THE NAME IS A JOKE. THE RECEIPTS AREN'T. RESEARCH CUTOFF ·

METHODS

The rules every number has to pass.

This page is the receipt for the receipts: how evidence gets labeled, what the water words mean, where the boundaries go, how dates work, and how corrections are handled. Research cutoff , latest verification .

01 / EVIDENCE CLASSES

Six labels. Every figure wears one.

The label answers the first question a careful reader asks: did somebody measure this, or did somebody calculate it?

Reported · primary
A government filing, permit, utility record, audited disclosure, or first-party statement. The number is what the source says, in the source's own boundary and unit.
Reported · secondary
A reputable secondary source used only when the original artifact is unavailable. Flagged so readers know the chain of custody has one extra link.
Modeled
Calculated from explicit assumptions and a versioned method, by us or by the cited author. Never presented as a meter reading.
Inferred
Deduced from multiple signals. The reasoning and a confidence level are shown alongside the value.
Definition
A term as defined by a standards body or agency such as USGS or DOE, applied to reported figures.
Analysis
Our framing of what several sourced figures do and do not establish together. Contains no new numbers.

02 / DEFINITIONS

Water words that are not interchangeable.

Definitions follow USGS and DOE usage. A disclosure that says only “water use” is analytically incomplete until it says which of these it means.

Withdrawal
Water removed from surface water, groundwater, or a utility system for use. Some returns to the source or sewer. Measures demand placed on infrastructure and source capacity.
Consumption
The portion of withdrawal not readily returned locally because it evaporated, was incorporated into a product, or was otherwise removed from the immediate water environment. Consumption equals withdrawal minus discharge.
Discharge (return flow)
Water sent back to a river, aquifer, or wastewater system after use. Quantity, timing, temperature, and quality decide whether it offsets local impact.
Replenishment
A separately estimated benefit from watershed projects intended to restore or improve water availability. It is not a reduction of a facility's peak withdrawal and may occur elsewhere or later.
WUE (water usage effectiveness)
Liters of site water per kilowatt-hour of IT energy. An efficiency ratio, not a total. Incomplete unless the numerator is identified as withdrawal or consumption.
Direct (on-site) water
Cooling-tower evaporation and blowdown, adiabatic cooling, humidification, domestic uses, and landscaping at the data-center site.
Indirect (electricity-related) water
Water consumed by the power plants and primary-energy supply serving the data center. Varies with grid mix, generator cooling design, and hydropower accounting.
Embodied water
Water used in semiconductor fabrication, equipment manufacturing, construction materials, dust control, concrete curing, and site dewatering. Often in a different watershed and year.

03 / BOUNDARIES

Every figure states its boundary.

Before a number appears on this site it has to answer: who said it, in which source, on what date; what facility, geography, period, and system boundary it covers; whether it is measured, proposed, modeled, inferred, or user-submitted; which metric it is (withdrawal, consumption, discharge, replenishment, electricity, or capacity); how confident the entity match is; and whether it has been corrected.

A total water footprint is direct site consumption, plus electricity-related consumption, plus primary-energy consumption, plus allocated hardware and manufacturing water, plus allocated construction water. Each term carries its own location, period, source type, method, and uncertainty. When those inputs are unknown, the honest output is a range with a confidence grade, not a crisp decimal.

What we never do.

  • Collapse withdrawal, consumption, discharge, and replenishment into undefined “water use.”
  • Present proposed demand as actual use, or nameplate capacity as operating load.
  • Present an estimate as a meter reading.
  • Add values with incompatible years, regions, or lifecycle boundaries.
  • Present a universal water-per-prompt number.
  • Infer ChatGPT, Claude, or Gemini water from provider-wide fleet totals.
  • Rank companies using incomparable disclosures.
  • Claim closed-loop or reclaimed-water cooling means zero footprint.
  • Claim that a small national share makes local risk irrelevant.
  • Publish precise facility locations or infrastructure details that create security risk.
  • Let an advertiser or sponsor preview, select, shape, or suppress evidence, conclusions, rankings, corrections, or coverage.

Dates and verification.

Four dates are kept distinct: the publication date of the source, the reporting period it covers, the event date where relevant, and the date we last verified the figure against the source. The research cutoff is the date after which new sources have not yet been reviewed. Each answer shows its own verified date; the sitemap, feed, and structured data carry the same dates, so a stale figure is stale everywhere at once, visibly.

Corrections policy and log.

Corrections are free to request and free to read. Email thirsttrapai@pm.me with the page, the figure, and the better source. When a figure changes, the new value replaces the old one on the page, the verified date is bumped, and a dated entry is added below. History is not rewritten silently and corrections are never paywalled.

No corrections logged yet. Launch content was verified on . The first correction will appear here with its date, the page, and what changed.

Source hierarchy.

Preferred in this order: government and regulatory records (USGS, DOE, LBNL, GAO, state audits such as Virginia JLARC, utility dockets, permits); audited or assured company disclosures; peer reviewed research; company blog posts and executive statements; reputable secondary reporting. Lower-ranked sources are used when they are the only source for a claim, and labeled accordingly. Sources are linked by URL on every figure so you can skip us and read the original.

The full analytical basis for the launch content is a research paper kept in the public repository, with thirty referenced sources. It is the reason the site exists.