runtime model extraction

Microservice evaluation has long suffered from the lack of representative benchmarks that simultaneously capture real-world scale, topology, and execution patterns. To address this, we propose Palette, a novel framework that introduces Graph Causal Models (GCMs) to microservice topology abstraction for the first time. Palette integrates call-branching probabilities, dependency ordering, and latency distributions to enable end-to-end generation of configurable macroscopic benchmarks directly from industrial-scale distributed trace data. Unlike conventional synthetic or simulation-based approaches, Palette-generated benchmarks—evaluated across multiple large-scale trace datasets—exhibit significantly higher fidelity, scalability, and reproducibility. This advancement substantially enhances the realism and practical utility of microservice system evaluation, enabling more rigorous and production-relevant benchmarking.

This study addresses the lack of systematic, large-scale analyses of structural properties in software feature models, which has hindered the understanding and evolution of variability models. For the first time, it systematically applies large-scale network analysis to 5,709 variability models drawn from 20 repositories. By constructing graphs capturing transitive dependencies and conflicts among features, and integrating graph modeling with network-theoretic and statistical analyses, the work uncovers cross-domain structural commonalities—such as dependency dominance, high centralization, and characteristic degree distributions—as well as domain-specific deviations. These findings provide novel empirical insights and a foundation for identifying pivotal features, guiding modular decomposition, and assessing structural fragility in variability-intensive systems.

Modern OLTP systems often suffer from frequent schema changes, missing primary/foreign keys, and fragmented execution traces, rendering traditional approaches—reliant on fixed schemas and manual modeling—costly and error-prone. This work proposes a fully automated pipeline that operates without predefined schemas by identifying quasi-key and timestamp columns, discovering inter-table relationships through statistical signals, and assembling and ordering events accordingly. To capture long-range dependencies across system events, the method incorporates a Temporal Convolutional Network (TCN). By eliminating dependence on ER diagrams, domain-specific templates, and stable schemas, the approach enables generalizable and scalable reconstruction of execution traces in dynamic information systems. Experimental results on TPC-H/E, synthetic, and real-world industrial datasets demonstrate 85% accuracy in event prediction and recovery of approximately 82% of true predecessor relationships, yielding high-fidelity process traces.

Highly variable event logs yield overly complex and poorly interpretable process models via automated discovery, while existing trace clustering methods largely neglect the probabilistic nature of activities and transitions, failing to capture real execution dynamics. This paper proposes a model-driven stochastic trace clustering method: grounded in stochastic process models, it introduces an entropy-based correlation measure derived from direct-follows probabilities and jointly optimizes trace assignment via structural alignment and generative likelihood. An efficient iterative algorithm ensures linear scalability. To our knowledge, this is the first approach to unify stochastic modeling with model-driven optimization in trace clustering, significantly enhancing control-flow pattern clarity and clustering quality. Extensive evaluation on multiple real-world datasets demonstrates superior behavioral representation accuracy and clustering stability over state-of-the-art methods, and reveals systematic effects of stochasticity on clustering performance ranking.

This work addresses the challenge that ROS 2 systems often embed their layered architecture implicitly within launch configurations, lacking an explicit, standalone architectural view and thereby hindering maintainability. To overcome this, the paper introduces the first approach that models the layered structure as a first-class architectural view in ROS 2. It proposes an automated recovery method combining deterministic parsing with a large language model (LLM) agent, where UML modeling and structural contracts constrain the LLM’s synthesis process, and architectural blueprints guide verifiable, high-fidelity reconstruction. Evaluation on three ROS 2 repositories—including an industrial-scale subset—demonstrates high precision across all abstraction levels, though recall for subsystems declines in complex systems due to the implicit semantics of launch configurations.

This work addresses the challenge that large language model (LLM) agents often produce redundant, exploratory, and non-deterministic execution trajectories that are difficult to reuse. To overcome this, the authors propose a skill-guided framework that extracts reusable structures from noisy trajectories and compiles them into near-deterministic workflows. The core innovations include a dependency inference mechanism based on evidence tuples—establishing strong dependencies only when parameters are uniquely traceable and flagging ambiguous relations as suspect—along with fine-grained binding-type categorization. The method integrates trajectory clustering, dependency rule mining, deterministic replay, and leave-one-out validation into a unified pipeline. Experiments demonstrate high precision (0.928) and recall (0.943) in dependency identification on the T1 dataset; for Venmo tasks, API calls are reduced from 34 to 11 while passing 15 of 21 test cases, and the system correctly rejects ill-posed or irreversible intents in Spotify and Todoist scenarios.

This study addresses a structural misalignment between producers and consumers of pretrained language models (PTLMs) on platforms like Hugging Face, which manifests as mismatches in model discovery, documentation, lineage tracing, and governance. Through surveys and qualitative analysis involving 50 model producers and 95 GitHub-based consumers, this work reveals significant discrepancies in how the two groups perceive the placement of critical metadata, motivations for lineage tracking, and priorities in model governance. These findings provide empirical grounding for improving model documentation standards, lineage-tracking tools, and governance frameworks, offering a novel perspective on optimizing the PTLM reuse ecosystem.

This work addresses the challenge of automatically recovering traceability links among software architecture documentation, models, and source code—a longstanding barrier to effective system maintenance and consistency assurance. To bridge this gap, we present the first end-to-end ecosystem for architecture-level traceability recovery, comprising a RESTful API supporting four distinct tracing pipelines, an interactive web-based frontend named TraceView, and TraceViz, an embedded visualization plugin for Visual Studio Code. The system integrates seamlessly into developer workflows through asynchronous task processing and caching optimizations, enabling intuitive exploration of traceability links directly within the IDE. All components are publicly deployed, and preliminary user studies indicate that TraceViz significantly enhances developers’ cognitive efficiency during software comprehension tasks.

Existing change impact analysis approaches rely solely on semantic similarity or structural dependencies, limiting their ability to comprehensively identify affected artifacts across heterogeneous software assets such as requirements, configurations, services, and tests. This work proposes a novel, training-free, and interpretable method that uniquely integrates semantic priors with multi-hop graph propagation. Specifically, it constructs a typed heterogeneous graph via static analysis, derives semantic priors from embedding-based cosine similarity, and diffuses impact through a row-normalized, decay-weighted propagation matrix controlled by a single parameter λ to balance precision and recall. Evaluation on five real-world change scenarios in a payment subsystem demonstrates the method’s capability to capture both structurally reachable yet textually disjoint artifacts and semantically related but structurally isolated ones, with demonstrated extensibility to operational assets such as container images and monitoring metrics.