Surveys reach sub-Milky Way masses at z>2 and numerical models produce realistic galaxies from gas and dark matter. Yet, a predictive theory that answers questions like "Why does a galaxy have its stellar mass, star formation rate, and size?" eludes us. Largely, this is because data capture different systems at different times while models describe the same systems at different times. I will illustrate how this schism allows multiple physical explanations to fit the facts, blurring our understanding of the forces that control a galaxy's mass and structural development. I'll then show how new data and inference techniques let us bridge this gap by reconstructing spatially resolved star formation histories to directly confront models in their native time domain, vastly increasing our empirical discriminatory power. Lastly, I'll discuss some new ideas for exploiting current data to provide robust, statistically powerful insights covering large parts of parameter space at z < 1. These methods could become routine in the JWST/GMT era, transforming the relationship between observation and theory in extragalactic astronomy.