AnesTRACE: Benchmarking Intraoperative Anesthesia from Multimodal Perception to Multi-step Decision-Making

Multimodal clinical decision-making requires reliable reasoning over heterogeneous evidence from electronic health records, medical images, and physiological signals. Existing models typically map these inputs directly to diagnoses without explicitly assessing evidence sufficiency, tool-use requirements, or diagnostic uncertainty. This paper presents MedTRACE, a tool-augmented multimodal clinical reasoning agent for evidence-grounded decision-making. MedTRACE uses modality-specific encoders to construct a unified patient-state representation and performs an iterative loop of hypothesis formation, toolaware deliberation, and evidence verification. It dynamically invokes visual grounding, evidence retrieval, and structured parsing tools to locate diagnosis-relevant regions, retrieve clinical knowledge and similar cases, and extract structured findings. The acquired evidence enters an evidence memory, where a consistency verifier confirms or revises the current hypothesis. MedTRACE outputs a diagnosis together with supporting evidence, an auditable reasoning trace, and calibrated confidence. Experiments on multiple multimodal clinical diagnosis benchmarks show that MedTRACE improves diagnostic accuracy by 5.4% and AUROC by 4.7 percentage points over the strongest baseline. It also improves evidenceselection F1 by 8.2 percentage points and visual-grounding IoU by 6.5 percentage points, reduces expected calibration error by 31.6%, and decreases unsupported diagnostic errors by 27.8%. These results demonstrate that active evidence acquisition and verification improve the accuracy, interpretability, and reliability of multimodal clinical decisionmaking.

Safety benchmarks assume that test-condition behavior predicts deployment behavior, an assumption that fails if models detect evaluation cues and adapt. This opens a gap between benchmark performance and deployment behavior: compliance measured under test conditions becomes an optimistic upper bound that overstates how safely a model behaves once the evaluation harness is removed. We characterize this evaluation awareness through eight experiments across 37 open-weight models and seven families. (i)Detection is moderate and training-driven (24/37 models exceed chance, best AUROC 0.714 vs.0.819 human, with instruction tuning dominating over scale). (ii)Detection shifts safety behavior (hard refusal drops 5.8 percentage points under hypothetical framing, and 21/140 HarmBench framing effects are significant, with compliance rising up to +30 percentage points. (iii)Representations survive behavioral collapse (probes retain AUROC 0.98 under rewrites that drive behavior below chance, and multi-layer steering causally moves three downstream tasks while random controls do not). (iv)These axes are weakly coupled (only 1/15 correlations are significant, the sole robust link being behavioral detection versus framing resistance, $ρ=-0.79$, $p<0.001$). We call this gap the benchmark illusion: because detectability, behavioral manifestation, and controllability vary independently, it is multivariate rather than a single number, so no single awareness score is a reliable proxy for deployment safety.

Real-world audio-visual understanding requires chaining evidence that is sparse, temporally dispersed, and split across the visual and auditory streams, whereas existing benchmarks largely fail to evaluate this capability. They restrict videos to short clips, isolate modalities, or reduce questions to one-hop perception. We introduce TraceAV-Bench, the first benchmark to jointly evaluate multi-hop reasoning over long audio-visual trajectories and multimodal hallucination robustness. TraceAV-Bench comprises 2,200 rigorously validated multiple-choice questions over 578 long videos, totaling 339.5 hours, spanning 4 evaluation dimensions and 15 sub-tasks. Each question is grounded in an explicit reasoning chain that averages 3.68 hops across a 15.1-minute temporal span. The dataset is built by a three-step semi-automated pipeline followed by a strict quality assurance process. Evaluation of multiple representative OmniLLMs on TraceAV-Bench reveals that the benchmark poses a persistent challenge across all models, with the strongest closed-source model (Gemini 3.1 Pro) reaching only 68.29% on general tasks, and the best open-source model (Ming-Flash-Omni-2.0) reaching 51.70%, leaving substantial headroom. Moreover, we find that robustness to multimodal hallucination is largely decoupled from general multimodal reasoning performance. We anticipate that TraceAV-Bench will stimulate further research toward OmniLLMs that can reason coherently and faithfully over long-form audio-visual content.