MIRN · Interactive Demo

Run the room twice

This crowd is invented. These are not real pedestrians and the figures below are not measurements of anybody — they come from a deliberately small model running in your browser, not from the MIRN engine itself. What the toy is faithful to is the construction: the same room runs twice from identical starting positions and an identical noise tape, differing only in whether the robot is there, so the true effect is known rather than estimated. Turn the dials and watch what the forecast metric says about it.

With the robot Same room, no robot The robot

Settings

How much space the robot demands — 0.60 m

How far ahead the forecaster guesses — 3.0 s

Which room

Placebo run — the robot is in the room but nobody can see it, so the true effect is exactly zero.

What the crowd really did 0.000 m Assumes: both runs available. Reads zero when the robot changes nobody’s path, by construction.

What the forecaster reports 0.000 m Assumes: people would have kept their speed and heading. On a true zero it reads 0.000 m, not nought.

Ordinary gap between two runs 0.000 m Assumes: nothing. Two runs of the same room with no robot in either — anything smaller than this is not a signal.

The real console carries thirteen dials and fifteen readouts, two crowd models, a Python oracle checking every formula the browser quotes, and a referee drill that withholds the second run and asks you to make the call yourself. It runs locally — there is no hosted copy, because the whole thing is static files. Source on GitHub.