The day the first bare world ran smooth
The very first thing The Long Watch ever drew was nothing much: a patch of bare voxel ground, no plants, no creatures, no weather. Just the floor of a world. And it ran badly — the view ground to a crawl over an empty field. The fix turned out to be the first real decision the whole project would lean on: spend detail only where it’s seen.
Even an empty field was too much
It’s easy to assume an empty world is a cheap world. There’s nothing in it. But the ground itself is enormous — one continuous surface that stretches from under the camera out to the edge of what you can see — and the naive way to draw it is to render every part of it at full crispness, the distant tiles as sharply as the dirt at your feet.
That is exactly what the first attempt did, and it buckled. Drawing the whole terrain at one uniform level of detail out to its full reach meant the graphics card was handed hundreds of separate chunks to draw every single frame, each one a fixed cost it had to pay whether you could make out the difference or not. The per-frame work blew well past the budget for smooth motion, and a flat, lifeless field stuttered. The lesson landed before there was anything interesting to look at: the bottleneck wasn’t complexity in the world. It was paying full price for ground nobody could resolve.
Draw the near ground in full, the far ground coarsely
The fix was a terrain that doesn’t treat all of itself equally. Right under the camera it draws in full detail, every contour crisp. The farther a patch of ground sits from you, the coarser it’s drawn — simplified in steps, until the long horizon is rendered very roughly indeed. Several distinct tiers of coarseness, the detail dropping down a notch each time the ground crosses a set distance from where you’re standing.
The reason this works is just how eyes work. At a distance you genuinely cannot tell the difference between crisp geometry and a coarse approximation of it — the detail is below what the picture can show you anyway. So we stopped sending it. With the ground at the camera holding maybe a quarter to half a metre of real-world size per voxel, and detail stepping down past the near band, the far ground costs a small fraction of what it did, and the field that used to grind held a smooth sixty frames a second. The draw count dropped sharply because most of the world was no longer drawn at full fidelity for no visible gain.
The first thing the project ever decided was not what to build. It was what it could afford to skip.
Trusting a brief that was right all along
The quietly satisfying part is that none of this was a surprise to the design. The original brief had called for exactly this kind of distance-based simplification from the start — full detail near, coarsening with range. The naive single-detail approach was the tempting thing, the shortcut that looks like less work; the design had already said it was wrong for a world this size. Measuring the stutter, and then measuring the fix, simply proved the brief right on first contact.
That — choosing to trust a written-down design decision once the numbers back it, rather than reaching for whatever’s quickest — became a discipline the project leaned on far beyond terrain. The broader habit of proving an assumption by hand before building on it is its own story; here it’s enough to say the very first place it paid off was the ground.
One setting for how far the world is drawn
Getting the far ground cheap on one screen wasn’t the whole win. The bigger one was making that cheapness travel. We wrote a single number into the world’s generation data — how far out the world should be drawn at all, a reach of a couple of hundred voxels — and made every screen that shows the terrain reuse that same number rather than each picking its own.
So the slow orbiting flyover you watch before you settle into a world, and the in-game camera you tend the world through, both draw the ground to the same horizon and coarsen it the same way out there. The far ground is cheap everywhere it appears, because there’s one definition of “far” and everything obeys it. Change the reach once, and every view of the world changes with it. There was no separate tuning to drift apart.
The same idea, grown up
What makes this small foundation moment worth telling is that it wasn’t really about terrain. It was the first time the project committed to a single principle: render what’s actually seen, and quietly cheapen what isn’t. Spend the machine’s effort where a person is looking, and almost nothing where they aren’t.
That instinct, born on an empty field, is the same one that shows up everywhere later as the world filled in. When the hillsides finally grew real trees, we drew a whole forest for almost nothing by describing one tree shape and stamping it in hundreds of places rather than treating each tree as its own expensive thing — told here. And once the world held creatures, we tiered it the same way: the animals near your view get the full, faithful simulation, while the rest get a cheap update that keeps them honest until you look back — told here. Different surfaces, one idea.
All three are the same bargain, and the terrain made it first — before a blade of grass existed, before anything was alive to spend detail on. A bare world that ran smooth wasn’t a milestone anyone would screenshot. But it was the bedrock the rest of the world got to grow on, and the first proof that you can make a place feel vast and continuous while only ever paying for the part of it you’re close to.
