Returns granular application-recorded guardrail decisions and clearly labelled model-judged evaluations without user content. Both collections have independent bounded offset pagination while summary metrics remain scoped to the complete filtered period.