scenario generation

Existing approaches struggle to effectively quantify the similarity and quality between synthetic and real data in evaluating tool-augmented agents. To address this gap, this work proposes SynAE, a novel framework that establishes the first multi-axis evaluation system tailored for multi-turn tool-use scenarios. SynAE introduces four fine-grained metric categories—assessing task instructions, tool invocations, final outputs, and downstream evaluation performance—to systematically measure synthetic data across dimensions of validity, fidelity, and diversity. Integrating natural language processing, trajectory modeling, and controllable generation techniques, the framework enables a reproducible evaluation pipeline and successfully identifies several representative failure modes in synthetic data generation. Empirical results demonstrate that such multidimensional assessment is essential for enhancing the reliability of agent evaluations.

This work proposes a novel approach to black-box testing of Functional Mock-up Units (FMUs) by integrating large language models (LLMs) with a human-in-the-loop mechanism. Addressing the inefficiency and poor interpretability of traditional FMU-based dynamic simulation testing—which relies on manually crafted scenarios—the method automatically generates structured Given-When-Then test objectives from FMU interface and functional specifications, and constructs complete test plans comprising input sequences and assertion oracles. Upon simulation execution, the framework produces visualizable logs and statistical evaluation metrics. The approach significantly enhances test design efficiency and result interpretability, facilitates test asset reuse, and demonstrates effectiveness on a lubricating oil cooling system by autonomously generating executable test scenarios and delivering objective-level pass-rate analysis.

Existing evaluation of generative AI is hindered by the scarcity of high-quality benchmarks, whose manual construction is costly and time-consuming. Method: We propose the first automated benchmark construction framework powered by collaborative large language model (LLM) agents, decomposing benchmark creation into four sequential stages—planning, generation, verification, and evaluation—integrating task decomposition, agent coordination, human-in-the-loop feedback, and explicit constraint-satisfaction assessment. Contribution/Results: The framework significantly enhances data diversity and metric reliability. Leveraging it, we construct the first high-quality benchmark specifically targeting planning and constraint-satisfaction capabilities in text generation. We systematically evaluate seven state-of-the-art models, uncovering shared failure modes and fine-grained capability disparities. Our work establishes a scalable, reproducible paradigm for evaluating generative AI capabilities, advancing both benchmark methodology and empirical analysis.

To address the “simulation-to-reality gap”—the difficulty of reproducing simulation-identified failure scenarios in real-world autonomous driving—this paper proposes a verification method based on formal scenario modeling and time-series matching. The method formally translates abstract scenario programs written in the Scenic probabilistic programming language into computable temporal matching rules, enabling precise retrieval of failure-relevant patterns from large-scale real-world sensor data. A key contribution is the design of an efficient, linearly scalable query algorithm that supports real-time pattern matching over long temporal sequences. Experimental evaluation demonstrates that the approach achieves higher recall accuracy for critical failure scenarios than state-of-the-art commercial vision-language models, while accelerating query throughput by several orders of magnitude. This significantly improves both the efficiency and trustworthiness of transferring simulation-discovered failures to real-vehicle validation.

Traditional static test sets inadequately evaluate foundation models’ diverse capabilities in open-ended scenarios. To address this, we propose ONEBench—a dynamic, extensible benchmarking paradigm that enables on-demand generation of customized evaluation suites targeting open capabilities, framing model assessment as a collective selection and aggregation process over sample-level tests. Our key contributions include: (1) the first unified, open-ended, and evolvable evaluation framework operating at the sample level; (2) a sparse measurement aggregation algorithm, a progressive sample pool construction mechanism, and a cross-modal unified interface (ONEBench-LLM/LMM); and (3) a robustness-aware scoring model with theoretical guarantees on identifiability and fast convergence. Experiments show that ONEBench achieves ranking stability >0.98 under 95% measurement sparsity, reduces evaluation cost by 20×, and attains >0.98 correlation with mean-score rankings on homogeneous data—enabling unified, efficient, and reliable assessment of both language and multimodal models.

This work addresses the challenge of systematically evaluating concept bottleneck models, whose applicability and failure mechanisms remain poorly understood due to the scarcity of real-world datasets with annotated concept labels. To bridge this gap, we introduce the first controllable synthetic benchmark that leverages parametric generation techniques to precisely modulate data modality, concept selection, annotation quality, and label completeness, thereby simulating diverse real-world relationships between concepts and predictions. This benchmark enables comprehensive evaluation of various concept bottleneck models across both decision-support and fully automated tasks, effectively identifying key performance determinants and characteristic failure modes. Our framework fills a critical void in the current evaluation landscape for concept-based interpretability methods.

Current benchmarks for medical large language models struggle to accurately predict real-world deployment performance, primarily because they fail to explicitly model the implicit assumptions underlying user–model interactions. This work proposes a conceptual distinction between “task assumptions,” which can be validated from dialogue data, and “outcome assumptions,” which require integration of outcome data and behavioral studies. To systematically evaluate how these assumptions contribute to the benchmark–deployment performance gap, the authors introduce BenchmarkCards—a novel framework coupled with a phased evaluation protocol. Retrospective clinical trial analyses and behavioral studies in case applications demonstrate that both types of assumptions contribute comparably to performance discrepancies, thereby validating the framework’s operational feasibility and practical utility, and significantly enhancing benchmark validity.

This study addresses the limitations of current robotic system validation, which relies heavily on manual selection of test scenarios, thereby hindering scalability and compromising reproducibility and reliability of conclusions. To overcome these challenges, this work proposes a compositional, scenario-based modeling approach that integrates declarative test specifications, plugin-driven scenario generation, containerized parallel simulation, and unified result analysis to establish the first modular and scalable automated verification framework. The framework enables systematic parameter variation across multiple dimensions and facilitates robust identification of systemic faults versus stochastic anomalies. Evaluated across 5,480 distinct scenario configurations with over 100,000 simulation runs, the approach accumulated 1,800 hours of simulated operation and 1,873 virtual kilometers, demonstrating its efficacy in discerning consistent system deficiencies from random irregularities.

Current evaluations of large language model (LLM) agents for database administration lack production-level fidelity and suffer from inconsistent tasks, platforms, and metrics, hindering meaningful cross-study comparisons. To address this, this work proposes DBA-Bench, a high-fidelity benchmark for database operations that introduces, for the first time, production-grade realism, outcome-oriented evaluation, and controllable scenario replay, covering four critical real-world operational gaps. Built on PostgreSQL instances under active workload, DBA-Bench integrates multi-source time-series and log observations, snapshot rollback, deep annotations, and an automated evaluation framework, encompassing 106 scenarios across seven task categories. In 848 automated runs, the best-performing agent achieved a safety pass rate of only 17.9%, starkly lower than human DBAs’ 93.4%, exposing significant limitations of current approaches in complex, safety-critical repair tasks.

This work addresses the limitation of existing agent evaluation benchmarks, which fail to capture the authentic interaction dynamics between developers and AI agents and inadequately reflect the distribution, diversity, and complexity of real-world tasks. To bridge this gap, the authors introduce a dynamic benchmark constructed from a large-scale corpus of real OpenClaw developer sessions. By leveraging conversation reconstruction, execution environment snapshots, and a deterministic validation mechanism, raw user requests are transformed into reproducible, automatically evaluable tasks. Sampling bias is mitigated through Jensen–Shannon divergence–based control to preserve the true task distribution. The resulting benchmark comprises 281 executable tasks; evaluations of 14 leading models reveal that even the best-performing system solves only 65.8% of them, highlighting a significant capability gap in current agents when operating in realistic software development scenarios.