household pipes and fixtures. A clean system grade is not a clean tap guarantee.
Grades reflect the last data update.
A grade describes conditions as of the most recent ingested data and may not match a system's status today. The utility's current CCR is always the authoritative real-time source.
Modeled is not measured. Nearby is not yours.
Private-well modeled grades are probabilities from regional USGS rasters, not tests of your specific well. Nearby-well measured grades come from real concentrations at the closest tested well in EPA/USGS WQP records, useful as context but still not your water. Only a lab test of your own well tells you what's actually in it.
WaterScore is informational, not advice.
It does not constitute health, legal, or environmental advice.
Disclaimer
WaterScore is for informational purposes only and does not constitute health, legal, or environmental advice. Data is sourced from publicly available federal and state databases and is updated periodically. Grades reflect conditions at the time of the last data update and may not reflect current system status. Always consult your water utility's Consumer Confidence Report and a licensed professional for site-specific guidance.
Sources cited on this page
- Lombard, M.A., et al. (2021). Arsenic probability model. USGS. Source for the private-well arsenic raster.
- Ransom, K.M., et al. (2021). Nitrate machine-learning model. USGS. Source for the private-well nitrate raster.
- Tokranov, A.K., et al. (2024). Predicting PFAS in private wells. USGS. Source for the private-well PFAS raster.
- Ward, M.H., et al. (2018). Nitrate intake and health. International Journal of Environmental Research and Public Health. Basis for the 5 mg/L pregnancy-mode nitrate threshold (not the well grading ladder).
- Jauniaux, E., et al. (2024). BMJ Open. Characterization of uncertain TTHM/HAA5 pregnancy-outcome evidence; basis for excluding both from pregnancy mode.
- Iuliano, K., et al. (2025). CCR readability, national sample. International Journal of Hygiene and Environmental Health, 114536. DOI: 10.1016/j.ijheh.2025.114536
- EPA SDWA_REF_CODE_VALUES.csv (ECHO data downloads). Canonical SDWIS violation code to contaminant name lookup.