Tool-using LLM agents must decide not only whether additional information is needed, but also which source can resolve the uncertainty. Existing proactive approaches often specialize in either user clarification or environment verification, without explicitly determining the appropriate information source for each decision. We formulate this problem as uncertainty routing among ACT, CLARIFY, and VERIFY, and propose PROUR, a proactive uncertainty routing framework. PROUR decomposes action uncertainty into disagreement across plausible user-goal interpretations, which signals user-side ambiguity, and the entropy remaining within each interpretation, which signals missing world-side evidence. To acquire information from the routed source, a query generator is trained with a mode-conditioned information-gain reward, targeting user-goal identification under CLARIFY and next-action identification under VERIFY. On τ-bench, PROUR achieves 28.17% average success rate across retail and airline, outperforming the strongest prior method by 4.57% while using 2.17 fewer interaction steps. The learned policy further generalizes to stronger task agents and transactional domains of τ3-bench without retraining, demonstrating the benefit of source-aligned uncertainty resolution for proactive agents.
Figures & tables
Figure 1: An example of clarification and verification in τ -airline trajectories. Clarification alone burdens the user with facts available from tools (left), while verification alone misses user preferences (middle). Combining them resolves the preference before retrieving system evidence (right).
Figure 2: Source distribution of 19,022 step-level events from τ -bench. World-side cases (blue) account for 89.1%; user-side cases (orange) account for 10.9%. See Appendix A.1 for the full taxonomy.
Figure 3: Overview of the Prour framework . (a) Decomposer estimates action-level aleatoric and epistemic uncertainty from candidate user-goal interpretations and routes to clarify , verify , or act . (b) Query Generator is trained with a mode-conditioned information-gain reward, using the ground-truth user goal as the clarification target and reference next action as the verification target.
Figure 4: Training dynamics under mode-conditioned information-gain reward.
Figure 5: Effect of the decomposer-call budget on performance and cost. (a) Success rate on retail, airline, and their average. (b) Average wall-clock time and total interaction steps per episode. The ∞ setting imposes no per-episode limit on decomposer calls.
Figure 6: Uncertainty trajectories over interaction. Prour reports AU, EU, and total uncertainty over the first 16 steps, while baselines report total uncertainty with decomposer applied post hoc to trajectories.
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 7: Held-out routing performance of our calibrated Prour and the prompted LLM Router. Rows denote ground-truth labels and columns denote predicted modes; each cell reports the count and row-normalized percentage. Prour applies the Qwen3-8B thresholds selected on the retail development set, whereas the prompted Qwen3-8B router predicts a mode directly, without calibrated thresholds. Both methods are evaluated on the same disjoint airline test set.
Metric
Qwen3-8B
GLM-4-32B
(τAU,τEU)
(0.25,0.20)
(0.20,0.08)
Macro-F1
0.714
0.711
act F1
0.700
0.742
clarify F1
0.683
0.706
verify F1
0.760
0.685
Appendix
Table 6: Routing F1 after model-specific calibration. Macro-F1 and per-class F1 evaluate three-way routing on the same development set.
Figure 8: Effect of decomposer invocation frequency on performance and cost. (a) pass @1 on retail, airline, and their average. (b) Average wall-clock time and total interaction steps per episode. Each x-axis value n denotes invoking the decomposer once every n agent decision steps.
Recent position papers argue that the classical aleatoric/epistemic uncertainty framework is insufficient for interactive large language model (LLM) agents and call for underspecification-aware, decomposed, and communicable uncertainty representations that can unlock new agent capabilities such as proactive clarification seeking and shared mental-model building. Practical deployment constraints -- black-box APIs, interactive latency budgets, and the absence of labeled trajectories -- rule out logprob-based, multi-sampling, and training-based methods, leaving prompt-based estimation as the most viable family for surfacing such signals at deployment time. We answer this call with a simple prompt-based decomposition that separates action confidence from request uncertainty (u), enabling the agent to ask for clarification when the task specification is ambiguous. To evaluate it, we introduce two clarification-augmented benchmarks (WebShop-Clarification and ALFWorld-Clarification) in which 50% of tasks are deliberately underspecified, and systematically compare the proposed decomposition against ReAct+UE and Uncertainty-Aware Memory (UAM) across five LLM backbones (GPT-5.1, DeepSeek-v3.2-exp, GLM-4.7, Qwen3.5-35B, GPT-OSS-120B) on these variants together with the standard WebShop, ALFWorld, and REAL benchmarks for fault detection. Averaged across the five backbones, the proposed decomposition improves clarification F1 on ALFWorld-Clarification by 73% over ReAct+UE and by 36% over UAM, and leads clarification F1 on every backbone on WebShop-Clarification and on four of five backbones on ALFWorld-Clarification, indicating that the gains generalize beyond a single LLM.
Gregory Matsnev
AI Talent Hub, ITMO University Saint Petersburg 197101, Russia
Large Language Model (LLM) agents often operate under underspecified user instructions, where latent uncertainty over user intent leads to erroneous tool actions. To address this challenge, we propose a goal-oriented clarification framework that aligns clarification behavior with ambiguity resolution. Central to our approach is the Information Gain Reward, a metric that quantifies the utility of clarification questions by measuring the Bayesian belief update towards the ground-truth goal induced by the clarification exchange. We train the clarifier (LLM) using this reward to optimize for high information gain, ensuring that clarifications effectively reduce uncertainty and improve task completion within the agent-tool-user environment. We validate our framework within a clarification-enhanced τ-Bench environment, conducting cross-agent evaluations across five heterogeneous backbones. Empirical results demonstrate that our method consistently improves the success rate by 3.7% over the no-clarification baseline, while adding only 0.3 total interaction steps on average.
Mengyi Deng, Zhiwei Li, Xin Li +4
Information Hub, HKUST(GZ) · Jilin University · et al.
Reliable uncertainty quantification (UQ) is essential for deploying large language model (LLM) agents in complex interactive environments. Existing UQ methods largely rely on local signals, such as token probabilities, predictive entropy, or per-step confidence, and therefore overlook the long-range dependencies through which errors accumulate across an execution trajectory. As a result, they may fail to identify agent failures whose causes originate several reasoning or interaction steps before the final answer. We propose RUPA (Relational Uncertainty Propagation for Agents), a trajectory-level UQ framework for LLM agents. RUPA represents an execution history as a directed trajectory graph in which reasoning states, tool interactions, and environment feedback are nodes connected by temporal and semantic dependency edges. It then propagates uncertainty over this graph to capture how execution risk accumulates and transfers across interaction steps. The propagated signal is combined with trajectory-level behavioral features and goal-alignment information to produce a confidence estimate for the full agent trajectory. We evaluate RUPA on representative agent benchmarks, including τ-2, Terminal-Bench-2, and GAIA, using 6 open-source LLMs spanning multiple model families. Experimental results show that RUPA consistently outperforms existing UQ methods by providing more accurate uncertainty estimates, enabling earlier failure detection, and improving uncertainty-guided agent execution across diverse agent tasks. These results demonstrate that explicitly modeling relational dependency is crucial to reliable UQ for long-horizon LLM agents, providing a practical foundation for trustworthy agent execution.