Change who goes first.
Change how fast you watch.
Build a variation from the edited prompt and compare it with the supplied program. The expected effects below are predictions for that build; testing a fresh agent-generated version is your next experiment.
Share the first turn
couriers in ID order
Rotate first priority each tick: 1-2-3, then 2-3-1, then 3-1-2; keep the existing reservation and collision rules.
Different couriers get the first chance at a contested square. This may improve fairness. Progress on arbitrary maps remains a separate property to test.
Sequential reservations favor whoever is considered first. Rotating that order changes who wins a contested move. The shortest-path calculation stays the same.
Set two couriers one step from D. On the tick where courier 2 has priority, it should enter D first. Check unique positions and forbidden swaps too.
Keep the other requirements unchanged.
Proposed variant: a fresh agent-generated build and test remain the next steps.
Make a default explicit
The prompt leaves the tick interval unspecified; the reference program uses 250 ms.
Add: Run at eight ticks per second, and keep One tick as one complete simulation update.
The same sequence plays in half the real time. A delivery should take the same number of simulation ticks.
The timer requests updates more often; routes and per-tick decisions are unchanged.
Run identical starting worlds for 50 ticks at both rates. Compare parcel totals and courier positions. Then compare elapsed playback time.
Keep tick-based jam detection unchanged unless you deliberately want its real-time delay to remain two seconds.
Proposed variant: a fresh agent-generated build and test remain the next steps.