Abstract
<jats:p>Mechanistically specified models are often treated as interpretable because each coefficient has an assigned role in the equations. That semantic assignment is necessary but not sufficient for inference: parameter interpretation also requires structural identifiability of the parameter-to-observation map and practical recovery under finite, noisy data. We formalize this distinction for the Continuous Moral-Social World Model (CMS-WM), a simulator with phenotype-state, world-state, and action-policy modules. The revised analysis makes the complete withdrawal-side policy explicit, defines the observation operator, proves exact softmax scaling and common-shift invariances, and evaluates effective withdrawal parameters in reproducible synthetic recovery experiments. In a principal batch of 30 ground-truth settings with 1,000 trials each, the ambiguity-linked coupling showed moderate rank recovery (Spearman rho = 0.52; median normalized absolute error = 0.19), whereas the friction-linked coupling was weak under correlated inputs (rho = 0.21; median error = 0.34). When friction was experimentally orthogonalized from ambiguity, friction recovery improved (rho = 0.68; median error = 0.087), demonstrating configuration-dependent rather than intrinsic interpretability. Pressure coupling and the withdrawal threshold remained insufficiently rank-recoverable in the principal design (rho = 0.37 and 0.30, respectively), despite substantial threshold activation. In contrast, out-of-sample action-regime prediction was stable: dominant-regime agreement was 100%, median predictive L1 distance was 0.0187, and median Jensen-Shannon divergence was 7.2 x 10^-5. These findings support a hierarchical reportability framework: regime-level behavior is reportable; the ambiguity-linked parameter is cautiously reportable; friction is design-conditional; and pressure coupling and threshold should not receive substantive mechanistic interpretation under the present observation model. The study is an internal synthetic verification of the model-observation pairing, not external validation of psychopathy or antisocial behavior.</jats:p>