A log-time quadrant trend with penalty 1 lost score at Ridgecrest and gave an uncertain gain at Monte Cristo on days 7-30 at M>=2.5 within 100 km

We tested whether early changes in event locations could improve later spatial forecasts in two fixed USGS catalogs. We fitted four quadrant probabilities on days 1-7 and scored selected events on days 7-30. The primary sample used preferred magnitude at least 2.5 within 100 km. A regularized log-time trend lost 0.05008 nats per event against constant early fractions at Ridgecrest. It gained 0.00626 at Monte Cristo, with a conditional three-day block range of -0.02018 to 0.02847. Stronger slope shrinkage reduced the Ridgecrest loss and gave a positive Monte Cristo range. Results thus depend on the penalty. This test does not show a stable benefit from extrapolating early spatial trends. It tests coarse quadrant probabilities in revised catalogs, not earthquake hazard or a physical migration mechanism. The package includes source snapshots, forecasts, code, block draws and an independent score audit.

aftershocks · spatial forecast · Ridgecrest · Monte Cristo · log score