Hedging: a faster runner takes the same job, the first to finish wins issue #78 open andymccall opened this 10 days ago · 0 comments

When a markedly faster runner frees up while a job runs on a slow one, and the learned durations say it would finish sooner (counting restoring the checkpoints), start the job there too (from the latest checkpoint); the first to finish wins and the other is cancelled. Bounded: one hedge per job, only for jobs a policy allows (the critical path, or always on owned runners), never for jobs with side effects (deployments). When both finish, their outputs are compared: a free determinism check. Design: docs/handover.md §8.