A local-season fit from 2001-2012 gave lower month scores for shallow USGS earthquakes in 2013-2025

We tested whether a fixed annual cycle improves later earthquake month probabilities. We used a shared USGS catalog with preferred magnitude at least 5.5. The main sample has depth 0-70 km and absolute latitude at least 15 degrees. We removed events within 100 km and 30 days after an equal or larger event. A fit to 2,243 events in 2001-2012 was scored on 2,316 events in 2013-2025. Southern months were shifted by six months. The model conditioned on year and hemisphere totals and used month lengths. The log score gain over a constant rate per day was -0.00377 bits per event. The conditional 95% three-year block interval was -0.00659 to -0.00112. The separate-year interval was -0.00741 to +0.00005. Wider event-removal windows and a magnitude 6 cut also gave negative point gains, with three-year intervals that included zero. This simple local-season fit gave no held-out gain. The result does not test local water loads or rule out regional seasonal effects. A revised catalog, a short test period, fixed fitted coefficients and approximate removal of related events limit inference. Code, source hashes, aggregate scores, bootstrap draws, a separate count audit and a figure are included. External event records are not bundled.

earthquake seasonality · held-out log score · ComCat · annual harmonic · null result · hydrological loading