Skip to main content
Intrinsic functions for Guardian safety and hallucination detection. The guardian-core LoRA adapter uses a <guardian> protocol tag in the last user message to specify system prompt, criteria, and scoring schema. This differs from other intrinsics, which rely on the instruction field in io.yaml.

​
Functions

​
FUNC policy_guardrails 

policy_guardrails(context: ChatContext, backend: AdapterMixin, policy_text: str) -> str
Checks whether text complied with specified policy. Uses the policy_guardrails LoRA adapter to judge whether the scenario described in the last message in context is compliant with the given policy_text. :param context: Chat context containing the conversation to evaluate. :param backend: Backend instance that supports LoRA adapters. :param policy_text: Policy against with compliance is to be checked :return: Compliance as a “Yes/No/Ambiguous” label (Yes = compliant).

​
FUNC guardian_check 

guardian_check(context: ChatContext, backend: AdapterMixin, criteria: str, target_role: str = 'assistant') -> float
Check whether text meets specified safety/quality criteria. Uses the guardian-core LoRA adapter to judge whether the last message from target_role in context meets the given criteria. Args:
  • context: Chat context containing the conversation to evaluate.
  • backend: Backend instance that supports LoRA adapters.
  • criteria: Description of the criteria to check against. Can be a key from :data:CRITERIA_BANK (e.g. "harm") or a custom criteria string.
  • target_role: Role whose last message is being evaluated ("user" or "assistant").
Returns:
  • Risk score as a float between 0.0 (no risk) and 1.0 (risk detected).

​
FUNC factuality_detection 

factuality_detection(context: ChatContext, backend: AdapterMixin) -> float
Determine is the last response is factually incorrect. Intrinsic function that evaluates the factuality of the assistant’s response to a user’s question. The context should end with a user question followed by an assistant answer. :param context: Chat context containing user question and assistant answer. :param backend: Backend instance that supports LoRA/aLoRA adapters. :return: Factuality score as a “yes/no” label (yes = factually incorrect).

​
FUNC factuality_correction 

factuality_correction(context: ChatContext, backend: AdapterMixin) -> float
Corrects the last response so that it is factually correct. Intrinsic function that corrects the assistant’s response to a user’s question relative to the given contextual information. :param context: Chat context containing user question and assistant answer. :param backend: Backend instance that supports LoRA/aLoRA adapters. :return: Correct assistant response.