Water loss distributions

• water-loss-distributions.csv — one row per population
• water-loss-distributions.json — the same rows, with the comparability caveat and the standards as named fields rather than prose

What is in it, and what is deliberately not

Every row is a population: a state, a county, or a band of systems of comparable size. There is no pws_id column and no individual utility appears as a row. That is the line, and it is not an oversight — a distribution is what a statistic is cited from, while a per-system file is a different artifact serving different people. The per-system figures are all published, one page at a time, under the index, each beside the archived cell it was read from. For those as files, ask on the terms page.

The caveat that must travel with these figures

“Water loss” is at least three different quantities depending on the jurisdiction. Texas, California and Georgia collect a component water balance and non-revenue water is computed from it; Massachusetts publishes unaccounted-for water as a percentage and collects no component balance at all. These rows are comparable within a state and are not comparable between states. Nothing here supports a cross-state ranking, and building one is the single most common way this data is misused.

Columns to read carefully

median_defaulted_pct is the share of reported apparent loss that the regulator supplied as a default rather than the utility measuring. Only TWDB flags this, so it is null outside Texas — which is an absence in what the other three states publish, not a finding that their filings are fully measured.

systems_over_standard_latest is null, not zero, where this corpus holds no published state threshold. A zero would read as “no system in this state is over the standard”, which is a claim about the state rather than about what we hold.

No filing here passes or fails a water balance check, and no source supports one. How these numbers are made.