Challenges in the relationship between theory and observation, such as the theory-ladenness of observations, can be interpreted as a call for general methods to evaluate theories independently of observational data. In this presentation, I will discuss such methods, with a focus on theories about mechanisms of higher cognitive functions. Central to the discussion is the definition of a 'mechanism' as a set of entities with specific properties that lead to systematic activity. I will explore the following key points as problematic: Theories based on proposing novel, dedicated mechanisms (i.e., introducing new entities), theories based on concepts that are challenging to link to any tangible entity, and theories that are only specified at a singular level of abstraction (i.e., overlooking entities at lower levels). All three aspects relate to the principle of parsimony, which favors theories requiring fewer assumptions. However, this discussion aims to refine the definition of 'assumption,' to reduce ambiguity in applying the parsimony concept, thereby enhancing its utility in evaluating theories in cognitive sciences independently of observational data.