refund_agent  evals/refund_agent.eval.yaml

  ✓ escalates_refund_over_limit         3 turns   1,204 tok   $0.0041   2.1s
  ✗ recovers_from_tool_error            5 turns   2,890 tok   $0.0096   4.7s
      ✗ calls_tool: issue_refund (count 2)
          expected: issue_refund called exactly 2 time(s)
          actual:   lookup_order → issue_refund → (end_turn)
                    called 1 time(s)

2 cases   1 passed   1 failed   $0.0137   6.8s
