LLMFactualityGuardrail

Builds a factuality guardrail for a reference text.

Builds a factuality guardrail with a threshold of 0.5.

The lower threshold is the only difference from apply; it does not guarantee that claims the context does not cover pass. The rubric scores 0.5 only for content that mixes supported and unverifiable claims, so content none of whose claims the context supports may still score below 0.5 and fail (see the class documentation).

Builds a factuality guardrail with a threshold of 0.9.