I express an agent's learning ability through a maximal menu of zero welfare. A simple duality links this zero set and posterior-separable learning costs that are common in the Rational Inattention literature. The novel representation helps to elicit the agent's prior belief and perceived learning ability in a simple, cost-effective and versatile way. Relative to existing approaches, the protocol also works when learning cannot be tightly controlled, and when learning produces lasting effects that alter the beliefs or costs in subsequent tasks. An experimental implementation is planned, and the manuscript will be updated once the findings are available.
