49% nothing, 51% everything

A forest on a grid. Fire is set along the top edge and can jump from a tree to each neighbour with a fixed chance. You would expect the fire to get a little further as that chance goes up. It does not work like that.

The rules, all of them. Every pair of neighbouring trees is given, once and independently, a gap the fire can cross (with the chance on the slider) or one it cannot. The top row is lit. Fire spreads across every crossable gap. The question is only whether it reaches the bottom row.

On an endless grid the threshold is exactly one half (proved by Harry Kesten in 1980). Below it every fire is finite; above it a fire can run for ever. On this page's grid, one tree taller than it is wide, the chance of crossing at exactly 50% is exactly 50% whatever the size, which the experiments check. The bigger the forest, the more the curve looks like a switch: at 400 trees across it crosses about 0% of the time at 48% and 100% of the time at 52%. The curve on the page is measured live from repeated forests of the size you choose. What it leaves out: wind, damp, clearings, firefighters. Real forests are not grids.