The cellular view of a conference
The sparsest and quietest room here — and nearly every collision is deliberate.
The conditions of this room
Four numbers decide what a crowd does: how tightly it is packed, how loud it is, how fast alcohol arrives, and how much attention is available for the people beside you. Everything below is either a published figure or derived from one.
What those conditions do to people
A conference floor is the opposite of every other venue here. At 0.45 people per square metre it is Fruin grade C, the only room modelled where personal space is merely intruded on rather than eliminated, and at 72 dB it has the longest speech range on the site: 3.24 metres, the full width of Hall's social zone.
That means a conversation can form across a gap, between people who chose each other, rather than between whoever happens to be pressed together. Attention is set at 0.90 — being there to meet people is the entire stated purpose of the format.
The output is unusual: a middling encounter rate of 12 per person per hour, but a large connected component of 435 people out of 1,200 — third-highest here, behind only a queue and a pilgrimage. A conference does not generate much contact, and converts an unusual amount of what it does generate.
What the model produces
Measured output from a standard run: 1,200 people, 4 hours of venue time, fixed seed, no parameter overrides. Every room on this site is run under identical conditions, so these figures are a like-for-like comparison rather than a set of individually flattering runs.
The crowd is a simulation and is presented as one. What is not invented are its constants: walking speed against density, the proxemic zones, the collapse of speech range with noise, the Widmark blood-alcohol curve and the conversational group limit all come from published work, cited above and listed in full in the engine.
Real places like this
Every gathering below is modelled as this kind of room. Attendance figures are published numbers for the period stated.
Questions
Does conference networking actually work?
In this model, disproportionately well. It produces fewer encounters than a nightclub but ends with roughly twice the connected network, because the room is quiet, sparse and attention is almost entirely available.
How far apart can you hold a conversation?
In a room at around 72 dB, roughly 3.2 metres — essentially Hall's full social zone of 3.6 m. Every 10 dB added to the room cuts that distance sharply.
How big are the largest tech conferences?
CES drew around 138,000 attendees in 2024 and Web Summit around 71,000 in 2023, though only a fraction of either is on any one floor at a time.
How this is modelled
Each person is one cell with a position, a body mass, a Widmark distribution ratio, an openness trait and a reserve of energy. They walk at the speed the local density allows, drink at the rate the room serves, and when two of them stay within speech range long enough, a bond forms. Most bonds are nothing.
What separates an encounter that creates something from one that does not is whether it bridges two groups that were not previously connected. A tie inside an existing group mostly restates what that group already knows; a tie across a gap carries information neither side had. That distinction is Granovetter's, and it is the mechanic the whole model turns on.
One parameter has no citation and should be read as an assumption: attention, the share of a person available to whoever is beside them rather than to a stage, a road or a rite. It is the variable that separates a street crossing from a dancefloor at similar density, and it is the one this model asserts rather than borrows.