Jev runs a city’s traffic

# Jev Traffic Sim

An interactive Chicago traffic simulator that runs **one experiment**: you watch a single car
cross real Chicago geography while the whole city simulates around it, and the same scenario
is also played out by two deterministic baselines.

**Live:** deploy URL is published with the release notes.

## The experiment

| Column | What it is |
| --- | --- |
| **Fixed** | Signals on a fixed rotation. No sensing, no adaptation. The do-nothing baseline. |
| **Adaptive** | Signals that react to local queue pressure and waits, with the same min/max green, yellow, all-red and starvation protection as everything else. |
| **Jev** | A citywide policy from an external evaluation model (TypeSafe AI's `jev` through the Vercel AI Gateway). It sees only citywide traffic state and answers bounded, typed questions. Its answer becomes corridor/region weights and a small pressure modifier. |

The visible run asks Jev for citywide policy. Fixed and Adaptive play the *same* scenario
headlessly, in a worker of their own, so the trip you are watching never stalls. The label
reports what actually governed the visible run.

**Nothing about the driver changes the scenario.** Tourist and Local are two ways of driving
the same trip in the same city: the tourist follows the shortest route and sits through
queues; the local knows the network and reroutes around trouble.

## Same-scenario fairness

A comparison is only shown when it is honest:

- all three runs share one **scenario fingerprint**: the scenario (trip, traffic level,
  driver, seed, horizon) is hashed into a short id, and results with different ids are never
  put side by side;
- all three runs are built from one **demand spawn list** and one **incident script**, so the
  weather in the city is byte-identical and only th