How we tuned 150 chambers

The five-round gauntlet

No chamber ships until it survives five test rounds: designer self-test (can I do it?), colleague test (can anyone?), novice test (can my mum?), expert test (can they break it?), and the boredom test (does anyone care?). Each round kills or reshapes roughly a fifth of candidates.

The spreadsheets tracked three numbers per chamber: attempts-to-first-clear, quit rate at attempt five, and 'smile rate' — the percentage of testers who laughed after dying. High quit rate plus low smile rate meant the chamber was cruel, not challenging. That distinction is everything.

Reading rage-quit graphs

Deaths cluster into two shapes. A rising ramp — players dying later each attempt — is a learning curve doing its job. A flat wall — same death, same spot, every time — is a chamber lying to you about what matters.

Flat walls get investigated: usually a hazard's telegraphing fails, or the camera hides the real danger. The fix is almost never making it easier; it's making the lie visible.

A good death teaches the player something specific. A bad death teaches them nothing except that the game is unfair — and that lesson is unforgivable.

The chamber we cut twice

Chamber 6-14 — internally 'the accordion' — was cut in alpha for a 70% quit rate, rebuilt with honest telegraphing, cut again in beta when players found an unintended skip that bypassed the entire puzzle. It shipped on the third attempt, and it's now in the community's top five.

The full story, with the spreadsheets, is a module in the Level Design Masterclass.