By: ValentinHirschi <valentin.hirschi@gmail.com>
Group finite campaign windows into shared parallel traversals Keep the original prefix partition and existing inner worker pool while amortizing small-window preparation and durable commits. Bind exact union admission and occurrence accounting to atomic multi-cursor publication; support configurable grouping on resume. Expose the policy in Python and native monitoring, retain the independent legacy/Epoch lane, and validate cancellation, resume, publication and exact application. Record counterbalanced four-loop gains, neutral five-loop R0 results and the limits of 150-worker scaling.