Our claims should be
easy to audit.
We show what we know, how we know it, and where the evidence stops.
Three labels, three different claims.
We performed the action or ran the test ourselves. The page identifies the date and relevant conditions.
A primary vendor source supports the statement. This establishes what is offered, not how well it performs.
Our conclusion based on disclosed facts and tradeoffs. Analysis is clearly separated from product claims.
No evidence, no score.
We do not turn documentation into a performance rating. A category receives a score only after sufficient hands-on testing. Overall scores, when introduced, will weight reliability, routing effectiveness, observability, developer experience, privacy controls, and effective cost.
Repeatable workloads over demos.
Tests use recorded workload categories, fixed success criteria, multiple runs, and explicit failure scenarios. We report context, medians, tail behavior, and material limitations. Vendors may fact-check product details but cannot preview or alter conclusions.
Every review carries a date.
We re-check high-change fields such as pricing and model access on a rolling basis. Material corrections are noted on the affected page. Readers and vendors can submit evidence of an error; source-backed corrections are made regardless of commercial relationship.