Scientists Shared 7 Years of Water Data — Flaws and All

A team of researchers has released something surprisingly rare: a long, honest record of how much water 24 municipal buildings actually used. Tethys covers 91 months — about seven and a half years — with a reading every hour, plus the software pipeline that turns messy raw meter data into something researchers and city engineers can trust.

What makes it unusual is the honesty. Most published water datasets arrive only after an undocumented cleaning process, which hides the defects of the system they came from. This team did the opposite. They report that raw data was available only 59.4% of the time, that four network-wide outages knocked out the entire estate at once for a combined 595 days, and that the aggregation step producing the released files silently created 205,200 impossible decreases in a cumulative index — like a car odometer rolling backwards. Fixing that bug takes one line of code.

Instead of quietly patching things, they published per-hour provenance: every hour is labeled either as resting on a real measurement (75.9%) or as unknown (24.1%). Because these meters only count upward, the readings on either side of a short gap actually let you calculate how much water passed through during it. So gaps aren't always lost information.

Why should you care? Software that predicts water demand and detects hidden leaks is only as good as the data behind it. When datasets are scarce, short, or pre-scrubbed, the tools built on them fail quietly. Honest, well-labeled data lets researchers build leak detection that genuinely works — and leaks caught earlier mean lower water bills, less wasted water, and more resilience when droughts hit.