Public records in. Honest map out.
This page explains how Culvert is built: where the data comes from, how the risk score works, how confidence is labeled, and what the map deliberately does not claim. If a statement isn't backed by a public record, it doesn't ship.
The standard
- Every layer carries its source. Federal, state, and municipal endpoints are credited on the map.
- Measured is labeled measured. Modeled is labeled modeled. Estimated geometry and inferred attributes are flagged, not passed off as records.
- Gaps stay visible. When the record is sparse, the map says so. We don't paper over holes with synthetic dots, and cities whose data didn't hold up are removed rather than faked.
Data sources
Everything on the map is drawn from public records: federal baselines like EPA SDWIS, state registries including NHDES, MassDEP, VT ANR/DEC, Maine MGS, USGS NWIS, and the RIGIS and VCGI E-911 datasets, and municipal utility GIS where cities publish it. Eleven public endpoints feed the current pilot, pulled by an open set of fetch scripts.
The proof-of-method city is Manchester, NH, mapped from real utility records: 12,028 pipe segments. Other cities are built from public records at varying depth, and each one is labeled with the confidence its records support.
The risk score
Each water-main segment carries a score from 0 to 100, banded into five levels:
| Level | Range | Meaning |
|---|---|---|
| Low | 0-20 | No factor in the record points to elevated failure risk. |
| Moderate | 21-40 | Some risk factors present; attention worth noting. |
| Elevated | 41-60 | Risk factors accumulating; worth watching. |
| High | 61-80 | Multiple factors compound; overdue for a closer look. |
| Critical | 81-100 | The strongest combination of risk factors in the record. |
Every segment is scored from the same public-data factors: age (weighted heaviest), material, and diameter, plus soil corrosivity, a break-rate term, and climate stress. Where a city publishes utility records, those factors use recorded values; where the record is silent, age is estimated from installation era and neighborhood, and the break term uses a regional baseline rather than counted events. Manchester, mapped from utility records, is the one city whose break contribution comes from its own recorded break events. Soil corrosivity is a declared factor that currently sits at the baseline, contributing zero across the New England pilot, because every mapped segment falls in the same soil-corrosivity class in the public record. Click any pipe on the map and its popup shows the exact contribution of each factor, and the contributions always sum to the score.
The model, in full. Two tiers. For the hydrant-grounded modeled cities (Boston, Burlington, Portland, Providence) the score is exactly: age score = min(100, age × 0.6); material score = 80 for cast iron, 40 for ductile iron, 10 for HDPE, 50 otherwise; diameter score = max(0, 60 − diameter × 3); final = age × 0.625 + material × 0.25 + diameter × 0.125, clamped to 10-95. No randomness, no hidden terms: recompute any pipe from its published attributes and you get its published score. The utility-GIS cities (Manchester and the rest) use a broader public-data model that also carries soil, a break-rate term, and climate stress, and estimates age from installation era where the record is silent; only Manchester's break term is grounded in recorded break events. In every city the per-factor contributions are published on each pipe and sum to its score.
Install-year estimates are floored at each city's documented water-system founding (Burlington: 1867, the year its municipal water works began operating). In July 2026 we also removed the remains of an earlier national prototype: 85 modeled-demo cities whose data didn't meet this standard. The pilot is the 23 cities listed on the map, nothing more.
What the score is: a transparent way to rank where attention is overdue, using the same factors published water-main failure research points to, with age weighted heaviest.
What the score is not: a prediction of which specific pipe breaks next. Nobody can responsibly claim that from public records, and we don't.
Confidence labeling
Every pipe carries a geometry-confidence label, shown in the map legend:
| Label | Backed by |
|---|---|
| High | Utility records |
| Moderate | Public records |
| Estimated | Inferred from public data (for example, street networks anchored to hydrant locations) |
| Unknown | Approximate placement only |
City pages also carry a coverage percentage, and it's real: it reflects how much of the system the public record actually documents, not how much we wish it did.
What Culvert is not
- It does not test your water or tell you whether it's safe to drink. Only a certified lab can do that.
- It does not detect PFAS in your glass. It shows where public records document contamination, testing, and proximity.
- It does not predict failures or give failure dates.
- It does not replace the engineers who run these systems. It makes conditions, records, and risk patterns legible so residents, reporters, and officials know where to look closer and what to ask.
Coverage today
More than 124,000 water-main segments across 23 cities in six New England states, plus statewide layers for wells, PFAS records, hydrants, and lead service line inventories where states publish them. Figures as of July 2026. Six New England states are the prototype; the same pipeline expands to the other 44.
Open the mapQuestions, corrections, records
If you're a reporter, an official, or a resident with a correction or a public record we should ingest, write to hello@littlewild.farm. Corrections are taken seriously and shipped.