The brief
A physical relief model of a city is one of the best public engagement tools there is. People who will not read a planning document will lean over a model of their own street and point at it. The problem is that a model says exactly one thing forever. It is built, and then it is finished.
Projection is the obvious escape, and it usually fails on the boring details. The projector cannot always be hung directly overhead, so the image arrives skewed. Getting new data onto the surface means preparing it in advance, so a session cannot follow the question a room is actually asking. And the audience stays an audience, because there is no route from looking at the model to putting something on it.
PARM-X is the software that fixes those three. Nottingham’s model and the City as Lab programme around it were already running. What was missing was a way to drive the projection live from a browser, so a model can show anything the city publishes, from any projector angle, while the people standing around it add to what is on it.
The audience turned out to matter as much as the mechanism. The same platform runs a public session in a library and an executive planning session, and in the second case the four hours that used to go into preparing cartographic outputs for a meeting now go into the meeting itself.
The insight
The architectural decision that made the rest tractable was to treat the installation as three independent browser windows rather than one application driving multiple displays. The operator console, the projected map and the public screen each run standalone and stay in step over BroadcastChannel.
That sounds like a small thing and it removes an entire class of problem. There is no server to deploy, no socket connection to drop halfway through a session in a building with awkward wifi, and no privileged display process to crash and take the others with it. A window can be closed and reopened mid-session and it rejoins the current state. For an installation that has to survive being run by someone who did not build it, in a room nobody tested in, that resilience matters more than elegance.
The second decision was to solve calibration properly. A scale-and-offset transform is quick and only works from directly overhead. Solving a full homography from four clicked corners means the projector can sit wherever the room allows, which is what determines whether the system can be installed somewhere that was not designed for it.
What it does not do
Calibration state lives in the session store, so a fresh browser session needs recalibrating. That is a deliberate trade for the moment, since a saved calibration that silently no longer matches the projector position is worse than one you know you have to redo, but it is friction at the start of every session.
Agent simulation is illustrative rather than predictive. Agents follow a road network and a set of origin and destination presets, and the flows are plausible, but there is no calibrated behavioural model underneath and the output should not be read as a forecast.
The platform is configured for one city. Retargeting is a matter of bounds constants and endpoints rather than architecture, but the scenario presets, the landmark sets and the curated media library are all specific to Nottingham and would need rebuilding.
