Characterize agent adaptation under alternative pricing schemes

Characterize how LLM agents adapt their information acquisition and resource allocation when evidence and tool prices differ from the common tariff used in JockeyBench.

Background

JockeyBench evaluates research expenditure under a single common tariff covering evidence cells, data-tool calls, and sandbox CPU time. The tariff enables comparison across configurations but does not reveal how agents would change their acquisition policies or resource budgets when prices change.

The paper explicitly states that adaptation under alternative pricing schemes has not been resolved, leaving the effects of pricing on evidence acquisition and resource allocation as an open problem.