跳到主要内容
Documentation

Fact Check Signal

Overview

版本:最新版

Fact Check Signal

Overview

fact-check decides whether a prompt should be treated as evidence-sensitive traffic. It maps to config/signal/fact-check/ and is declared under routing.signals.fact_check.

This family is learned: it relies on the fact-check classification path under global.model_catalog.modules.hallucination_mitigation.fact_check.

Key Advantages

  • Separates factual verification from general domain routing.
  • Helps decisions choose safer plugins or stronger models for evidence-sensitive traffic.
  • Keeps verification policy visible instead of burying it inside later plugins.
  • Exposes both positive and negative labels such as needs_fact_check and no_fact_check_needed.

What Problem Does It Solve?

Not every prompt needs the same level of factual grounding. If the router treats all traffic the same, creative prompts can be over-constrained while factual prompts can be under-protected.

fact-check solves that by detecting which prompts should trigger evidence-aware routing behavior.

When to Use

Use fact-check when:

  • factual claims need stricter routing or plugins
  • creative prompts should bypass expensive verification paths
  • hallucination mitigation depends on an early routing signal
  • you want factuality handled as routing policy instead of post-hoc repair

Configuration

Source fragment family: config/signal/fact-check/

routing:
signals:
fact_check:
- name: needs_fact_check
description: Queries with factual claims that should be verified against evidence.
- name: no_fact_check_needed
description: Creative or opinion-heavy prompts that do not need factual verification.

Define only the labels your decisions will reference. The learned classifier decides which one fires.