The index is built from filings utilities submit to their own state regulator. Nothing here is a survey, an estimate of an estimate, or a proprietary adjustment. Every figure is either as-filed or computed from as-filed values, and every page says which.
Sources
Who files and when is on each state’s own page. What follows is what changes how the index may be read.
Texas. Water loss audits filed with the Texas Water Development Board. Who files, and when. TWDB publishes no per-audit validation status, so this site displays none. We checked all five source files; the only validation-adjacent column is a waiver implementation status covering 42 systems. We have asked TWDB whether the per-audit status is available and will publish it if it is.
Massachusetts. Unaccounted-for water and residential gallons per capita per day, published by MassDEP from Annual Statistical Reports, required of municipal suppliers averaging 100,000 gallons a day or more. The performance standards.
California. Validated water loss audits filed with the Department of Water Resources. Who files, and when. DWR publishes six columns naming individuals, so what this site holds and cites is a redacted extract and the column letters in our click paths are the extract’s. The grain is the system-year and not the supplier-year: one investor-owned utility files eighteen rows for a single year, one per system it operates. Keying on the organization would collapse them, and counting rows without collapsing would weight that filer eighteenfold.
Georgia. Validated water loss audits filed with the Environmental Protection Division. Who files, and when. EPD publishes one workbook per audit year and the headers do not agree with each other: column counts, volume-column spellings and even the sheet name change across the series, so columns are resolved per file and a missing one stops the build instead of silently nulling a field. From 2021 the headers stopped declaring units, settled from EPD’s own arithmetic and from continuity across the break. Variable production cost is declared in no year and is not published here at all.
Wisconsin. Schedule W-15 of the annual report every regulated water utility files with the Public Service Commission under Wis. Admin. Code ch. PSC 185. The PSC calls W-15 the input to an AWWA M36 audit, and it is not an audit: no validity score, no leakage index, no meter accuracy, no validation status, and no price of any kind, so no Wisconsin figure is valued here. The export carries street addresses in free text, so what this site holds is a redacted extract with those columns dropped.
Indiana. Water loss audits compiled by the Indiana Finance Authority. Who files, and when. The audit year appears nowhere in the bytes: no column, no header, no cell. Each file is named for the report and not the year it covers, so the period comes from a mapping this site holds against the sentence each report states it in, and an archived file outside that mapping stops the run instead of being dated by its name.
Presentation scale is detected per file and converted on the way in, so every volume in the index is million gallons. A file reporting in thousands and read as millions puts every figure derived from it out by three orders of magnitude, and none of them looks wrong.
Why the water balance is not a check
A water audit is an identity:
system input = billed authorized + unbilled authorized + apparent loss + real loss
This index used to report a pass or fail on it. It no longer does, because in none of the sources we hold can the identity fail, and publishing it as a finding was wrong.
TWDB’s audit worksheet computes water losses as system input minus total authorized consumption, then splits that into apparent and real. Both relationships hold on all 9,744 Texas filings on record, so the identity reduces to input = input. The residual was exactly zero on 9,743 of them, and the exception is a one-digit transcription error in TWDB’s file rather than a disclosure failure.
MassDEP collects no component balance at all, so there is nothing there to test either.
Every filing is therefore inapplicable, because nothing was withheld and nothing is checkable, or insufficient, where it does not publish enough components to decompose system input. Neither is a pass. No filing here carries closed or failed; both remain available for a future source that publishes independent components. Each state page states its own counts.
The arithmetic is still computed, because a non-zero residual means the components the regulator published contradict each other. That is a defect in the export, and the page says so.
What is checkable: the defaulted share
Apparent loss has three components: customer meter inaccuracy, unauthorized consumption, and data handling error. Where a utility does not measure the last two, TWDB supplies a default value and flags it in its own export. Unauthorized consumption is a default on 95.2 percent of filings and data handling error on 30.4 percent.
So a reported loss figure can be substantially an assumption, and each page states the share: the volume from defaulted components against the apparent loss the filing reports.
14.4 percentThe median filing’s defaulted share. 1,329 filings are more than half. Seven exceed 100 percent, stating a negative customer meter accuracy loss, which pulls the reported total below the sum of its own defaulted parts.
TWDB’s own data-validity score does not separate the two: 68.4 out of 100 on filings using at least one default against 70.0 on those using none. The defaulted share is a separate fact from the regulator’s quality metric: it is the one per-filing value in this data that can be shown false, which is why it replaced the balance verdict.
Where a filed figure contradicts the audit’s own arithmetic, we publish the filed value, say why it cannot be true, and exclude it from anything computed. A leakage index far outside the unavoidable minimum the same audit computes, a reported loss of exactly zero, and a unit price a thousand times a plausible one all qualify. An impossible number and an absent number are different findings.
Do not read any verdict in the bulk data as a pass or a failure.
The recoverable figure
Apparent loss is valued at the retail rate. It is water that reached a customer and was never invoiced, so it is revenue, recoverable inside a budget year, with no capital work. Real loss is valued at variable production cost, because stopping a leak saves treatment and pumping, not retail price.
Where a filing carries no retail rate, a state median is substituted and the figure is marked modeled everywhere it appears, including in the CSV. That median comes from the rates utilities filed themselves, is published with the count behind it, and is recomputed on every build: for Texas, $4.75 per thousand gallons across 9,493 filed rates, 2015 to 2024. Where a state publishes no rates, as Massachusetts and Wisconsin do not, there is no median and no modeled figure, and the value is unavailable rather than estimated. Where a filing gives total loss without splitting apparent from real, the whole volume is valued at variable cost and labeled a deliberate floor.
Every dollar figure is a loss volume valued at a unit price, and it is only as good as that price. Retail prices are read between $1.00 and $40.00 per thousand gallons, and variable production costs between $0.10 and $25.00. Outside those, TWDB’s own cost of losses is omitted instead of repeated, because that column is the price multiplied by the volume and the identity holds exactly on every filing carrying both.
Published loss dollars are rounded to three significant figures. A fifth of filed variable production costs are round numbers someone chose instead of computing: 298 filings at exactly $1.00, 222 at $2.00, 214 at $0.50. A figure derived from one of those does not support eight digits.
A recoverable figure is an order of magnitude, not a collections target. It says whether the problem is worth an afternoon or a quarter. Which accounts, which zones and which meters needs your billing history, your meter reads and your production data reconciled against each other, and no public filing contains that.
What we do not do
We do not rank by loss percentage. A small percentage on a large system is worth more than a large percentage on a small one, and percentage rankings punish small rural systems for arithmetic.
We do not recompute a utility’s filed figures to make them close, and we do not publish a check that cannot fail. When we found the balance-closure verdict was a tautology on every filing we held, we retired it and said so here.
We do not publish a figure we cannot source. An absent value renders as an em dash, never as a zero.
Corrections
If a number here is wrong, it is wrong in the filing or wrong in our reading of it. Either is worth an email to ammar@municorn.us, with the system and the year. We correct the page and note what changed and when.