Implementation:Truera Trulens Block Output Decorator

Knowledge Sources	TruLens TruLens Guardrails
Domains	Guardrails, Safety
Last Updated	2026-02-14 08:00 GMT

Overview

Concrete tool for blocking application outputs that fail a quality or safety check at runtime, provided by the trulens-core library.

Description

The block_output class is a Python decorator that evaluates the output of a method after execution. If the output scores below the threshold (or above for inverse metrics), a safe default value is returned instead.

Usage

Import and apply @block_output to methods whose output needs quality or safety validation before reaching the user.

Code Reference

Source Location

Repository: trulens
File: src/core/trulens/core/guardrails/base.py
Lines: L199-268

Signature

class block_output:
    def __init__(
        self,
        feedback: core_metric.Metric,
        threshold: float,
        return_value: Optional[str] = None,
    ):
        """
        Args:
            feedback: Metric instance for output evaluation.
                Must accept a single argument.
            threshold: Minimum score to allow output.
            return_value: Value to return if output is blocked. Defaults to None.
        """

Import

from trulens.core.guardrails.base import block_output

I/O Contract

Inputs

Name	Type	Required	Description
feedback	Metric	Yes	Feedback function returning a float (single argument)
threshold	float	Yes	Minimum score to allow output
return_value	str	No	Safe response returned when output is blocked

Outputs

Name	Type	Description
(decorator)	Callable	Wrapped function that blocks low-scoring outputs

Usage Examples

Block Unsafe Outputs

from trulens.core.guardrails.base import block_output
from trulens.core import Feedback
from trulens.providers.openai import OpenAI

provider = OpenAI()
f_criminality = Feedback(provider.criminality, higher_is_better=False)

class SafeApp:
    @block_output(
        feedback=f_criminality,
        threshold=0.5,
        return_value="Sorry, I can't provide that information."
    )
    def respond(self, question: str) -> str:
        return self.llm.generate(question)

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment