This work proposes a formal framework for analyzing the selection of cognitive architectures as a phase process. The model distinguishes between local learning dynamics and global architectural transitions, introducing concepts such as selection functions with thresholds and hardness parameters, reproducibility, variability, phase transitions, and phase locking. The framework emphasizes that architectural transitions constitute rare phase events dependent on specific configurations of selective and reproductive parameters. Cognitive complexity is not treated as an intrinsic evolutionary attractor but as a contingent consequence of particular phase regimes.