Levels of proof

  1. Official source: the owner of a system, policy, standard, or law explains it.
  2. Reviewed research: experts reviewed a study before it was published.
  3. Direct result: we measured a result in a system we own or may study. We also share the method and limits.
  4. Expert review: we use shared proof to reach a reasoned view.
  5. Idea to test: the idea may be true, but it needs more proof.

Labels on each page

Each article says what kind of proof it uses. It may use official guidance, our own review, a direct case study, or a mix. This keeps our advice from sounding stronger than its sources.

Case study rules

  1. Name the subject, owner, question, measures, and set dates.
  2. Save the result before the change and define the comparison.
  3. List each change with its date.
  4. Check that source date ranges do not overlap.
  5. Share the full totals and comparison values.
  6. Name other things that may have caused the result.
  7. Separate facts, our view, possible causes, and lessons for others.
  8. Save the source files and share safe proof.

How we measure AI answers

We set the questions before a test starts. We save each prompt, product, mode, date, place, answer, source link, fact check, and repeat. We count mentions, citations, correct facts, advice, visits, and leads on their own. One answer is never called a stable world rank.

Claim list

Important claims need set wording, a source, a check date, an owner, and clear limits. This covers roles, prices, study results, client results, and comparisons. We remove or narrow a claim when its proof is old or gone.

How we use sources

We prefer direct sources. A source must support the claim beside it. Fame does not make a source strong. We save access dates and copies when facts may change. A source list cannot fix a claim with no support.

How we use AI

AI may help sort notes, review code or proof, suggest tests, and question drafts. It is not a fact source. A person checks each important claim against direct records. We do not give private client data to an AI tool without an approved agreement.

Updates and fixes

Each article has an update date and next review date. We fix major errors fast, retain the prior text in version control, and add material fixes to the correction log. We give major method changes a new version.