Mathematical models of dynamic complex systems primarily rely on differential or time-recursive equations. In cognitive sciences, however, the scenario is distinctly different from disciplines like physics and engineering that operate on foundational principles like Newton's laws or Maxwell's equations, facilitating the derivation and interpretation of these equations. In contrast, cognitive sciences necessitate an empirical approach due to the absence of such foundational principles. In essence, we are required to infer the governing equations of cognitive systems (behavioral and neural) based on what we can practically observe. However, these observations are inherently constrained. Given the lack of cognitive first principles, we face uncertainties regarding what to measure (i.e., state variables) and the duration (time scales and dimensionality) for which measurements should be taken. Prior research from our group and others indicates that any of these factors can compromise the entire machine learning process, resulting in models that lack replicability, explainability, predictability, and do not translate into traditional intervention mechanisms informed by control theory