Accountability
Editorial policy
Who writes it, who checks it, how we use AI, what stops a page going out, and how often we come back to it.
A description of this site's own process, every element of which is enforced by rules G1 to G12 in build.py and specified in SPEC.md section 3.
Most editorial policies are aspirational documents nobody checks. This one describes rules that are executed on every build, and a violation of any of them stops the site being written at all.
How is content on this site reviewed before publication?
Every page has to name who stands behind it, carry publication, modified and next-check dates, say whether AI helped draft it, and grade how solid the evidence is with a written reason. If any of that is missing, or a check date has gone by, the build stops and writes nothing.
For YMYL subjects the rule tightens: an AI-assisted page must have a subject-matter reviewer distinct from its author, or it cannot publish. One page is currently held back under that rule.
How AI is used
Plainly: AI is used for drafting, outlining, restructuring and metadata on this site, and the aiAssisted field on each page records whether it was used there.
It is not used to supply facts, figures, or judgments. Every figure comes from the statistics ledger, where each entry was fetched from its primary source on a recorded date. Every documented claim links to the documentation.
The byline names a person, never a model, because the byline identifies who is accountable for the page being correct.
That field is a factual statement about production, and it is deliberately not a lever. The YMYL gate cannot be satisfied by changing aiAssisted to false — a rule you can edit your way around is not a rule.
The review cycle
Every page gets a next-check date ninety days out. When one goes by, the build stops. That turns "we should keep this current" from a good intention into something that blocks the work.
A review is not a glance. It covers: re-verifying each cited figure against its primary source, checking that linked documentation still says what the page claims, re-checking platform-behavior rows that decay fastest, confirming internal links resolve, and reconsidering whether any claim should move between evidence grades.
That last one matters most. A claim graded Speculative today may become Documented when a platform publishes something — and a claim graded Documented can become wrong when a platform changes its rules.
Corrections
Errors get corrected on the page, and the modification date moves. Where a correction changes a conclusion rather than a detail, the page says what it previously said and why it was wrong.
This site argues that search content decays silently and that undated pages are unreliable. Quietly editing an error out would be the same failure.
What stops publication
The complete list, enforced in code rather than in process:
- Any required field missing.
- A review date that has lapsed.
- An author or reviewer who is not in the entity file, or who has no stated credential.
- An invalid evidence grade, or a grade with no stated reason.
- A figure in prose that is not traceable to the statistics ledger.
- An incomplete ledger entry.
- A dead internal link.
- A glossary term that does not exist.
- A broken heading hierarchy, or more than one top-level heading.
- A pillar with no spokes, or a spoke that does not link to its pillar.
- An AI-assisted YMYL page with no distinct subject-matter reviewer.
- A page claiming its evidence is documented without citing documentation.
More in About
Evidence grades
Three labels, one on every page, saying whether a claim is documented by the platform, held by practitioners, or genuinely a guess. The grade is required, and a page can't build without one.
What we sell
The library is free. The audit is the paid product. Nothing here is sponsored and there are no affiliate links, which is why a recommendation on this site is just a recommendation.
Myths & unverified stats
The most-quoted numbers in search marketing can't be traced to anybody who measured them. Here's why none of them are on this site, how to check one yourself, and the claims we refuse to repeat.
How this site is built
The stack, the rules that stop a bad page going out, and an honest list of what we haven't checked. If the site is the argument, this is where you go to test it.
Privacy
No analytics, no cookies, no trackers, no third-party requests, and no form wired to any endpoint. The on-site search runs in your browser. This page explains how to verify all of that rather than take it on trust.
Contact
How to reach the people who publish this site, and an honest note about why the address is not printed here yet.