| 167 | |
| 168 | |
| 169 | def aggregate_ops(results: list[ScenarioResult]) -> AggregateOps: |
| 170 | if not results: |
| 171 | return AggregateOps() |
| 172 | |
| 173 | durations = [r.ops.duration_ms for r in results if r.ops.duration_ms is not None] |
| 174 | costs = [r.ops.est_cost_usd for r in results if r.ops.est_cost_usd is not None] |
| 175 | |
| 176 | return AggregateOps( |
| 177 | tokens_in_total=sum(r.ops.tokens_in for r in results), |
| 178 | tokens_out_total=sum(r.ops.tokens_out for r in results), |
| 179 | duration_ms_p50=_percentile(durations, 50), |
| 180 | duration_ms_p95=_percentile(durations, 95), |
| 181 | tool_calls_total=sum(r.ops.tool_call_count for r in results), |
| 182 | est_cost_usd_total=round(sum(costs), 6) if costs else None, |
| 183 | ) |
| 184 | |
| 185 | |
| 186 | def _percentile(values: list[float], pct: float) -> float | None: |