NURSERY

what is happening

A cellular automaton has a rule you can write as a number: so many live neighbours, such a next state. A continuous automaton has no such number. The neighbourhood is a smooth two-dimensional weighting — a kernel, usually a ring or a stack of rings — and the rule is a curve mapping how much of a substance that kernel finds nearby onto whether there should be more of it. There is no ruleset string. There is a set of functions.

Almost every set of functions is boring. Draw one at random and the world dies, or fills, or churns forever without ever closing into a thing with an outside. What this page hunts is the fourth outcome, where the field settles into a bounded lump that maintains itself — and hunting it by hand is hopeless, because the growth width alone spans two orders of magnitude and the band inside it that works is a few thousandths wide. So the rules are bred. A rack of candidates is run headless, measured, and sorted into creatures and failures; the creatures breed, a quarter of each generation is drawn fresh so the population cannot collapse onto one disc, and one elite slot is held for each kind of behaviour so a rare mover is not out-competed by the four hundredth still body.

Keeping something puts it on the shelf and makes it a heavily favoured parent. Culling removes it from the gene pool. Neither is required — the triage fills a shelf unattended — but a person's taste is a better fitness function than any number here, and that is where it enters.

Half of the first rack is not random. A rule drawn from nothing turns out to be a creature about one time in seventy — that is the density of this space, not a fault in the search, and every prior tried against it moved the failure around and left the rate where it was. Breeding solves it completely the moment there is anything to breed from: from nothing to two thirds creatures inside three generations. So the page starts from rules the nursery has already caught, and fills the rest of the first rack with fresh draws. They are a starting population and not a floor — cull them, and the search carries on into the parts of the space nobody has been to.

what a unit is

A unit is a creature written down, and it is deliberately not a sprite. The file carries the rule, the seed integer the creature was grown from, the settled body as a small grid of values, and every number that was measured off it. Anything holding a unit can put the body back into a world and keep simulating, which is the point: a unit dropped into a life sim is still alive when it lands. It can be pushed, starved or cut in half, and it responds, because what was exported was the physics and not the picture.

Nothing in the record is a timestamp or a random id, so the same creature always writes the same file, and the file states the observation window its numbers were taken in — a speed measured over a hundred and sixty steps is not the same reading as one measured over two hundred and fifty-six. A reader needs the update rule, which is nine lines of arithmetic, and nothing else from this repository.

reading the verdicts

• diedthe seed did not take, or the world emptied
• floodedit filled the dish and lost every edge
• churnalive, but it never closed into a body
• stilla bounded body that holds its position
• pulsebounded, in place, and breathing on a period
• gliderbounded and travelling in a straight line
• roverbounded and travelling, but turning as it goes

Gliders are rare. In this rule family a body that travels usually curls, and a run of three hundred rules often finds none at all — that is the space and not a fault in the search.