An Asymmetric Framework for Inferential-Token Safety: Normality Inputs and Coordination Constraints for Conjunction Introduction/Elimination

Safety is commonly framed as robustness of truth across relevant counterfactual cases. Dutant & Rosenkranz (2024) show that natural counterpart relations can produce coordination failures for conjunction introduction and elimination (CI/CE), undermining safety preservation and encouraging asymmetric treatment of closure. This paper offers a Carnapian explication of inferential-token safety: a counterfactual evaluation of tokens ⟨α,b⟩ (case plus basis) that makes normality and coordination parameters explicit. A normality input Rel = Rel_E^k(α*) fixes relevant cases; a coordination constraint (CTA) requires structural replay of the same basis in each β∈Rel, with diagonal replay mandatory when formable. A minimal implementation CN^k returns the diagonal token if replay succeeds and ∅ otherwise. Comparing weak+coverage (NV), strict, and indeterminacy (⊥) treatments of ∅, the framework yields (i) a decomposition of CI/CE preservation into truth, coordination, and ability stability, and (ii) under strict/⊥, propagation of falsity or replay failure, blocking local bootstrapping. The framework is deliberately conditional and leaves any bridge to knowledge safety open.