By Ryan Richardson · Published 8 October 2026
Setting the total you're prepared to lose, and a ceiling a machine cannot raise, is step 56 (Part IX), marked moderate difficulty, almost nobody does this. The hard part isn't choosing a number. It's choosing it before the decision gets personal.
A partnership once built an AI system for analysing a specific sport's technique from video. It demoed beautifully, the way these things always demo beautifully, and topped out at about 50% accuracy in production, after real money and two years of work. The kill happened on a number written down before any money moved, before anyone had started caring about the answer. It stung, and the kill happened anyway: the whole point of setting a number while indifferent is obeying it once that indifference is gone, because that's the only moment the number is worth anything.
In an ad account, the same reflex runs smaller and weekly: a loss ceiling set while nothing is running, enforced by a rule rather than relitigated in the moment. Four numbers get written down before any build gets a dollar, and none of them get revised once the project starts: a validation bar (most people asked have to say they'd pay before building starts), a hard ceiling on cost per customer against what that customer is worth, dated milestones instead of a general sense of progress, and a fixed count of failed fixes on the same bug that means the design is wrong rather than the patch.
A written loss ceiling, set and dated before a campaign or build begins, enforced by an automated rule rather than a daily judgement call, with the four pre-commitment numbers attached to it.
Relitigating the ceiling once a project feels close to working, which is the exact moment the number exists to override. Setting a ceiling with no enforcement mechanism, so it functions as a guideline rather than a rule.
| Claim | Value | Source |
|---|---|---|
| Partner-built product outcome | topped out at about 50% accuracy in production after two years of work, killed on a pre-written number | The Sixty Steps manuscript |
| Four pre-commitment numbers before any build | validation bar, cost ceiling vs customer value, dated milestones, fixed count of failed fixes | The Sixty Steps manuscript |
| Matrix status for this step | Moderate difficulty, almost nobody does this, automated tooling | The Sixty Steps matrix |