CrowdSynthetic

The whole simulation is driven by a single layout.json file. You define zones, a stage, and exits, and the geometry shapes how the crowd moves and where risk accumulates. Change the file and the entire run adapts. That makes it easy to test the same scenario logic against different floor plans.

Each zone carries a capacity, and the simulator tracks how close it's getting to full. Risk isn't one global number. It's computed per zone, so you can see exactly which section is under pressure. That granularity is what makes the prediction useful instead of just decorative.

Rather than only flagging an area as full, CrowdSynthetic reads the trend and forecasts ahead. It surfaces messages like "FRONT in X seconds" or "FRONT saturated," giving you a moving estimate of when a bottleneck will hit. This is the core safety value: a warning that arrives before the crowd packs in. Timing matters more than the raw count.

When risk crosses the threshold, the sim switches itself from NORMAL into EVACUATION mode without you pressing anything. The crowd behavior changes, exits come into play, and the run holds on a final freeze-frame after the transition back. It's a self-contained demo. No buttons required. That automated loop is the reason the project exists.

A heatmap sits over the venue view and shows crowd density at a glance. Dense pockets light up, so you can spot congestion hotspots visually instead of reading numbers. Call it heatmap crowd visualization without the dashboard. It's the kind of overlay that turns a wall of raw coordinates into something a room full of people can actually read during a live demo. Useful for spotting a looming pinch point before the metrics spell it out.

The tool generates plain-language recommendations based on the current risk state. Instead of only a red number, you get a suggested action. It reads like an advisor sitting next to the operator. The wording comes from the simulation's own rules rather than a large language model. So while it sounds smart, the suggestions follow the same fixed thresholds that drive the rest of the risk engine, which makes them predictable and easy to audit but never particularly surprising or creative.

Every run writes its metrics to concert_risk_log.csv and concert_risk_log.json. That means you can replay a session's numbers, chart them, or compare layouts later. For anyone doing analysis rather than just watching, this is the part that turns a demo into a dataset.