Most interesting finding: showing Jev a corrected example barely moved it (p=0.01). Putting the rule in its criteria helped (0.69). Asking "does this rule apply?" as a separate yes/no question worked…