Can someone review this PR by Laminar? https://github.com/OpenHands/benchmarks/pull/464 It makes sure that the timeline accurately reflects real execution time. Currently, the timeline will be overestimated when there are more tasks than workers. I validated it by triggering a 5 instances swebench job