Structured observation of how AI evaluators surface, compare, validate, and eliminate a specific organization in matched buying scenarios: what is said, what is recommended, what evidence is requested, and where confidence visibly breaks down.
Why this is marked experimental: a capability cannot be marked operational until it derives from published, stable method (the build enforces this). M-1 is under active development and no observation method has yet been published, so this capability is offered as what it is — careful, honest, early-instrument work, with sampling conditions and variance disclosed rather than hidden.
What this cannot tell you: why the evaluator behaves as it does (mechanism), or whether its behavior will survive the next model update (see Q-3).