Test exclusions hide causal errors

This proposed study asks whether knowledge can drift during human-AI handoffs when a recipe still passes its usual checks. The key idea is that a predecessor may pass along a rule that discourages one diagnostic test: a deliberately chosen change that separates recipes which otherwise behave alike. In a harmless recipe simulator, successors would receive either that inherited exclusion, the same earlier tests as an unordered record, or a replacement rule that samples tests uniformly or guarantees diagnostic coverage. The recipes, test totals and later outcome exposure would match. If inherited exclusion led people to choose the diagnostic test less often and later miss source-specific differences, beyond established learning models, it would support this proposed mechanism.