- AI assistants draft
- I use AI assistants, Claude above all, to draft pages and to write the site's code. A draft is not published because a model wrote it fluently.
- Claims about me
- A story about a client, an outcome or a number that is not on my track record does not go up. A page that describes a method describes a method, not something that happened to a company I will not name.
- Numbers
- A statistic appears with a link to where it comes from. If I could not find the source, the number is not there. A worked example says it is an example and shows the arithmetic.
- Charts
- The plan chart on the cost page draws only what this site already publishes: the 30, 60 and 90 day structure, the description of each price and the steps of a venture sprint. Each bar carries the sentence it comes from, and the build stops if that sentence is no longer on the page that publishes it.
- The bad reviews
- Every tool entry in the glossary links the low-star reviews a buyer would find, on Trustpilot, Capterra and sites like them, says what each one claims, and says whether the claim holds against the vendor's own pages. Where nobody has reviewed a product yet, the entry says so, including for the product I founded. The build stops if an entry loses that section.
- Checked before every deploy
- The build stops if a post claims experience that is not on the approved list, repeats text from another post, cites a source it does not list, leaves a model's stray output in the text, or links to a blog post that does not exist.
- Tools that use AI
- The diagnosis is written by an AI model from your answers (Claude by default; the page names the model that wrote yours), and I do not read it before you do. The case it points to comes from my recorded work; the match is the model's judgment. The audit and the calculators run fixed rules and arithmetic, and so does the quiz, which says "not yet" when the answers say so.