Credit Without Ground Truth: Auditing Step-Level Credit Assignment in LLM Agents Against Executed Replay

• URL: http://arxiv.org/abs/2608.19760v2
• Date: Tue, 25 Aug 2026 17:45:26 GMT
• Title: Credit Without Ground Truth: Auditing Step-Level Credit Assignment in LLM Agents Against Executed Replay
• Abstract summary: We audit against policy-conditional ground truth from executed replay in a single-agent tool environment.<n>No step-level credit signals we audit show reliable incremental fidelity beyond marginal-matched shuffled control.<n>A confidence-only router recovers pivotal steps at chance level, but cuts judge cost by 13.1% per turn.
• Score: 0.6768558752130311
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Audited against policy-conditional ground truth from executed replay in a single-agent tool environment (ALFWorld), none of the step-level credit signals we audit -- LLM-judge scores, outcome-conditioned logprob ratios, or the policy's own confidence -- shows reliable incremental fidelity beyond its own marginal-matched shuffled control. Correcting for replay-target reliability leaves implicit fidelity bounded near zero and judge fidelity inconclusive at the achieved target reliability. Existing evaluations grade these signals against annotated step *correctness*; we audit them against step *contribution* -- what re-sampling the policy's own alternatives at each decision point and rolling forward changes about the outcome -- and they come apart. The ground truth is structured: 30.5% of decision points where it is defined exhibit a nonzero replay contrast at the achieved sampling resolution, and measurability is model-dependent -- the fraction of points with no policy-supported counterfactual differs twofold (13.1% vs. 26.8%) between two similar-scale policies. The failure mode is identifiable: implicit credit echoes the policy's fluency (median rank correlation +0.75, replicating at +0.70 in a second family under a corrected instrument), while outcome conditioning adds no causal information (partial correlation -0.004, Qwen). A confidence-only router recovers pivotal steps at chance level, but cuts judge cost by 13.1% per turn (14.0% per trajectory). In a seven-arm pre-registered training experiment, no arm reliably outperforms the untrained policy, and the checkpoints' apparent instrument signature is statistically consistent with mediation by effective training dose in this design -- sparser credit retains fewer examples, an order-of-magnitude spread in optimizer steps -- not credit content. Comparisons of credit rules must match effective sample size, or they measure dose, not credit.

