Method / Blog
How we publish articles.
Articles are decision aids with a named author — not keyword shells.
What does every article have to pass?
- Named human byline — never a brand Person in schema.
- A direct answer in the first 60 words of the article.
- A genuinely complete answer to its question — articles under 400 words fail the audit as thin; padding to hit a word count is a defect, not depth.
- No near-duplicates — a sampled word-overlap (Jaccard) gate runs on every build; template-generated series and mass-appended shared sections both fail it.
- Markdown mirror + inclusion in
llms-full.txtwhen indexable. - Exactly one JSON-LD
@graphafter the merge gate.
What did we refuse to do?
In August 2026 we removed a shared "enricher" that appended identical decision sections to all 455 articles to pass a word-count check — it passed the metric while doubling measured duplication. We also deleted thirty template-generated articles that were 96% word-identical, replacing them with ten hand-written ones. The standard now: one argument per article, written by a person, or the article does not exist. Fewer, better pages.
How do corrections appear?
Substantive fixes are listed on /corrections/ with date and page. The standards live in the editorial policy.