How we checked the model

We ran the model for past weeks using only records available before each week in the historical catalog. These retrospective tests are not real-time validation.

26 weeks checked. Average weekly error: 6.6 earthquakes.

The simple historical-average baseline had lower error in this test. We cannot claim that the more complex model is better.

These tests concern regional counts of magnitude 2 and greater events, not predictions of location or magnitude. The 35-year archive is shown separately as historical context.