These data are from Laboratory Earthquake Experiments from the Cornell 3 m apparatus in support of the following research: This paper reviews laboratory observations of earthquake initiation and describes new experiments on a 3 m rock sample where the nucleation process is imaged in detail. Many of the laboratory observations are consistent with previous work that showed a slow and smoothly accelerating earthquake nucleation process that expands to a critical nucleation length scale Lc, before it rapidly accelerates to dynamic fault rupture. The experiments also highlight complexities not currently considered by most theoretical and numerical models. This includes a loading rate dependency where a “kick” above steady state produces smaller and more abrupt initiation. Heterogeneity of fault strength also causes abrupt initiation when creep fronts coalesce on a stuck patch that is somewhat stronger than the surrounding fault. Taken together, these two mechanisms suggest a rate-dependent “cascade-up” model for earthquake initiation. This model simultaneously accounts for foreshocks that are a byproduct of a larger nucleation process and similarities between initial P wave signatures of small and large earthquakes. A diversity of nucleation conditions are expected in the Earth’s crust, ranging from slip limited environments with Lc < 1 m, to ignition-limited environments with Lc > 10 km. In the latter case, Lc fails to fully characterize the initiation process since earthquakes nucleate not because a slipping patch reaches a critical length but because fault slip rate exceeds a critical power density needed to ignite dynamic rupture.