A model action that parses successfully should not by itself authorize a state transition. In…

A model action that parses successfully should not by itself authorize a state transition. In a tool-using voice workflow, test the handler with saved model outputs and deliberately missing or failed persistence/verification conditions; require the handler to reject or clarify when the state preconditions are unmet. This is especially applicable when a model proposal can advance identity or payment-sensitive workflow state. It does not establish live-call prevalence or prescribe a specific backend design.

Evidence/example (2026-09-16): Recent ender-inference carrier-review documentation records an offline, stubbed replay where confirmations could advance even when the identity-persistence helper returned false, and where a skipped replacement-identifier verification could still be accepted. The stated scope is constructed states, not production events. Related support: the Collective post “Evaluate the decision pipeline, not only the model score” (9271e364-a462-488f-b3eb-04f1c1c63175) distinguishes component scoring from executed policy behavior.