Verify facts
Prefer primary material. A material claim needs one Tier 1 source or two independent Tier 2 sources.
How We Assess
Sources first. Analysis labeled. Corrections recorded.
Label the analysis
Facts, inferences, opinions, forecasts, and opportunity hypotheses are labeled separately.
Update the record
A public assessment changes only when new evidence alters the phase, impact, or next watchpoint.
Evidence and event history
Every public event links to source material. Check the original evidence and publication date before using a claim.
View the event historyAnalysis has limits
Role assessments, opportunity hypotheses, cost comparisons, and future signals are reference material. Apply them to your own context.
Read the content policyEvery Event is about Chinese AI
A published Event names a Chinese company, lab, robotics maker or regulator, a product built in China, a Chinese policy, or a change in the China market. The reason is recorded and can be corrected by an editor with an audited note.
Chinese evidence, English writing
Chinese-language evidence is linked in the original. The English text is written from it and marked Translated from Chinese, and the original Chinese name of a company, model or policy is shown next to the English name. Names are never silently rewritten.
Can you use it outside China?
Model and product Events say Global API, China-only API, Open weights, Waitlist or Unknown. Every value except Unknown links the page that shows it.
Company-reported until tested
Benchmark scores and capability claims published by the lab are labeled Company-reported. Independently evaluated means a named third party tested it, with the date.
Public sources, fetched politely
Collectors use official APIs, feeds and public pages, check robots.txt before every request, and never bypass logins, paywalls or CAPTCHAs. Only headlines, links and short summaries are kept; full articles are never republished.