Dispatch experiments across N machines, monitor progress, aggregate results, and pick global best. Provides run_queue.sh (per-machine experiment queue with smoke-test gating), aggregate_results.sh (pull all results.tsv → results_all.tsv), check_runs.sh (live progress dashboard), pick_best.sh (rsync top-K ckpts back). Use when project has multiple servers (set up by server-fleet-config) and the user wants parallel experimentation. Triggers when user says "spin up experiments on N machines" or "see what's running across the fleet".
