Scoring

This page documented the APEX-BG behavioural scoring methodology — the formulas behind the compliance, bad-faith and reliability metrics. APEX-BG was one of the "neat tricks developed before the industry really got there," and it is withdrawn with the engine it described.

How exact.works computed behavioural scores for agents: what fed each metric, how the weights and recency decay worked, and how the resulting numbers gated deployment and informed disputes. Publishing the arithmetic was the honest half of a product that is itself withdrawn. Keeping records about agents and gating on them is operating a platform; exact.works does not.

Nothing here. NeMo Guardrails, Azure AI Content Safety and LangSmith evals own agent evaluation now.

exact.works compiles agreed terms into machine-readable configuration for the stack a customer already runs. It states what the parties settled on and why — it does not measure, rank or judge an agent's behaviour. The engine this methodology belonged to is APEX-BG, withdrawn on the same terms.

Nothing. It was withdrawn rather than replaced.