| Status | Pass or fail for this case | Start with the failures |
| Score | The scorer’s output, normalized where supported | A low or withheld score means the scorer could not decide — inspect the case |
| Cost | The model-call cost for this row, in USD | Spikes often mean a runaway prompt or tool loop |
| Latency | Model-call wall time | Compare against your latency budget |
| Provenance | Complete or incomplete | Incomplete provenance cannot be used as release evidence |