We had a question: when is a puzzle game too hard? The usual answer — “when players drop off” — is too slow and too lossy. People rage-quit quietly. They don’t fill out forms.
So we built a little tool. It sits on the test build’s telemetry pipeline and watches for a single signal: keypress frequency on the delete key within 500ms of a failed signal sync. That’s it. That’s the whole thing.
We call it the One Bad Word Detector, because that’s the exact moment the player has just said one.
What it told us
Puzzles 1 through 14 are fine. Puzzle 15 is not. Puzzle 15 was supposed to introduce the splitter, and we’d layered it on top of a board that also had its first forced capacitor interaction, and playtesters were hitting the delete key at a rate that would’ve alerted a small village.
We split puzzle 15 into two puzzles. The word rate dropped. People kept playing. The capacitor still has opinions, but now they’re introduced kindly.
Our guiding principle is no pressure. The one bad word detector is how we audit that principle.
We will probably open-source the detector at some point. It’s 80 lines of code and one very judgmental dashboard.