To develop earthquake prediction research as a sound science, it is essential to construct physical models with predictive capabilities and to monitor the observables that constrain the parameters of the models. The present paper proposes an approach toward this goal by using the following recently published results as building blocks: (1) The extrapolation of laboratory results to the field observations on earthquake faults by Ohnaka and his colleagues; (2) The discovery of temporal change in frequency dependent seismic attenuation and in the magnitude-frequency relation related to stress re-distribution after 1995 Hyogoken Nanbu earthquake by Hiramatsu and his co-workers; (3) The discovery of a systematic relationship between moment magnitude and the sub-event size from strong motion records for earthquakes in eastern north America and elsewhere by Beresnev and Atkinson; (4) The estimation of seismic energy from an earthquake taking into account the scattering and absorption loss by Sato and his colleagues, which will enable an accurate estimation of temporal change in apparent stress, and (5) The computer simulation of earthquake fault ruptures by Ward. These simulations can assimilate available data from geological studies on active faults in the modeling of earthquake occurrence. The key concept unifying all of the above studies is the slip weakening model of the rupture over a heterogeneous fault plane. Slip weakening friction might be too simple, but we believe that it is a good starting point for our purpose.