The model. Mark Granovetter's threshold model (1978) gives every person a threshold: the share of the room that must already be standing before they join. A few instigators have threshold zero. In each round everyone whose threshold is at or below the current standing share stands up (people who stand stay standing). If thresholds are normally distributed with mean μ and spread σ, the share standing next round is F(r) = f0 + (1 − f0) Φ((r − μ)/σ), and the cascade ends at the first fixed point r = F(r) at or above f0.
The lesson. Two rooms with the same average threshold can end very differently: a wider spread puts a few people with low thresholds in the room, which can start a chain reaction, while a narrow spread leaves a gap that nobody bridges. Small changes near a tipping point change the outcome a lot. This is why the same crowd can behave "all or nothing".
Simplifications and honesty. This is a toy model of social influence, not a measurement of any real audience: thresholds are invented draws from a normal distribution, everyone sees the whole room equally, nobody sits back down and people have no friendships or seating effects. It is related to, but is not, the Asch conformity experiment (which measured individual responses to a unanimous group). A room of 144 is small, so randomness matters; the exact formula describes the limit of a very large room.
Checks. (1) The formula's fixed point found by iterating F is compared with the first crossing found by scanning F(r) − r with bisection. (2) A separate agent-based simulation of 100,000 sampled thresholds, run round by round, is compared with the formula's final share.
The theatre seats, people, stage and carpet were modelled and baked in Blender; the agent model and rendering run in Three.js r160.