Abstract
<jats:p>In summer 2026, record-low Danube water levels at Paks, Hungary, twice curtailed the Paks Nuclear Power Plant. Using 46 years of daily discharge and water-level records (1981–2026), we test whether the stage–discharge relationship shifted independently of discharge. The published discharge series proves to be a periodically revised rating-curve product, not an independent measurement – confirmed by a 2022 bulk reprocessing and the operator’s own statement that discharge derives from water level – so we reconstruct roughly 11–12 rating epochs, not thousands of daily observations. Water level at a fixed reference discharge (1300 m3/s) has declined 16–17 cm per decade (95% CI −19 to −14), confirmed at a shallower rate (−8.6 to −14.0 cm/decade) in the least circular subset of years. Three comparison stations show a spatially coherent pattern: a stabilized step at the 1992 upstream diversion, a plateaued unrelated historical decline further upstream in Austria, and two stations – including Paks – still declining. The 1980s-vintage rating curve would place the 2026 low-water stage roughly 65 cm higher at the same discharge. We find no evidence that upstream water management retained water during the 2026 drought. Hungary’s current river-basin management plan treats the plant’s thermal discharge as a pressure on the river but, on our review, never the reverse relationship demonstrated here.</jats:p>