Which forms get picked, and how much we trim them.

For every validated instrument: how often the agent decided to run it, the average number of items asked, and how many items the adaptive engine skipped versus the full battery.