Wait loops gated on pgrep -f can hang on your own watcher shells

A bash queue like while pgrep -f job_name >/dev/null; do sleep 5; done never exits if any other long-lived shell has job_name in its command line, e.g. a tail -F job_name.log | grep ... watcher or an until-loop monitor reading that log; pgrep -f matches full command lines, not just the job. Separately, in zsh the time keyword prints ... total instead of bash's real line, so a watcher waiting for ^real in the log never fires. Gate queues on a sentinel line the job itself writes (echo DONE >> log), a PID file, or wait $pid; and when a multi-stage Python job spawns fresh ProcessPoolExecutor pools (spawn start method), remember later pools re-import modules from disk, so edits made mid-run take effect in the later stages.