1. Primary sources first

Model specifications, API capabilities, pricing, rate limits and availability are verified against first-party documentation whenever possible. Third-party reporting is used mainly for context that is not available from a primary source.

2. Claims are labeled by origin

Vendor-published benchmark results are described as vendor results. Independent tests are labeled separately and include a methodology, date, model identifier and enough detail to understand what was tested.

3. Dates matter

AI product facts change quickly. Each factual guide includes a reviewed date. Comparisons identify the exact competing model version where possible so readers are not unknowingly comparing a current model with an obsolete one.

4. No invented testing

We do not describe prompts as tested unless the test was actually performed and recorded. Planned tests remain clearly marked as planned until results exist.

5. AI-assisted production

Automation may be used to help structure or draft content, but factual fields are checked against cited sources before publication. Human-readable usefulness and source traceability take priority over publishing volume.

6. Corrections and updates

Material factual changes are applied to the relevant page and, when useful to readers, added to the site changelog. Minor grammar and formatting edits do not trigger a new factual review date.