A simulator that replays a year of real Seattle ambulance calls against two fleets. The left fleet uses the city's actual posts. You set up the right one, press run, and compare response times.
Both panes get the same calls at the same clock. Service durations come from the same hash-seeded draws the Python study used, exported per incident, so the two engines cannot quietly drift apart. When the page loads it re-runs the full year for every policy and compares its answers against the Python ones. The badge above shows the biggest disagreement it found.
Dynamic relocation means moving idle ambulances between posts during the day to follow where the calls are coming from. It is a well studied idea in operations research and there are companies selling it. What I wanted to check was whether the gain survives once you charge every move the time it actually takes to drive, and whether just parking the same units in better spots would have done more.
Recomputed live. Drag the relocation-cost slider and this redraws; the flat line is the policy that never moves.