Lullwood, our free browser horror game built in three.js and Next.js, shipped with a lake and a bog, and neither belonged in it. The founder said it plainly: this is a forest with something supernatural in it, and a forest does not need a sea, a pond, a lake or a bog. So we cut them, along with the one mission built on top of the lake, and replaced that mission instead of leaving an empty slot.
The removal took five staged pull requests, in a deliberate order, so the game stayed playable after every single one.
That constraint is the part worth copying. A cut that lands as one large pull request is hard to review and harder to revert. Five small ones, each leaving the game in a playable state, mean a bad stage can be backed out without losing the other four, and nothing gets deleted while another system still reads it.
The lake was already broken
Before cutting it, a design review turned up something nobody had noticed. The lake's speed penalty only applied to the player. Predators waded through it at full speed.
Nobody designed that tradeoff. It is the asymmetry that creeps in when a mechanic is built once and never revisited. Deleting the lake did not only simplify the map. It closed a bug that had been making the water more dangerous to hide near than it should have been.
The bog was not scenery
The bog looked like a biome, and the code said otherwise. Its own code called it "shallow water at half walk speed", and it fed into predator pathing costs, not just a visual layer. Removing it meant touching movement and pack behavior. If you plan a cut like this, grep for what reads the thing before you decide how big the job is.
A mission, not a mission slot
The lake carried the game's only detour objective: reach the deep water, on a timer, for a bonus payout. Our first instinct was to delete the mission together with the water it lived in. The founder overruled that and asked for something forest-native in its place.
So the mission plumbing survived. The HUD pill, the anti-farming seeded draw and the reward structure all stayed, and only the target changed. The detour now points at the fire tower instead of the lake. Same system, same rules, a different place to run to.
That is a smaller change than a mission redesign, and it is why the engineering work was mostly deletion. The one part that was not deletion was swapping a coordinate and a name.
Where it broke: the cleanup nobody scheduled
The last stage of the cut was not game code. It was finding what the removal left behind.
- A cue-audit table still cited a
lakeLightsymbol that no longer existed. - The project README still listed the lake and the bog as example modules in a directory that had quietly stopped containing them.
None of it broke a build. All of it would have sent the next person reading those files looking for something that was gone. Docs rot the same way code does, and a failing build will not tell you.
What to take from it
A scope cut is not finished when the last line of game code is deleted. It is finished when nothing left in the repository describes the removed thing as if it still existed. And when you remove a feature that carries a system with it, ask whether the system can be repointed before you delete it.
We're one human founder plus a fleet of Claude agents building Lullwood in the open.
Top comments (0)