Faithfulness Metrics

Measures the internal coherence of the reasoning chain. A faithful chain must avoid contradictions and maintain a valid logical flow from premises to conclusion.

• Key Tests: Checking for logical fallacies, non-sequiturs, or contradictory statements within the same chain.
• Example: If a model states 'All mammals are warm-blooded. A whale is a mammal. Therefore, a whale is cold-blooded,' the chain is logically inconsistent and unfaithful.
• Evaluation Method: Often assessed by having a verifier model or rule-based system check for formal logical errors in the step sequence.

Assesses whether the facts and data cited in the reasoning steps are accurate and verifiable against a knowledge source.

• Core Principle: Each declarative statement in the reasoning (e.g., 'The capital of France is Paris') must be factually true.

• Challenge: Models may generate plausible-sounding but incorrect 'facts' that still lead to a correct final answer by chance, indicating low faithfulness.

• Evaluation Method: Compare statements against trusted knowledge bases (e.g., Wikipedia, proprietary databases) or use a retrieval system to verify claims.

Evaluates if each reasoning step is pertinent to solving the problem and if the chain would fail without it. Extraneous or irrelevant steps indicate 'reasoning noise.'

• Key Question: Is this step required to derive the final answer, or is it decorative?
• Example: For a math problem, generating a historical anecdote about Pythagoras is irrelevant. A faithful chain includes only necessary calculations and logical inferences.
• Evaluation Method: Human annotators or advanced models judge the necessity of each step, often by attempting ablated reasoning chains.

Granularly scores the accuracy of each individual inference or calculation, not just the final output. This is the foundation of Process Supervision.

• Distinction from Final Answer Correctness: A model can guess the right answer (e.g., '42') with a completely wrong reasoning process.

• Importance: Critical for debugging model logic and for high-stakes applications where the process must be auditable and trustworthy.

• Evaluation Method: Requires fine-grained labeled datasets where each reasoning step has a verifiable truth label.

The highest-order test: does the reasoning chain causally explain the final answer? This determines if steps are genuine reasoning versus post-hoc rationalization.

• Core Problem: A model may generate a correct final answer, then fabricate a seemingly logical chain that didn't actually guide its internal computation.

• Evaluation Challenge: Requires probing the model's internal decision-making, often via counterfactual testing—if a step is altered, does the final answer change accordingly?

• Advanced Methods: Use input perturbation or causal mediation analysis to establish the actual influence of generated steps on the output.

The technical approaches used to measure the dimensions above.

• Human Annotation: Gold standard but costly. Annotators score steps for factuality, relevance, and logical flow.

• Automatic Metrics:

  • NLI-based: Use a Natural Language Inference model to check if the final answer is entailed by the reasoning chain.
  • Fact-Score: Decompose reasoning into atomic claims and verify each against a knowledge source.
  • Self-Consistency: If multiple sampled reasoning paths converge on the same answer, faithfulness is more likely (but not guaranteed).

• Process Reward Models (PRMs): Specialized models trained to predict the correctness of a single reasoning step.