Related papers

• Beyond Local Accuracy: A Protocol-Level Identifiability Audit for Controlled LLM Reasoning Evaluation [6.047519836191728]
  We formalize a protocol-level identifiability audit over a finite behavioral policy class.<n>The audit requires zero model calls and resolves our diagnostic case.<n>This case shows how evaluation-design validity can be checked structurally before model inference.
  arXiv  Detail & Related papers  (2026-08-13T14:49:47Z)
• When Can You Trust Offline Evaluation of Equal-Cost Top-k Allocation? A Controlled, Reproducible Benchmark and Practitioner's Guide [0.0]
  We benchmark six estimators across five datasets and two known-effect sweeps, and validate the mechanisms against a non-simulated paired reference.<n>Sharpening a logger from the target's own score barely moves overlap over the tested range; action-level disagreement collapses it.<n>Cross-fitting the outcome nuisance alone leaves the reuse bias in place and makes it worse.
  arXiv  Detail & Related papers  (2026-08-12T18:10:10Z)
• SVR: Self-Verifying Refinement via Joint Verdict-Confidence Reinforcement Learning for Adaptive Test-Time Compute [62.3458279176813]
  Self-Verifying Refinement is an oracle-free multi-turn reinforcement learning framework.<n>It learns to use self-verification as a compute-control policy.<n>It achieves a macro-average accuracy of 0.563 with only 2.99 inference turns on average.
  arXiv  Detail & Related papers  (2026-07-30T16:20:58Z)
• ContainmentBench: Trace-Based Evaluation of Post-Injection Containment in Tool-Using LLM Agents [4.018837636907658]
  We introduce a sandboxed, trace-based benchmark that measures benchmark-defined endpoint policy compliance.<n>We find that aggregate logged-spread rankings change with evidence-stage composition and denominator choice.<n>These results show that a terminal policy label is not a sufficient statistic for operational post-exposure containment.
  arXiv  Detail & Related papers  (2026-07-27T04:51:20Z)
• Critic Experience Bank: Self-Evolving Step-Level Confidence Estimation for LLM Agents [7.774009529494805]
  LLM agents act in external environments where each action changes the state that later decisions condition on, and where a single wrong step can waste interaction budget or trigger irreversible side effects long before the final failure is observed.<n>We introduce the method (methodshort), a self-evolving critic framework in which an LLM critic accumulates evidence from its own past judgments and their observed consequences.
  arXiv  Detail & Related papers  (2026-07-14T06:17:09Z)
• What Accuracy and Gradient Cosine Miss: Evaluating Feedback Alignment via Scale Stability, Reference Validity, and Depth Utility [48.10132234701036]
  We propose a diagnostic evaluation protocol based on three checks -- scale stability, reference validity, and depth utility -- together with per-layer rather than aggregate cosine reporting.<n>Across multiple architectures and methods, our protocol identifies all failures with wide calibration margins.
  arXiv  Detail & Related papers  (2026-06-19T06:04:17Z)
• AgentFairBench: Do LLM Agents Discriminate When They Act? [2.3004655342211078]
  AgentFairBench is a cheap, reproducible, multi-domain benchmark for demographic disparity in the actions of LLM agents.<n>It spans three regulator-anchored domains: hiring, lending, and medical triage.<n>A NumPy-only harness computes counterfactual flip rate, mean absolute score difference (MASD), action-rate disparity, and tool-invocation disparity.
  arXiv  Detail & Related papers  (2026-06-15T13:50:26Z)
• DFAH-Bench: Benchmarking Observable Agent Instability in Financial Decision-Making [0.7699235580548228]
  DFAH-Bench is a replay benchmark that measures observable behavioral instability in financial agent decision-making.<n>We find that frontier models can agree on decisions 95% of the time while following the same tool path only 77% of the time.<n>We identify three behavioral profiles: pattern matchers that achieve near-perfect agreement by collapsing to a single output regardless of input, stable executors with relatively consistent tool-use processes, and trajectory divergers that reach the same conclusions through materially different tool paths and evidence contacts.
  arXiv  Detail & Related papers  (2026-06-10T03:31:09Z)
• Causal Agent Replay: Counterfactual Attribution for LLM-Agent Failures [0.0]
  Causal Agent Replay (CAR) models an agent run as a structural causal model.<n>It applies a do-operation to a step, and re-executes a trajectory forward under the same policy.<n>CAR is open source and runs on hosted or free local models.
  arXiv  Detail & Related papers  (2026-06-06T17:44:23Z)
• Step-wise Rubric Rewards for LLM Reasoning [72.17879367869503]
  Reinforcement Learning with Verifiable Rewards (RLVR) is widely used to improve reasoning in large language models.<n>We find 18.2% of steps in correct responses are wrong yet positively rewarded.<n>Across six mathematical reasoning benchmarks, SRaR improves average accuracy over RaR by 3.57 points.
  arXiv  Detail & Related papers  (2026-05-17T07:08:14Z)
• Beyond Accuracy: Policy Invariance as a Reliability Test for LLM Safety Judges [26.595399077062638]
  LLM-as-a-Judge pipelines have become the de facto evaluator for agent safety.<n>Existing benchmarks treat their verdicts as ground-truth proxies without checking whether the verdicts depend on the agent's behavior or merely on how the evaluation policy happens to be worded.<n>We operationalize it as three testable principles: rubric-semantics invariance under certified-equivalent rewrites, rubric-threshold invariance under intentional strict-to-lenient shifts, and ambiguity-aware calibration.
  arXiv  Detail & Related papers  (2026-05-07T12:49:09Z)
• How Independent are Large Language Models? A Statistical Framework for Auditing Behavioral Entanglement and Reweighting Verifier Ensembles [46.63622714488747]
  Shared pretraining data, distillation, and alignment pipelines can induce hidden behavioral dependencies, latent entanglement.<n>In practice, this manifests as correlated reasoning patterns and synchronized failures.<n>We develop a statistical framework for auditing behavioral entanglement among black-box language models.
  arXiv  Detail & Related papers  (2026-04-08T23:32:06Z)
• Claw-Eval: Toward Trustworthy Evaluation of Autonomous Agents [66.97968363332465]
  We introduce Claw-Eval, an end-to-end evaluation suite addressing all three gaps in agent benchmarks.<n>Claw-Eval comprises 300 human-verified tasks spanning 9 categories across three groups.<n>Every agent action is recorded through three independent evidence channels.
  arXiv  Detail & Related papers  (2026-04-07T17:43:18Z)
• CARE What Fails: Contrastive Anchored-REflection for Verifiable Multimodal [84.71254539482369]
  Group-relative reinforcement learning with verifiable rewards (RLVR) often wastes the most informative data it already has the failures.<n>We present CARE, a failure-centric post-training framework for multimodal reasoning that turns errors into supervision.<n> CARE improves accuracy and training smoothness while explicitly increasing the share of learning signal that comes from failures.
  arXiv  Detail & Related papers  (2025-12-22T16:34:21Z)
• Reference-Free Rating of LLM Responses via Latent Information [53.463883683503106]
  We study the common practice of asking a judge model to assign Likert-scale scores to free-text responses.<n>We then propose and evaluate Latent Judges, which derive scalar ratings from internal model signals.<n>Across a broad suite of pairwise and single-rating benchmarks, latent methods match or surpass standard prompting.
  arXiv  Detail & Related papers  (2025-09-29T12:15:52Z)
• TrustJudge: Inconsistencies of LLM-as-a-Judge and How to Alleviate Them [58.04324690859212]
  Large Language Models (LLMs) as automated evaluators (LLM-as-a-judge) has revealed critical inconsistencies in current evaluation frameworks.<n>We identify two fundamental types of inconsistencies: Score-Comparison Inconsistency and Pairwise Transitivity Inconsistency.<n>We propose TrustJudge, a probabilistic framework that addresses these limitations through two key innovations.
  arXiv  Detail & Related papers  (2025-09-25T13:04:29Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.