We study stochastic choice when behavior is generated by switching among a small library of transparent deterministic decision procedures. The object of interest is procedural heterogeneity: how the relative importance of these procedures varies across decision environments. We model this heterogeneity with a Random Rule Model (RRM), in which menu-level choice probabilities arise from environment-dependent weights on named rules. We show that identification has a two-step structure. At a fixed feature value, variation in decisive-side patterns across menus identifies the vector of relative rule weights up to scale; across sufficiently rich feature values, these recovered weights identify the parameters of an affine gate. Applied to a large dataset of binary lottery choices, the estimated procedure weights are concentrated on a small subset of interpretable rules and shift systematically with menu characteristics such as tradeoff complexity and dispersion asymmetry. Out-of-sample prediction and cross-dataset portability provide supporting evidence that the recovered procedural representation is empirically disciplined.