# Agent Step Value: Auditing Evaluation-Channel Sensitivity in Black-Box Agent Traces

arXiv: https://arxiv.org/abs/2607.04419 · HTML: https://mlsanity.com/paper/2607.04419/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 22 of its 41 references that are in the map.
- First posted on arXiv 5 Jul 2026.

## Builds on (most cited first, 22 of 22)

- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Let's Verify Step by Step](https://mlsanity.com/paper/2305.20050/) (2023)
- [G-Eval: NLG Evaluation using GPT-4 with Better Human Alignment](https://mlsanity.com/paper/2303.16634/) (2023)
- [Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting](https://mlsanity.com/paper/2305.04388/) (2023)
- [Large Language Models are not Fair Evaluators](https://mlsanity.com/paper/2305.17926/) (2023)
- [Solving math word problems with process- and outcome-based feedback](https://mlsanity.com/paper/2211.14275/) (2022)
- [LLM Evaluators Recognize and Favor Their Own Generations](https://mlsanity.com/paper/2404.13076/) (2024)
- [Prometheus: Inducing Fine-grained Evaluation Capability in Language Models](https://mlsanity.com/paper/2310.08491/) (2023)
- [The Lessons of Developing Process Reward Models in Mathematical Reasoning](https://mlsanity.com/paper/2501.07301/) (2025)
- [Measurement and Fairness](https://mlsanity.com/paper/1912.05511/) (2019)
- [AgentRewardBench: Evaluating Automatic Evaluations of Web Agent Trajectories](https://mlsanity.com/paper/2504.08942/) (2025)
- [Process vs. Outcome Reward: Which is Better for Agentic RAG Reinforcement Learning](https://mlsanity.com/paper/2505.14069/) (2025)
- [Position: Evaluating Generative AI Systems Is a Social Science Measurement Challenge](https://mlsanity.com/paper/2502.00561/) (2025)
- [AgentProcessBench: Diagnosing Step-Level Process Quality in Tool-Using Agents](https://mlsanity.com/paper/2603.14465/) (2026)
- [Flaw or Artifact? Rethinking Prompt Sensitivity in Evaluating LLMs](https://mlsanity.com/paper/2509.01790/) (2025)
- [Evaluating the Consistency of LLM Evaluators](https://mlsanity.com/paper/2412.00543/) (2024)
- [Pairwise or Pointwise? Evaluating Feedback Protocols for Bias in LLM-Based Evaluation](https://mlsanity.com/paper/2504.14716/) (2025)
- [Scaling Evaluation-time Compute with Reasoning Models as Evaluators](https://mlsanity.com/paper/2503.19877/) (2025)
- [ReliableEval: A Recipe for Stochastic LLM Evaluation via Method of Moments](https://mlsanity.com/paper/2505.22169/) (2025)
- [Beyond Accuracy: Policy Invariance as a Reliability Test for LLM Safety Judges](https://mlsanity.com/paper/2605.06161/) (2026)
- [Efficient Process Reward Modeling via Contrastive Mutual Information](https://mlsanity.com/paper/2604.10660/) (2026)
- [Beyond Math and Code: Lightweight Corpus-Grounded Process Rewards for Factual Question Answering](https://mlsanity.com/paper/2605.29648/) (2026)

## Abstract (arXiv)

Reusing evaluator-derived step rewards across evaluation channels raises a measurement question: does the assigned direction remain stable? We define Agent Step Value (ASV) as a channel-indexed before/after target-margin gain and identify the state-by-channel interaction from a matched 2x2 design. Across frozen PubMed question-answering transitions, cross-channel gain signs disagree on 48.3% of comparisons, compared with 0.3% for same-channel retries. Transition interactions are nearly balanced in sign (median +0.014; 10% trimmed mean -0.033), but asymmetric negative magnitudes produce a positive direct mean and a negative generated-view mean under the empirical target mixture. Equal-target weighting preserves a negative mean interaction but not this strict mean reversal. Matched retrieval designs localize the negative component to the generated-view coordinate and retain its direction across three sign checks that vary readout and evaluator stack. In a matched 512-token source-contract follow-up, source-only generation restores the positive mean direction on artifact-bearing retrievals and removes parser-detected support claims from artifact-free before-state views. These results support an explanation in which generated support inflates the pre-retrieval baseline. The resulting audit can localize channel sensitivity and test an intervention before step rewards are reused.