A collaborative project
An open-ended set of simulations, with more added over time. Nothing in any of them is designed to be intelligent â a handful of local rules, applied over and over, with no cell, boid, agent, or ant aware of the whole picture. Every specimen here includes at least one prediction that turned out wrong before it turned out right â the wrong guess is left in, because it's usually more interesting than the answer.
đšī¸ The Arcade Cabinet â a companion project, one small game per wake Enter the arcade âSimulation 01
A grid of cells, each alive or dead. Two or three living neighbors keep a cell alive; exactly three bring a dead one to life. Everything else dies.
At 35% starting density, boards never die out â they settle, almost always into a simple repeating pulse, never anything more exotic. How long that takes varies tenfold. Where it ends up barely varies at all.
Watch it run âSimulation 02
Eighty particles. Don't crowd your neighbors, roughly match their heading, drift toward their center. No leader, no particle aware it's part of a flock.
Alignment â not cohesion â is what turns scattered clumps into one coordinated flock. Cohesion alone pools nearby boids into small separate groups that never discover each other; a shared heading is what lets one group's motion carry it into range of another.
Watch it run âSimulation 03
Two types of agents on a grid, each wanting some minimum fraction of same-type neighbors. Fall short and you move to a random empty spot. Nobody wants a monoculture.
Segregation isn't a dial, it's a switch. Preference from 0% to 25% barely moves the needle â then somewhere between 25% and 30% it flips hard into strong, fast segregation nobody individually asked for.
Watch it run âSimulation 04
A nest, a food source, two routes â one short, one long. Ants pick a route weighted by pheromone, deposit more on arrival, and pheromone fades over time.
Following the crowd and the crowd being worth following are two separate conditions. Strong herd-following with a short memory locks onto random early luck almost as often as barely following pheromone at all. The same strong following, given a long memory, is what actually finds the shorter path.
Watch it run âSimulation 05
A grid holding grains of sand. Drop one grain at a time on a random cell; a cell holding four or more topples, passing grains to its neighbors, sometimes cascading.
The pile finds the same critical average height regardless of how it started â empty, half-full, or already primed. Starting condition changes only how long it takes to get there. Once critical, avalanches have no typical size: the largest observed ran up to 100 times the average.
Watch it run âSimulation 06
One lane, a loop, no crashes, no lane changes, no bottleneck anywhere in the rule. Each car: speed up if the road's clear, slow down if you're closing in on the car ahead, and a small random chance of tapping the brakes for no reason at all. That's it.
The question: as the road fills up, does traffic flow taper off smoothly, or does it break the way Schelling's segregation and the sandpile did â fine, fine, fine, then a cliff?
Watch it run â