The collapse that wasn’t about tuning: where the world puts its first animals
A routine repair to a long-running population probe turned up something nobody had gone looking for: at the settings we’d shipped, a brand-new world’s animals collapsed to almost nothing within a fraction of the years the design needs them to survive. The first read was a tuning mistake. Finding the real one took two more days, two rounds of measurement, and a date nobody had thought to check.
A probe repaired to tell the truth
The tool we use to soak-test a world over the long haul had quietly been running with disease turned off and the soil’s depletion floor disabled, dodging two of the game’s real pressures on a population without anyone noticing. Fixing the probe to match what actually ships was meant to be a small correction. Instead it uncovered the trajectory that had been hiding underneath: one seed opened with 670 creatures, fell to 122 by day 12, to 18 by day 38, and to 5 by day 51, while the plant count kept climbing the entire time, from 3,238 to over 50,000. The design needs a world to hold together for roughly 351 in-game days. This one was dying by day 51 — about 14 percent of the way there.
The knobs that weren’t it
The instinct was to treat this as tuning: some rate had drifted, and turning it back would fix it. So we swept the obvious levers one at a time. A wider breeding search radius didn’t save species from dying out. Better breeding odds didn’t either. Cutting predation didn’t either. Letting animals roam farther, so a world would feel less crowded, made things worse instead — a bigger range pulled newly arrived animals away from the small groups that let them find each other at all.
We’d swept parameters this way before, on a hundred worlds at once, back when there were only four species to balance. This time the sweep pointed away from every rate we tried and toward the map itself. Widening the playable area, without adding a single animal or touching a single rate, kept far more species alive: 42 of 65 over twenty in-game days, against 25 of 65 in the unmodified world, with fewer total animals on the ground than any attempt that just added more of them. The ability to grow a world’s usable area without growing its population had been built weeks earlier, for an unrelated reason; here is where it turned out to matter.
Whatever was killing off species responded to space, not to how many animals were on the ground.
Why more species meant fewer neighbors
The reason space mattered traces back through months of growth, not one afternoon of tuning. Every animal looks for a breeding partner the same way: a search within a fixed radius around itself, for another adult of its own kind. That radius never moved. What changed, slowly, was how many species were sharing the world’s animals. Early on, with four species written, each averaged around 167 animals, and an adult had roughly a 95 percent chance of finding a partner somewhere in range. By the time the roster reached 65 species, the average had fallen to about 10 animals per species, and the odds of finding a partner had fallen with it, to about 17 percent. Nobody touched a breeding number to get there. The roster simply grew, update after update, while the radius held still.
Sealed off before the world even starts
Chasing the space explanation further turned up the sharper fault underneath it. A direct count of where founding animals land showed that a third to nearly half of every starting species’ roster — the exact share moved with the random seed — had zero breedable partner within range from the moment the world existed, before a single day had passed. Across all the animals in a starting world, about three in ten individuals were isolated from birth.
A world’s founders are placed once, when the world is born, and mostly never move again after that. For a meaningful share of species, the outcome was decided the instant the world was created — before any death, disease, or hunger ever touched them.
Clustering the founders
The fix that followed didn’t touch a single death, breeding, or predation rate. It changed how a new world scatters its founding creatures in the first place: animals of the same species now land clustered near each other, and every species is guaranteed a minimum starting group instead of however few the random scatter happened to produce. Tested across multiple seeds, the species-locked-out-from-day-one problem didn’t turn up again on any of them.
Population followed the fix on every test world, even though roughly eight in ten deaths there still came from predation, exactly as before. One scenario’s low point of 309 creatures recovered to 476 and climbing; another’s low of 279 climbed to 1,018, past where it had started.
A verdict that needed a new date
The day after the placement fix shipped, a routine review caught something the investigation itself had missed: the original day-51 collapse table, the one that had started all of this, had been measured the day before the fix landed and never rerun. Measuring again, after the fix, told a different story. Both retested worlds recovered instead of collapsing.
What hadn’t changed was a quieter number sitting underneath the first: species count was still declining, slowly, even as total population climbed. Extra animals were piling into whichever species were already going to make it, while the rarer ones kept thinning regardless — a rising headcount that doesn’t mean a healthy roster.
What changed was the diagnosis, not the verdict.
The fix that shipped that Monday changed nothing about how an animal dies. It changed where the world puts its animals on the day it’s born, and that was enough to turn a collapse that looked total into a species count that’s still worth watching.
