Five rules. They exist because I got two claims wrong early and had to correct them.
Everything on this site follows five rules. I'd rather publish the standard than ask you to take the results on faith.
Fully automated, AI-assisted, rule-based, or human-approved. These are not the same thing, and collapsing them is the most common way automation writing misleads. When I say something is automated, the label tells you what kind.
If I claim a dollar figure, the line items are in the piece. You should be able to add them up and get my number — or tell me I'm wrong.
Approximate numbers are marked approximate, with the basis stated. Where a count was verified on a specific date rather than live, the date is in the text.
Time saved, maintenance hours, error rates, response-time deltas, cost per action. I don't measure any of them, and I say so in the case study rather than estimating something that sounds good. If you see a write-up with confident numbers in all of those categories, ask how each one was instrumented.
If I can't point to a real system that backs a statement, it comes out — even when it's the more interesting sentence. Two claims in early drafts of these case studies were wrong and were corrected before publication. One guardrail I had described as technical turned out to be policy. That correction is in the published version.
A wrong answer is worse than "I don't know."