Teaching the world to crowd: a balance that reaches where you aren’t looking
For a long time the world could only crash or boom, never rest — and the reason was not where we kept looking for it. The fix that finally held is almost dull to state: when a patch of ground gets too crowded for a kind of creature, that creature grows a little likelier to die there. The interesting part is why that plain idea settled a world that cleverer ones couldn’t. It comes down to a single structural fact about how a living world is simulated.
Most of the world is somewhere you aren’t
To run a world full of creatures on modest hardware, we don’t simulate all of them the same way. Only the handful near the camera get the full, expensive, moment‑to‑moment treatment — getting hungry, seeking food, moving, eating. The vast majority live far off‑screen as cheap statistics, updated coarsely and rarely. We built that split earlier so the world could be enormous and still run; here it matters for a reason we didn’t fully reckon with at the time.
The catch is buried in the word coarsely. An off‑screen creature isn’t hunting for food, because simulating the search would cost what the tier exists to save. Its hunger is held roughly flat by the cheap update — the world simply assumes it fed eventually — so hunger never really builds out there, and never bites. The watched few starve; the unwatched many don’t. That asymmetry is invisible most of the time. It became the whole problem the moment we tried to balance the world through hunger.
Why a hunger‑keyed lever can’t hold a world
The intuitive way to stop a population running away is to make crowded animals breed less: when food gets scarce, slow down the births. We chased exactly that idea, and it could not settle the world. The reason isn’t that it’s a bad idea in the abstract — it’s that it acts through hunger, and hunger only exists, in any meaningful way, for the creatures near the camera.
So a hunger‑keyed brake reaches the small on‑screen population and leaves the great off‑screen multitude untouched. And the off‑screen bulk is precisely where a population runs away — it is most of the world by a wide margin, and it was breeding at full tilt no matter how scarce things got near the player. We were tuning a lever that pulled on a few hundred creatures while the thousands behind us climbed unchecked. No setting of that lever was ever going to work, because the lever could not reach the part of the world that was overflowing.
We did the honest thing and shipped that mechanism switched off rather than pretend it was solved — one piece of a longer story about why deferring a balance can be the right call. But the dead end taught us the real shape of the problem, and the shape pointed straight at the fix.
A balance is only real if it reaches the part of the world you’re not watching. The most elegant rule is worthless if it only governs the creatures on screen.
Route the pressure into death, not hunger
If the trouble is that hunger never reaches the unwatched many, then stop going through hunger. The fix routes the crowding pressure straight into the odds of dying. When too many of a kind pack into one patch of ground, each one there simply grows a little likelier to die — the world’s own carrying‑capacity pressure, applied directly.
The reason this reaches everywhere is the quiet center of the whole thing. Crowding doesn’t care where you’re looking. It needs nothing but a count: how many of a kind are gathered in a place. And that is a fact the world always knows — as cheaply for the unseen thousand as for the creature standing beside you. Hunger has to be simulated, and that’s what made it expensive to track off‑screen and cheap to skip. A headcount per patch is already there, near and far alike. So a single steady dial reaches every creature at once, no matter where you happen to be looking.
It slots in as just another named cause of death. A creature in this world doesn’t die on a timer — it wears out as pressures accumulate, and when one of them wins, it dies of something you could name. Crowding takes its place on that same scale, sitting between being hunted and going hungry, ordered so the most acute thing claims the death first. Each kind of creature carries its own sense of how crowded is too crowded: grazers can pack in close before it bites; the hunter at the top needs a great deal of room before the world ever leans on it.
And the dial is gentle by design. Below a comfortable density it does exactly nothing — no penalty at all. Only as a patch overshoots does the extra risk appear, and it climbs slowly from there, the more crowded the steeper. There’s no wall to slam into, no all‑or‑nothing cliff like the one the old approach kept tipping over. A place self‑limits before it can run away, and it does so out of sight just as surely as in view.
The dial that runs backwards
Once the pressure could reach everywhere, one more surprise was waiting — and it’s the kind of thing only a food web does. You’d expect that turning the shared crowding dial up would leave fewer grazers. Past a point, it leaves more.
The logic is worth following slowly, because it’s genuinely counter‑intuitive. A stronger crowding pressure presses on every kind of creature, predators included. But the predators lean on the grazers for their living, so culling the predators harder than the grazers loosens the grip on the grazers from above. With fewer hunters bearing down, the prey are released to climb — even as the same dial is, in theory, supposed to be holding them down. The pressure that thins them directly also thins the thing that was thinning them, and past a certain strength the second effect wins.
So the grazers’ sweet spot isn’t the highest setting. It’s a middle one — strong enough to hold them, not so strong it strips out the predators that also hold them. We found that middle empirically rather than by reasoning our way to it; the method — sweeping many candidate worlds in parallel and reading the outcomes side by side — is its own story. What belongs here is the mechanism beneath the result: a single dial whose effect on one creature runs through every creature it’s tangled up with, which is exactly why an ecology can’t be tuned one species at a time.
Where it left the world
With the dial settled, breeding went back on for good, and four kinds of creature now persist together with no hand on the scale: many grazers, fewer scavengers, a scarce hunter on the ground, and scarcer still a hunter above it. What that settled world feels like to watch over is told from the other side, and the call to ship it in the shape it actually took — not the one we’d sketched — is its own note; the engineering claim here is narrower and, to us, the more durable one.
A balance held across two or three in‑game years — and an in‑game year here is about a hundred and ninety days — is not a balance we’re standing over, topping up by hand. It is a property of the world: a count per patch, a gentle rising risk, a comfortable density for each kind of life, and nothing else. That it reaches the unwatched corners is not a nice extra. It is the entire reason it works where the cleverer ideas didn’t.
The rule that holds a world isn’t the cleverest one. It’s the one whose reach doesn’t stop at the edge of what you can see.
Which is, in the end, the same bargain the whole game makes. You don’t win; you tend. And a world is only worth tending if it can hold itself together where no one is looking.
