Every config over the full benchmark, baseline included. Bars are sorted best first; hover a bar for its value.
Pick a metric. Each config's running total builds up task by task on a shared axis, so the lines are directly comparable.
Change in cost versus the baseline over one pass of the set, with a 95% confidence interval (paired bootstrap over the units of the set). A tool whose interval includes 0% has no measurable effect on cost. Success and quality are for the selected set only.
| Tool | Cost change | 95% interval | Pass rate | Quality |
|---|