Devlog
The systems, the measurements, and the times a number turned out to be wrong. Written as we go, including the parts that did not work.
Follow by RSS, or leave an address below and get one email when the game is out.
No. 11
We moved the whole game somewhere else. The renderer needed a reskin, the economy needed nothing at all, and the only thing that really broke was the website — which spent four days describing a place the game no longer had.
No. 10
The harness worked, the runs completed, the numbers looked plausible, and a whole column of results turned out to be counting something other than what it was labelled. Here is what survived the correction, and why.
No. 09
Our release build completed successfully and produced an app with the entire economy missing. Then the check we wrote to catch that skipped, in exactly the state it existed to catch. Both were found the same way.
No. 08
A test asserted that our delete-account button was disabled, and it passed while the code behind the button would happily delete an account on an empty box. Here is the shape of that mistake, and the nine mutations we used to find the rest of them.
No. 07
A check that cannot come back negative is decorative. We have a rule about that, we enforce it on the game, and this week it caught me three times in four days on the website instead.
No. 06
Every printer needs a clear tile in front of it. Until this week "front" was the same fixed direction for every machine in the shop, which made clearance a tax you paid rather than a space you could plan.
No. 05
Every safety check in this project has been watched failing on a real problem before its passing result was believed, because a check that cannot come back negative is indistinguishable from one that works.
No. 04
There was a stretch of the game with nothing meaningful to buy. Four different fillers were built and measured, and all four failed in exactly the same way — by making the run longer instead of fuller.
No. 03
A single corrupt value could destroy a save, and every safety net we had built failed at once — silently, and in a way that was invisible by reading the code.
No. 02
Pinch-to-zoom ran straight into a physical constraint rather than a taste question — crisp pixels and fluid scaling are, at the naive level, mutually exclusive. The fix was to stop treating it as one decision.
No. 01
A probe that plays the real simulation as a flawless player turned out to be the most useful thing in the repository, because it measures the one thing a developer cannot feel.