Four ways to put water in a world: the one we built, and the parts we took from the rest
Before a drop of water existed anywhere in The Long Watch, we wrote four complete designs for putting water in a world and set them side by side. Not sketches: four proposals argued through to the end, each costed against the terrain we actually had. Only one of them was ever built.
The survey that came first was blunter than we expected. The world held no water of any kind. No sea level, no lakes, no rivers, no drainage, nothing pooled in any hollow: one height for every column of ground, and a single skin laid over it. Every river and lake and fish in our own design notes was a promise we had written down and never paid.
The four on the table
Each proposal answered the same question a different way. What is water, exactly, in a world made of stacked ground?
The first made water a real substance: its own volume in the world, occupying space the way ground does. It was the showy answer and by a distance the largest job. It also had to live somewhere in the world’s classification of country, and the only room for it was a ninth climate zone on a list the terrain holds at a fixed length. Adding one meant a permanent change to how every world is generated, in service of three fish.
The second set one flat water line across the whole world, then swept the flooded ground into separate connected bodies, so an enclosed pond and an ocean were different things from the start. Feasible, cheap, and clear-eyed about ponds. It carried a flaw we kept circling: it took the water line off the terrain surface as it currently stands, and that surface is being reshaped, a hair at a time, by the water that carves the ground.
The third used a single flat line as well, but treated open water as a tag laid over the one terrain we already had. No second surface. No new terrain data. Swimming arrived as a variant of walking that rides the water line and stays inside the wet. Modest, and a little disappointing to read: a fish that skims the top of a lake is not what anyone pictures when you say swimmable.
The fourth was the third with the disappointment taken out. Real swimming through the whole water column, and species carrying stated traits for walking or swimming, land or water. It also read the water line off the eroding surface, exactly like the second.
The line that settled it
Two of the four would have made the map of a world’s water depend on how long you had played it.
Erosion here is slow and real. Rain soaks in, runs downhill, and grinds the land down a hair per beat, so a valley you have tended for a long while is measurably not the valley it started as. Measure a water line against that surface and a shoreline becomes a function of the world’s age. Nothing dramatic would happen; coasts would just creep, a little, across in-game centuries. But two people who had grown the same world from the same seed and lived in it for different lengths of time would be looking at different maps.
The third design was the only one specified to read the version of the terrain’s shape that has no erosion in it and no clock: the base the world is generated from, before anything wears it down. So the shorelines are settled the moment a seed is chosen, and stay settled. That a world grows identically on every machine is a rule this whole project already lives under; what mattered in the comparison is that exactly one of the four satisfied it for nothing.
A second criterion cost nothing to satisfy and would have cost a great deal to miss. Treating open water as an overlay across the existing climate zones, rather than as a climate of its own, meant none of the existing country had to move and no saved world needed a new shape to be read back. The first design failed on exactly that point, and it is why a lake in a meadow is still a meadow with water over it.
One base, a graft, and a shelf
The comparison did not produce a winner.
The third went in as the base. Onto it we grafted the fourth’s full water-column swim and its walk-or-swim traits, because the surface-skimming fish was the one part of the third nobody was willing to defend. And we took the second’s pass for telling physically separate bodies of water apart and set it on a shelf, unbuilt, with a note about when we would need it.
Four designs went in and one came out, wearing parts of a second and holding a third in reserve.
Then everything landed switched off. The rule for what counts as open water, the overlay, the swim mode, the arithmetic that decides how many fish a stretch of water can carry, and a first three-species web of a pond plant, a fish that grazes it and a fish that eats that one: all built in a single day, and every piece dormant. The water line sat parked so far beneath the ground that no world could tell the difference.
The shelf came down the next day
The reserved part was wanted within a day of the water being switched on.
The first flooded world broke into eight physically separate ponds, lakes and a stretch of ocean, and nothing in the simulation knew they were separate. A hunter in one fed on prey in another, straight through solid ground. Teaching the water where its walls are is told in full elsewhere. What belongs here is that the failure was not a discovery. The design we did not build had described that exact problem in writing, and described the remedy, before a line of the winning one existed.
Where the write-up expected to be weakest
The comparison closed with one more useful act: it said plainly where it thought it would fail, and that was not where the argument had been.
Swimming was understood. The reproducibility question was answered. The real unknown was whether a see-through water surface, with the world visible through it and the light changing as the camera crosses under, could hold its frame rate on the actual modest laptop this game has to run on. So an opaque fallback went into the record in advance, to be taken without complaint if the frame budget failed, and then we spent a day proving the card before building either the water surface or the other feature riding the same bet. It held. The fallback was never spent.
The comparison fixed the ceiling as well, and said so out loud: still water only, and what a flat line doesn’t do logged as intended rather than delivered. What the four designs bought, for a day of writing and no code at all, was a build that sat dormant for a day, went live on a single changed number, and has not had to be taken apart since.
