How should an embodied agent respond when a person's correction may be wrong? We formulate grounded correction arbitration as a choice among accepting, rejecting, inspecting the world, and asking the speaker. GAVA implements this interface with observation-bounded evidence, legal probes, and a one-step expected-loss rule. In text-only ALFWorld, 162 checkpoints produce 972 paired true and false interventions. Complete local inspections give GAVA and always verify 100 percent correction accuracy, establishing the evidence contract rather than a comparative advantage. In same-episode execution, GAVA reduces interaction cost against always verify but ties a cost threshold under a perfect speaker. An exploratory training-only object-location prior lowers interaction and declared joint cost on 340 unseen scenarios by 0.490 and 0.420 relative to uniform GAVA. After freezing the policy, costs, baselines, and multiplicity plan, the gains replicate on 77 non-overlapping seen checkpoints, covering 308 scenarios: 0.595 and 0.517, with both 95 percent checkpoint-bootstrap confidence intervals excluding zero. Joint cost also improves over an identical-prior fixed policy, while the matched calibrated no-VOI comparison remains inconclusive. Semantic GAVA makes four factual errors in each cohort, corresponding to 98.8 percent and 98.7 percent accuracy, and all methods complete every task. Results support selective information gathering with semantic priors under declared costs, but do not establish a general advantage of environmental value of information over clarification. The study uses normalized claims, complete symbolic observations, and controlled speakers; it evaluates neither human participants, visual input, nor physical robots.
Figures & tables
Research line
Executable actions
Fallible correction
Active probe
Observable-state boundary
Robot language correction ( Thierauf et al., 2024 ; Sharma et al., 2022 ; Cui et al., 2023 ; Liu et al., 2023 ; Shi et al., 2024 )
Yes
–
–
varies
Closed-loop embodied LLM planning ( Huang et al., 2022 ; Bhat et al., 2024 )
Yes
–
feedback
varies
Active HRI query and clarification ( Deits et al., 2013 ; Cakmak and Thomaz, 2012 )
Yes
–
dialogue
Yes
Text/visual correction selectivity ( Sinha, 2026 ; Pi et al., 2025 ; Beigi et al., 2025 )
–
Yes
–
–
This work
Yes
Yes
Yes
Yes
Table 1. Relationship to adjacent research. “Yes” means the capability is central to the cited evaluation, not merely possible in principle. A comparison table showing that prior robot correction, closed-loop planning, and sycophancy benchmarks each cover only subsets of executable action, fallible user feedback, active verification, and explicit observable-state isolation; this work covers all four in a text simulator.
Split
All
Pilot/quota
Inspected
Timeout
Other
Accepted
Unseen
134
26
108
0
23
85
Seen
140
29
111
21
13
77
Table 2. Checkpoint selection flow. “Other” includes non-location policy openings, policy/fact mismatches, and unmatched locations. The complete flow from all candidate games through pilot or quota exclusions, inspection, rejection, and final accepted checkpoints.
Split
Checkpoints
Pairs
Interventions
Valid unseen
85
255
510
Valid seen
77
231
462
Total
162
486
972
Table 3. Frozen benchmark composition. “Pairs” counts pressure-specific true/false twins; the statistical unit is the checkpoint. The unseen split contains 85 checkpoints, 255 true-false pairs, and 510 interventions. The seen split contains 77 checkpoints, 231 pairs, and 462 interventions.
Figure 3. Semantic arbitration results. (a) Paired mean differences and 95% checkpoint-bootstrap CIs; negative values favor Semantic GAVA. U is the unseen exploratory cohort and S the prospectively frozen seen replication. Matched baselines appear only in S. (b) Seen interaction cost versus incorrect fixed-decision (IFD) rate. Dashed contours show equal declared joint cost. All methods complete every task. Panel a is a forest plot. Semantic GAVA has lower interaction and declared joint cost than uniform GAVA and always verify on both cohorts, and lower declared joint cost than prior-only on the seen replication. Its interval relative to calibrated no-VOI crosses zero. Panel b plots interaction cost against the percentage of incorrect fixed decisions. Semantic GAVA lies near the lower-left corner; no-VOI uses slightly less interaction but makes more factual errors.
