The United States Geological Survey publishes an automatic alert for every significant earthquake on Earth. It is called PAGER, it goes out within about half an hour, and it estimates how many people were exposed to strong shaking and how many are likely to have died. Near the bottom of the one-page version is a short table headed “Historical Earthquakes”: up to three previous earthquakes near the same spot, each with its date, its distance, its magnitude and its death toll. It is the part of the alert a reader without a seismology degree can actually weigh.
Every one of those comparisons is at least eighteen years old. Across 730 earthquakes that carried a PAGER alert between 10 January 2018 and 9 August 2026, the product files contain 1,712 historical comparisons. Sorting them by date, the most recent is a magnitude 6.6 off Gisborne, New Zealand, on 20 December 2007. There is nothing after it — not one row, in eight and a half years of alerts.
The reason is a file rather than a policy. PAGER reads its comparisons from a spreadsheet called expocat.xlsx that sits in USGS’s own public source repository for the software. It holds 5,646 earthquakes, the first on 29 February 1960 and the last on 26 December 2007, with columns for magnitude, depth, shaking deaths, total deaths and the population exposed at each level of intensity. That the live product is reading that exact file is checkable: the row PAGER printed for a Chiapas alert in July 2026 — event 199109180948, magnitude 6.2, region name “UK”, country code GT — is byte-for-byte the row in the spreadsheet.
The selection rule is in the same repository. PAGER takes every catalogue entry within 400 kilometres of the new epicentre and ranks what it finds by shaking deaths, then by maximum intensity, then by how many people that intensity reached. Magnitude is never used. So the table is not a list of the biggest nearby earthquakes, or the most recent ones; it is a list of the deadliest ones the spreadsheet knows about.
That matters most where the record has moved. When a magnitude 6.8 struck the Kumamoto region of Japan on 28 July this year and PAGER issued a red alert, its three comparisons were earthquakes from 1997, 2001 and 2005, between 131 and 231 kilometres away. The magnitude 7.0 that killed dozens in the same prefecture in 2016 struck 12.5 kilometres from this year’s epicentre — comfortably inside PAGER’s own search circle. It is not that the software weighed it and preferred something else. It cannot see it.
The same gap shows up wherever a modern disaster is the obvious reference. The 2010 Haiti earthquake sits 96 kilometres from the epicentre of the magnitude 7.2 that hit Nippes in 2021, and does not appear in that alert’s comparisons. The magnitude 8.2 at Tehuantepec in 2017 is 116 kilometres from a magnitude 7.3 off Mexico in July this year. The 2020 Zagreb earthquake is 58 kilometres from Petrinja later the same year. All four are inside the radius; all four are in USGS’s own ShakeMap Atlas; none is in the spreadsheet PAGER reads.
Measured across the whole sample, the missing material is not marginal. In 579 of the 580 alerts that printed any comparison at all, at least one magnitude 6.0 or larger earthquake since 2008 had occurred inside PAGER’s 400-kilometre circle — 22,250 such earthquakes in total, a median of 31 per alert. Of the 728 alerts whose product files were readable, 562 printed the full three comparisons and 148 printed none.
None of this is concealed, exactly. PAGER’s reference page credits EXPO-CAT, a catalogue described in a 2009 paper by Allen, Marano, Earle and Wald whose published span is 1973 to 2007. But that page currently returns a server error and had to be recovered from the Internet Archive, and the printed alert itself carries no caption, no catalogue name and no date range above the table. A reader looking at the one-page PDF has nothing to tell them the list has a ceiling.
The catalogue’s age shows in smaller ways too. Of the 1,712 comparison rows, 1,389 — 81 per cent — carry the region name “UK”, which is not the United Kingdom: the 1991 earthquake PAGER labels that way carries the country code GT, for Guatemala. It is an unfilled placeholder that survives into published text, so alerts have reported earthquakes as having “struck UK”. And 539 of the rows, just under a third, carry no total death toll at all.
The newer data exists. USGS’s ShakeMap Atlas, which the agency describes as the calibration set for exactly this kind of loss estimation, holds 14,865 earthquakes, 3,003 of them from 2008 onward. Turning those into PAGER comparisons would mean regenerating the exposure catalogue behind them rather than copying a column across, which is real work rather than a configuration change. But the choice being made now is not between old data and new; it is between a table that says where the ceiling is and one that does not.
A note on the sample: it was built from two queries against USGS’s event service — magnitude 6.3 and above for 2018 to 2024, and 6.0 and above from 2024 to today — so the population is not uniform across the whole period. The 2007 cutoff is present in both halves and in every year within them.