With only one configuration of runs (e.g. only with_skill), aggregate_results() emitted a delta computed against an invented zero baseline, so the benchmark report claimed a pass-rate/time/token improvement that no comparison supported. |
||
|---|---|---|
| .. | ||
| agents | ||
| assets | ||
| eval-viewer | ||
| references | ||
| scripts | ||
| tests | ||
| LICENSE.txt | ||
| SKILL.md | ||