The one number that was secretly a cliff
A number can be right for months and then, with nothing touching it, become catastrophically wrong. This is the story of one such number: the dial that governs how hard overcrowding presses on a population. It never changed. The world underneath it did — and the day the ground shifted, a gentle brake quietly became a cliff that wiped out whole herds overnight.
The cliff turned out to be the single hidden cause of three problems we’d been blaming on three different systems. Walking it back to a brake fixed all three at once, with one edit and no other lever moved. That is the kind of bug worth writing down — not because the fix was clever, but because the failure mode is one any long-lived system can fall into without anyone making a mistake.
A number that was correct for the world it was born in
Some background, kept narrow. When a patch of ground holds too many of one kind of creature, that creature grows a little likelier to die there — crowding routed straight into the odds of dying. Why we settled on that mechanism rather than a hunger-keyed brake, and how it reaches the off-screen majority where a population actually runs away, is its own story; here we only need one fact about it. The strength of that pressure is a single tunable number, and the pressure it produces rises in proportion to how far a population sits above the ceiling its land can support, multiplied by that one strength dial.
When the feature first went live, we hand-tuned that dial against a fixed notion of how many creatures a place could hold — one ceiling, a few hundred, shared across the whole world. Against that ceiling the number was exactly right. When a herd ran past its limit, the pressure shaved off the excess and the population settled back down to where the land could keep it. Everything coexisted: rabbits common, ravens fewer, foxes rare, birds of prey rarer still. The number worked because it had been fitted to that one ceiling, and the ceiling didn’t move.
Then the ground changed shape
Later we changed how the ceiling itself is decided. Instead of one authored cap for everywhere, each species’ limit is now read live off the land — the forage a patch grows, the fertility beneath it, the biome, the animal’s body and way of living. A grazer in the country it’s built for gets the full measure of the ground; the same grazer in a biome it isn’t suited to finds the same patch holds only a small fraction as many. That change — capacity derived from the land rather than typed in by hand — is told from the player’s side elsewhere. What matters here is the arithmetic it left behind.
Some of those derived ceilings are tiny. A heavy grazer in a sparse, cold, or arid biome might be supportable only at a fraction of a fraction — a ceiling near zero, far below the old shared cap of a few hundred. And the crowding dial had never been re-tuned for that new reality. It was still the number we’d fitted to a ceiling of a few hundred.
The number never broke. The ground beneath it did — the moment a limit stopped being a big fixed figure and started being a possibly-tiny reading off the land.
Why a big number plus a tiny ceiling is a wall
The math is the whole of it, and it’s worth following slowly. The crowding pressure scales with how far a population sits above its ceiling, multiplied by the strength dial. With a ceiling of a few hundred, even a sizeable dial is fine — a herd a little over its cap gets a gentle nudge downward. But against a ceiling near zero, over the cap effectively means the entire herd. Multiply that overshoot by the same large dial and you get an enormous death pressure: not a nudge but a wall. Any population even slightly above its tiny cap was annihilated wholesale — the whole cohort gone within about a single in-game day, instead of being trimmed toward what the patch could actually hold.
So a grazer set onto land whose ceiling was tiny was, in effect, born dying. The same dial that read as a soft brake under the old rules had silently become a cliff under the new ones — with nobody having touched it. That is the unsettling part. There was no bad edit to find, no regression to bisect. A value that had been stabilizing for months had simply outlived the assumption it was calibrated against.
One root, three faces
What made this worth a postmortem is that the single mis-scaled dial was wearing three different masks, and we’d filed each as a separate bug against a separate system.
- Grazers that couldn’t establish anywhere harsh. Herds placed into cold, dry, and tropical land crashed straight to zero within days, leaving no standing prey base at all. We’d been reading this as a problem with where animals were being placed or how forage was set. The player-facing telling of that surprise is its own note.
- An upside-down result in the temperate test. The predator-free control group would boom — overshooting toward a couple hundred animals — and then crowding-crash so hard it ended up below the run with a predator present. Removing the predator left fewer prey than keeping it: an ecology reading exactly backwards.
- Predators that grew over an empty world. With the prey gone, the hunters kept multiplying anyway, because nothing yet tied their breeding to whether there was anything left to eat. (That gap had its own fix, told separately.)
Three symptoms, three suspects, one cause. The first two were the cliff directly — the third was unmasked once the cliff stopped erasing the prey base out from under the question. Chasing them one at a time, we’d have re-balanced placement, re-balanced forage, and re-tuned three systems, and the world would still have been falling off the same edge.
Turning the cliff back into a brake
The fix was almost embarrassingly small for the size of the problem: turn the one dial down, hard. We’d confirmed the diagnosis by sweeping the strength across a range and watching what each value did to a herd on thin ground. Across the harsh end of that range — the cliff — the herd was erased entirely, zero survivors, every time. Only at the gentle end did a population survive and settle at a sustained level matched to what its patch could actually feed. So we lowered the dial to the gentle end, and crowding went back to shaving off the over-the-limit excess instead of annihilating the cohort.
With that single change, all three faces resolved at once, and no other lever moved. A tropical grazer that had been crashing to extinction now holds steady at a modest sustained number; the cold and arid grazers find their own levels too; and the temperate result turned right-side up. As a companion change in the same close-out we also wired predator breeding to slow when the prey base collapses — so hunger finally feeds back into reproduction — which is the third face, and a story of its own. Together those two changes closed this chapter of the ecology work: the food chain is a working loop, and the prey base finally matters.
The shape of the durable lesson
The throughline is a number that is correct only relative to its assumptions. The crowding dial was fitted to a single shared ceiling of a few hundred. When we replaced that one ceiling with many environment-derived ones — some of them near zero — a value that read as stabilizing became exterminating, and the change that did it touched the ceiling, not the dial. The dial was a passenger; it had been measured against a world that no longer existed.
That is the failure mode worth carrying out of this game and into any long-lived system: a constant calibrated under one regime can quietly turn destructive when the regime shifts beneath it, with no edit to the constant and no error anywhere to catch. The defense isn’t vigilance — nobody re-audits a number that’s worked for months. The defense is to notice, when you move the ground, which old numbers were standing on it.
We left ourselves a note for next time, too. The gentle-brake margin we landed on is tight, and a sturdier shape exists: reshape the crowding pressure so it scales with each patch’s ceiling automatically, so a tiny cap can never multiply a big dial into a wall in the first place. That removes the cliff by construction rather than by careful tuning — a constant that can’t become a cliff no matter how small the capacity gets. We’ve parked it for when it’s needed. For now, the simplest possible fix — one number, walked back from a cliff to a brake — let the herds hold their ground.