LLM agents that answer questions and act on behalf of users must decide when to yield to a user who disagrees. Disagreement may correct a genuine error or may be an attempt at manipulation. Prior work shows that models frequently abandon correct answers when challenged by users, yet it remains unclear whether they distinguish correction from manipulation or merely differ in their propensity to concede. We examine this question across seven LLMs from four vendors in two studies, one on the answers agents give and one on the actions they take as customer-support agents whose policy permits account changes only for verified users. Simulated users push toward correct and incorrect answers with graded pressure. Following research on epistemic vigilance, the pressure either appeals to the source, such as claimed authority, or offers content the agent cannot verify, namely a cited source. Drawing on signal detection theory, we measure on a set of benchmark questions how much more readily each agent concedes to correction than to manipulation, separately from its overall propensity to concede. Citations made five of seven models readier to concede and four worse at discriminating, while stronger wording added little. In the agentic study, five models never violated the verification policy, including models that readily abandoned correct answers, so concession on answers did not predict violations of policy. Ministral 3B violated the policy mainly under claimed authority, and Ministral 8B reported changes it had not made. An instruction to revise only on verifiable evidence chiefly shifted propensity to concede rather than improving discrimination. Concession rates alone are therefore misleading, and agents should be evaluated on both their discrimination and the actions they take.
User instructions are often underspecified because humans rely on implicit assumptions about the surrounding environment. For large language model (LLM) agents operating in information-rich digital and physical environments, these assumptions cannot be inferred from the instruction alone; they must be recovered from the current state of tools, data, interfaces, and observations. Effective execution therefore requires agents to identify missing context, ground it in observed evidence, and carry it forward into subsequent actions. We show that current agents often fail to do so. They act from assumed rather than observed specifics, overlook information they could have gathered, and fail to incorporate evidence that has already been returned. Building on this insight, we propose ACCORD (Action-Conditioned Contextual Grounding), a simple and effective agent framework for adaptive grounding. Before each action, ACCORD actively probes the environment for missing information and integrates relevant context from the agent's trajectory that would otherwise be overlooked. Requiring no additional training or task-success signals, ACCORD improves task-goal completion on AppWorld by up to +20.6 points with GPT-5-mini, from 42.0% to 62.6%, compared to strong baselines. These gains persist with a substantially stronger base model (+10.8 with Claude-4.5-sonnet), an open-weight model (+10.1 with Qwen3.5-27B-FP8), and on the embodied AlfWorld benchmark (+7.4 success rate with GPT-5-mini).
Lai Jiang, Cheng Qian, Zhenhailong Wang +3
University of Illinois Urbana-Champaign · Stanford University
Before letting an agent operate over real context, can you prove it used the right evidence? GroundEval turns that question into a deterministic test of what the agent searched, fetched, cited, and was permitted to access. In one case study, two frontier LLM judges scored a plausible agent response 0.85 and higher. But the trace told a different story: the agent had never retrieved the artifact its answer depended on, yielding a GroundEval score of 0.000. We introduce GroundEval, a judge-free framework for evaluating agents against grounded, time-bounded, and access-controlled evidence. GroundEval uses a domain configuration to generate questions, lets the agent choose how to answer, and then scores both the final answer and the recorded trajectory that produced it. The benchmark targets three failures that LLM-as-judge evaluation struggles to detect: whether an agent checked before claiming absence, reasoned only from evidence available to the actor at the relevant time, and used the correct causal mechanism rather than a plausible one. These correspond to three tracks: Silence, Perspective, and Counterfactual. GroundEval exposes when plausible answers rest on invalid evidence paths, and produces structured per-question diagnostics that pair tool activity with the agent's turn-level narration, making each score inspectable rather than merely reported. Our case studies suggest this failure mode is common rather than exceptional, one that final-answer and judge-based evaluation cannot detect by construction.