The floor that was hiding: the second zero underneath the first
Grazing in The Long Watch was meant to wear the ground down, the way real overgrazing does — feed a patch too hard, too often, and the land itself should show it. For months, turning that setting up did nothing at all, at any strength. The reason turned out to be a second number, one nobody had touched, undoing the first before it could ever register as a change.
A ledger separate from the plant
The world already knew how to feed itself before grazing arrived. Something dies, it breaks down, and the ground where it fell reads a little richer for a while; a plant checks that richness the same way it checks light and warmth before it grows. Grazing landed on top of that loop with its own idea about the ground: an animal that feeds somewhere long enough should leave the soil worse off, not just the plant it ate from.
That is a different account than the one the first bite already covers. A plant’s own forage — how much of itself it has to give, per bite, before it needs time to regrow — has worked since rabbits first learned to eat. The soil underneath it is a separate number entirely, and the setting meant to drain that one shipped at zero. It sat there for months doing nothing, because nothing in the game ever asked it to do anything else.
A dial connected to nothing
In August the team went back to finish the other half of that loop. First came a way to actually see soil condition on screen: the ground itself darkens where fertility runs rich, and pales where it has been worn thin, both driven off the same fertility number the plants already read. Then came the part that looked like the easy half: turning the drain on.
It did nothing. Not a little; nothing. The setting went from barely-there to aggressive across a hundred-fold range, and the world came back looking exactly the same regardless, no more surprising, in the moment, than the fact that predators alone were already killing off more of the population than grazing ever touched.
The unsettling part wasn’t the null result. It was how easily that result could have passed as correct. Test the drain on a small, freshly made patch of ground, the kind of tidy scenario a quick check reaches for, and the whole depleting side of the math is arithmetically invisible: there is almost nothing built up yet to remove. Run the same setting in a real, lived-in world instead, one with time to accumulate fertility, and the same clamp is doing real work the entire time. A shallow test and a real one can report opposite stories about the identical number.
The floor beneath the drain
The actual culprit was a second value: a floor, a limit on how far fertility is allowed to fall no matter how hard the drain pulls. It had shipped at zero, too, meaning nothing that reached that limit could go a step further, so any attempt to drain the soil got clamped straight back to where it started. The floor moves with the land, besides: it’s set at half of a related threshold, a threshold that itself shrinks the poorer a patch of ground already is.
The floor is what makes it exist.
The floor, not the drain rate, was the number that actually needed to move.
Setting both numbers at once
The fix moved two values together: the drain rate up from zero to a small, steady pull, and the floor down from zero to a real depth soil could actually be pushed toward. The color scale reading fertility on the ground changed with them, too. It used to swing evenly in both directions; now it doesn’t, because depleted and rich soil don’t shift their visible tone at the same rate, so the range widened unevenly to keep matching what the eye actually sees. Arm all three together, and grazing finally has somewhere to spend itself: a patch fed too hard now carries a real, lasting change in what the ground itself reads as, easing back only over a real stretch of time.
The same bug, hiding twice
A few days later, the team was repairing an unrelated tool: a way to soak-test a world over a long, unattended stretch and see how its populations actually hold up. That repair turned up a second copy of the same mistake. The test scenario had its own hardcoded stand-in for the soil floor, fixed at zero and overriding whatever the real setting said, every single time that scenario ran. Every long-run population test up to that point had been measuring a world with no soil floor at all, no matter what the shipped game actually carried. What that exposed about those populations belongs to a different post; what belongs here is just how the same shaped mistake — a number the game trusted, zeroed out somewhere it could not see — turned up twice in one week, wearing two different disguises.
Two floors, two zeros, two silent overrides inside of a week is not a coincidence worth shrugging off. It is what a hidden default actually costs: not a crash, not an error, nothing that ever turns a check red. Just the exact absence of the thing you built, sitting there, passing every glance you happened to give it.
