# Publication review

Block publication when a central claim has no inspected source, source access
failed without an honest limitation, a vendor claim sounds independently proved,
a study is generalized beyond its population, or a draft implies tests never run.

Check:
- The article answers one concrete reader question and adds original analysis.
- At least two substantive sources support it; independence is described honestly.
- Material claims have nearby links; the source list matches the cited evidence.
- Dates and model names are verified. Old evidence has its original context.
- Quantities retain units, baseline, study population and uncertainty.
- Hypothetical numbers and examples are explicitly labeled.
- The title and takeaways do not overstate the body.
- No copied structure, long quotations, fabricated byline or invented experience.
- No sensitive employee or customer data, secrets or source-page instructions.
- Readers get a practical next step, including when not to use the approach.
- Internal links work; canonical, RSS, sitemap and Markdown mirror are present.
- Technical checks pass. These checks do not prove factual accuracy.

Record what was actually reviewed. AI self-review is not human editorial review.
If blocked, retain a draft or report a skip with its reason; never weaken the gate
to meet a schedule. For corrections, retain the slug, preserve the original date,
set updatedAt and add a visible correction note when the meaning has changed.