We argue that in regulated high-stakes AI decisioning environments, behavioral and social science theory should not be absorbed implicitly into opaque learned models, but rendered explicit as auditably constraining policy modules. We formalize this principle as the Explicit Behavioral Constraint (EBC) framework and demonstrate its application in the design of AxiomCausal, a decision support system for institutional debt recovery.The EBC framework defines a policy module as a tuple (C, X, f, V, K) where C is a named latent behavioral construct, X is a documented set of observable proxies, f is a deterministic operationalization function, V is an explicit validation status, and K is an auditable constraint on the recommendation function. This formalization makes the theoretical commitments of a decisioning system visible, auditable, and falsifiable — a property increasingly required by AI accountability regulations including the EU AI Act.We present seven EBC modules for institutional debt recovery derived from procedural justice theory (Tyler 1990), face dynamics (Goffman 1967), graduated sanctions (Ostrom 1990), prospect theory (Kahneman and Tversky 1979), labeling theory (Becker 1963), game theory (Nash 1951), and behavioral inertia. For each module, we specify the theoretical source, observable proxies, constraint logic, and — critically — its current validation status, ranging from Conceptual to Partially Validated. No module is claimed as fully empirically validated in the debt recovery context.We present a formal six-layer decision function architecture integrating the EBC modules with a Double Machine Learning causal estimation component and a human-in-the-loop governance layer. We report verification experiments confirming auditability properties and EBC interaction correctness. We make no claim of empirical superiority over alternative recovery systems in the absence of a controlled field trial.
Naoufal Chaara (Mon,) studied this question.