How weak can the preference for same-color neighbors be and still sort the board? At the default similarity-threshold of 35, the salt-and-pepper mix coarsens into single-color clumps a few patches across and stops changing within about 15 ticks.
Try this While it runs, raise similarity-threshold from 35 to 75 (it is read every tick, so no Reset is needed). The board churns for about 200 ticks, then separates into large blue and orange regions with the empty patches lined up along the borders. At 76 it never settles, because a household with 8 occupied neighbors then needs 7 of them to match.
About this model Where it ends up
Thomas Schelling's 1971 model of residential segregation, on the grid used by Uri Wilensky's NetLogo Segregation model. Households that are content to be a local minority still sort a random mix into single-color regions.
Each patch is empty, blue or orange. Each tick, every household counts the occupied patches among its 8 neighbors (the board wraps at the edges) and is satisfied if at least similarity-threshold percent of them share its color, or if it has no occupied neighbors. Then, in random order, each unsatisfied household moves to an empty patch picked uniformly from the whole board, with no check that it will be satisfied there. density and blue-share are read only at setup; similarity-threshold is read every tick.
Two differences from the NetLogo version: there, an unhappy household makes random jumps of up to 10 patches until it lands on an empty one, and the run stops once everyone is happy. Here any empty patch is equally likely, and the run keeps ticking after the board has stopped changing.
