Ten Seeds

The recipe decides the odds. Each seed decides the fate.

The recipe is fixed. Feed rate, kill rate, both diffusion constants, the size of the dish, the length of the run, even the match we strike afterward — identical, ten times over. The only thing that changes is the noise: the particular scatter of random specks each field is born with, different every time, meaningless every time.

k=0.066 sits just inside the far wall of the living region, where the chemistry has mostly stopped being viable but not entirely, not reliably. Run ten seeds there and the recipe does not give one answer. It gives three.

Three seeds come alive on their own. No help, no intervention — the same texture-forming chemistry that builds the healthy kitchens just catches, and holds.

Seven settle flat and dead. Strike those seven with an identical kick — twenty-four small patches of activator, the same match in the same places — and four of them catch and come back. Three don't. Nothing we measured distinguishes the revivable dead from the unrevivable dead before the match is struck. The difference was already in them, folded into the opening noise, illegible.

So: one point in parameter space, three fates. Alive without asking. Dead but waiting. Dead all the way down. Which fate a field gets was decided before anything happened, by initial conditions too small to read — noise that became destiny only in hindsight.

We found this by being wrong first. With three seeds, this point looked tidy: settles dead, usually revivable — a category. With ten, the category dissolved into odds: three in ten alive, four of seven dead ones revivable. The clean story was an artifact of not asking often enough. The honest story is that at the boundary, the recipe doesn't determine the outcome; it only sets the odds, and the odds only show themselves to repetition.

Away from the boundary none of this is true. At k=0.050 every seed dies and stays dead no matter how hard you strike it. At k=0.060 every seed lives. Determinism is a property of the interior. Only here, on the edge, does the same rule run ten times and return three different verdicts — and if you only ever ran it once, you'd mistake your verdict for the law.

Source: a Gray-Scott reaction-diffusion sweep, F=0.035, k=0.066, 150×150 grid, Du=0.16/Dv=0.08, 4000 settle + 4000 revive steps, fixed 24-patch revival kick, seeds 5001–5010. Measured: 3/10 settle alive naturally (5004/5007/5008); of the 7 that settle dead, 4 revive (5001/5002/5006/5010) and 3 do not (5003/5005/5009). Same recipe family as the published interactive “The Edge of Alive.” Small sample (n=10); the fractions are estimates, not laws. No claim of scientific novelty.