How it is scored

The rules behind the Owner's Audit, written down so you can check them. They are fixed and the same for every owner. If they change, this page and its date change.

What is asked

The layers

Every question belongs to one layer. The layers are a ladder: direction, processes, information and numbers, roles and handover, automation. Under all of them runs a safety floor, and around all of them the improvement loop.

RuleHow
Layer score100 × (1 − mean answer / 3) over the questions that apply. Broken below 40, Weak below 60, Holding below 80, Strong at 80 and above.
Keystone floorEight questions are keystones (for example: what happens if you disappear for 30 days; whether there is a plan for illness). If one of them has the worst answer, its layer is capped at Weak whatever the mean, and the question is named. The mean cannot hide one catastrophic answer.
What comes firstThe lowest rung that is Broken or Weak; if none, the lowest that is Holding. Direction is always the first page, whatever its score, because it decides which processes are worth building.
In the free auditLayers are scored from the 20 stage-1 questions, not all 69. Automation is not measured. The read-out says so.

The levers

Eight levers group the questions by the move that fixes them (make it safe, write it down, count what matters, hand it over, make it pay, step out of small decisions, demand that does not need you, choose the direction). Lever pressure = 100 × mean of its answers / 3. A lever is in play if it has one answer at the worst, or two at the second worst. The three with the highest pressure are shown, by tier and then by pressure. In the program, four at most, with what each needs first.

The six numbers and the contradictions

Six numbers are asked before any opinion question: days without you, hours last week, share of the biggest customer, revenue that repeats, months of cash, people who can price a job. They are not scored. They are shown beside the scores, and again at day 85. When a number and an answer do not agree (for example "runs 30 days without me" and 12 days as the number), the read-out says so, with two possible readings and a one-week test. It never decides which one is true.

What we do not do

What is AI and what is not

Sources

The question-behaviour effect: Wood, Conner and colleagues, "The impact of asking intention or self-prediction questions on subsequent behavior: a meta-analysis", Personality and Social Psychology Review, 2016. A small but reliable effect; we say so and leave it there.

Rules version: question bank v2, layers v2, levers v2, numbers v1, contradictions v1 (4 November 2026). Page dated 5 November 2026.