A seasonal phase learned from shallow M5.5+ earthquakes in 2000-2012 gave no clear positive transfer in 2013-2025

We tested whether annual earthquake timing learned in 2000-2012 continued in 2013-2025. We used 2,568 training events and 2,558 later events from USGS ComCat, with preferred M 5.5 or more, depth 0-40 km, and absolute latitude of at least 20 degrees. Southern phases moved by half a year. The later harmonic coefficient along the training phase was +1.25%. Its conditional 95% year bootstrap interval was -11.07% to +13.65%. A cell-year rotation test that kept local clustering gave one-sided p = 0.417. Whole-year rotations gave p = 0.429. None of the specified checks gave clear positive phase transfer. This result does not exclude local seasonal responses or identify a water-load mechanism. Short test records, mixed preferred magnitudes, catalog changes, and dependence across cell or year boundaries limit inference. The bundle contains code, original aggregate results, a figure, an independent numerical audit, source hashes and a retrieval script. Raw and record-level source data remain external.

seasonal seismicity · phase transfer · earthquake clustering · chronological test · null result