Engineering · Events & stadiums

Egress in under twelve minutes: the math behind safer crowd flow

Stadium egress is the most consequential ten minutes of any event. A privacy-first density layer can turn the post-match crowd into a routing problem, not a bottleneck. Here's the math and the architecture.

A packed stadium crowd silhouetted against the floodlights from the upper bowl
Sixty thousand people, ten minutes, and the routing layer that holds it together

The final whistle is a routing problem. Sixty thousand people stand up at the same instant, walk the same direction, and pass through a small number of doors. Every minute of egress beyond the planned envelope is a minute the safety brief gets uncomfortable. The math that holds the night together isn't crowd surveillance; it's density routing.

We don't have a stadium pilot running yet. This post is about how we'd configure one.

01 · THE NUMBERS THAT DRIVE THE BRIEFThe numbers that drive the brief

Most major venues operate to an egress target: eight to twelve minutes from full bowl to empty, depending on geometry and licence. Beyond that envelope, two things happen at once: dwell at the gates grows, and the transport hub beyond the venue can't absorb the throughput. Both are failures of routing, not capacity. The gates exist. The trains run. The crowd just doesn't know which gate is moving fastest, and the platform supervisor doesn't know how many people the venue is about to release in the next ninety seconds.

A privacy-first density layer would fix the routing on both sides:

  • Inside the venue, the platform watches every gate, ramp and stairwell. It ranks them by live density and dwell, and surfaces the under-used routes attendants can push the crowd toward.
  • At the transport interface, the platform reads the transit operator's own platform counts (or its public feed) and paces the venue's release rate to match. The concourse holds the crowd for ninety seconds; the next train arrives; the hold releases.

Neither side requires knowing who anyone is. Neither side stores video. Both sides count.

02 · WHAT A WORKED EXAMPLE WOULD LOOK LIKEWhat a worked example would look like

A 60,000-seat venue, post-match egress, a four-zone test deployment on the busiest concourse and its two feeder ramps. The egress envelope on the licence is twelve minutes. The platform's job is to keep it inside the envelope and surface the routing nudges in time for the attendants to act on them. A worked timeline of a typical alert sequence:

  • t+00:00 - final whistle. Bowl at 96% of seated capacity. Concourse density baseline.
  • t+00:48 - main gate sustained 92%. Attendant team on the feeder aisle is pinged: redirect to the under-used gate. Density falls within 90 seconds.
  • t+02:30 - transport interface 86%. Platform near capacity. Next train 3 minutes. The platform holds 60 people at the concourse exit.
  • t+04:14 - main gate 64%. Recovered. Holds released.
  • t+11:32 - bowl 87% emptied. Inside the envelope. Post-event report queued.

The events would be unremarkable. That is the point. A routing layer that moves a thousand people across an aisle thirty seconds before the bottleneck would have formed is a layer nobody notices, because the bottleneck doesn't form.

03 · THE TWO CHARTS A SAFETY SUPERVISOR WOULD READThe two charts a safety supervisor would read

Two charts would come out of every event.

The first is density-by-zone-over-time: one row per gate, ramp, stairwell and concourse segment, colour-coded against the threshold matrix. The supervisor reads it like sheet music: teal where the venue was inside its envelope, amber at the edge, coral where the platform pushed an alert. The second is the egress curve: total people remaining inside the bowl, by minute, plotted against the licence target.

Two questions get answered every week from those charts: did the licence target hold, and what would a different gate state have changed.

04 · WHAT THE ARCHITECTURE WON'T DOWhat the architecture won't do

It will not identify a single attendee. The model is on the device, frames don't leave, the data the platform sees is the anonymous count the dashboard exposes. There is no person-of-interest list, no attendance log keyed to identity, no biometric inference of any kind. A venue that needs that has the wrong vendor, and most venues that operate at this scale know exactly why the wrong vendor is the wrong vendor.

The duty of care doesn't require surveillance. It requires routing. The routing is what we count.

05 · WE'RE LOOKING FOR THE FIRST STADIUM PILOTWe're looking for the first stadium pilot

If you operate a venue and the egress envelope is the room you spend most of the safety brief defending, talk to us about a pilot.

If your building can't answer one of these questions yet, we should talk.