Editorial policy
Claims about metrics have to survive the fixtures.
This policy explains who is responsible for AI Stats articles and docs, how importer and methodology statements are checked, when an updated date is earned, and how to request a correction.
Last reviewed 2026-08-09.
Review and accountability
Metric and importer claims are checked against the current code before they are published.
- A feature is described as available only when the adapters, dashboard registries, or public pages already implement it. Roadmap items stay on the roadmap.
- When an export format does not contain a metric, the writing says Unavailable and names the capability-matrix reason. It never substitutes a zero, an estimate, or a typical value for Claude, Gemini, or ChatGPT web tokens, costs, or missing model names.
- AI Stats guidance is product documentation, not financial, legal, or workplace advice. Readers remain responsible for how they interpret their own imports.
Corrections
Specific correction reports go to the team that can amend the article and the product references it cites.
- Email the page URL, the statement at issue, and a primary source or a reproducible import result when you have one. The correction link on every article prepares that mail.
- Material corrections update the article's modified date and flow into metadata, the sitemap, and the RSS feed. Spelling and link maintenance do not earn a freshness date.
- When a correction changes how a shipped metric should be read, the methodology page, the capability matrix, and the related dashboard copy are reviewed with it.
AI assistance
AI may help organize a draft. It is never the accountable author or the final reviewer.
- The team may use AI tools to outline, to find questions worth verifying, or to clarify prose. Every factual and product claim still has to survive human review against the current adapters and registries.
- No named expert persona is invented for authority. The organizational byline identifies who is responsible for publication and corrections.
- Imported user exports are not part of the editorial workflow. Parsing stays in the reader's browser; editorial review uses fixtures and public pages only.
Primary sources
Real export fixtures, adapter tests, and first-party provider documentation take priority.
- Capability claims prefer sanitized fixtures in this repository and the contract suite that parses them. If a cell and an adapter disagree, the test fails and the prose is wrong.
- Provider documentation is linked where it substantively supports an export-format claim. Product behaviour is verified in the current application rather than inferred from a marketing page.
- Secondary round-ups may provide context. They do not replace a fixture, an adapter test, or a vendor export guide for a metric-quality claim.
Freshness and revision dates
Dates describe real publication and substantive revision, not routine rebuilds.
- Every article records its publication date. An updated date appears only when the guidance materially changes after publication.
- Feature-dependent pages are reviewed when the related adapter, dashboard, or methodology behaviour changes. The changelog is the historical product context an older article was written against.
- A route's sitemap date is derived from the same revision metadata the page and its structured data use, so a deployment alone does not manufacture a freshness signal.
Topic clusters
Articles sit in a small set of durable subject areas. Each cluster points at a cornerstone page — the methodology centre, or a guide that states what the export formats actually contain.
Report a correction
Use the correction link on an article or email the team directly. Include the URL, the exact statement, what you believe is wrong, and when available a primary source or a reproducible import result.
Email an editorial correction