We first explore a series of retrospective earthquake interactions in southern California. We find that the four shocks in the past 150 yr brought the Ridgecrest fault closer to failure. Examining the 34 hr time span between the 6.4 and 7.1 events, we calculate that the 6.4 event brought the hypocentral region of the 7.1 earthquake 0.7 bars closer to failure, with the 7.1 event relieving most of the surrounding stress that was imparted by the first. We also find that the 6.4 cross‐fault aftershocks shut down when they fell under the stress shadow of the 7.1. Together, the Ridgecrest mainshocks brought a 120 km long portion of the Garlock fault from 0.2 to 10 bars closer to failure. These results motivate our introduction of forecasts of future seismicity. Most attempts to forecast aftershocks use statistical decay models or Coulomb stress transfer. Statistical approaches require simplifying assumptions about the spatial distribution of aftershocks and their decay; Coulomb models make simplifying assumptions about the geometry of the surrounding faults, which we seek here to remove. We perform a rate–state implementation of the Coulomb stress change on focal mechanisms to capture fault complexity. After tuning the model through a learning period to improve its forecast ability, we make retrospective forecasts to assess model’s predictive ability. Our forecast for the next 12 months yields a 2.3% chance of an Garlock fault rupture. If such a rupture occurred and reached within 45 km of the San Andreas, we calculate it would raise the probability of a San Andreas rupture on the Mojave section by a factor of 150. We therefore estimate the net chance of large San Andreas earthquake in the next 12 months to be 1.15%, or about three to five times its background probability.