Evidence model

Claims are smaller than articles.

Each consequential claim records a proposition, fact-or-inference type, sources, confidence, owner, review date and disposition. Sources record title, publisher, URL, publication/access date, locator and source type. Attachments are untrusted research inputs; a citation is not permission to reuse text or media.

01 / Event vocabulary

plannedannouncedlimited testcompletedpartialreversedsupersededoutcome unverified

Dates preserve precision. An announcement date, effective date, evidence-review date and page-modification date are not collapsed into “updated today.”

02 / Verification pass

  1. Identify the canonical intent owner so two pages do not answer the same job.
  2. Check edition, region, device, plan, effective date and exception.
  3. Use primary sources for rules, transactions and regulatory acts.
  4. Add independent, worker or player evidence before interpreting consequences.
  5. Search for a contrary reading and expose a narrow conflict.
  6. Do not turn documentation into a claim that hands-on testing occurred.

Freshness and failure behavior

Historical pages change slowly; product, service and policy pages need more frequent checks. Every page shows its last-reviewed date. When current evidence is insufficient, the affected detail is omitted or described as unresolved instead of being filled from memory.

AI assistance and rights

AI assistance is disclosed. No human credential or review is implied. Original prose, accessible HTML tables and CSS diagrams are preferred. Third-party images remain blocked until exact permission and accessibility records exist.

Method

Claims separate documented facts from interpretation and uncertainty. Read our method.

Updates

Material factual revisions are recorded publicly. View the correction log.