Özdemir, Konrad, Kirchdorfer, Lukas, Amiri Eliasi, Keyvan et al. · Zenodo (CERN European Organization for Nuclear Research) · 2025
Business process simulation (BPS) is a key tool for analyzingand optimizing organizational workflows, supporting decision-making byestimating the impact of process changes. The reliability of such esti-mates depends on the ability of a BPS model to accurately mimic theprocess under analysis, making rigorous accuracy evaluation essential.However, the state-of-the-art approach to evaluating BPS models hastwo key limitations. First, it treats simulation as a forecasting prob-lem, testing whether models can predict unseen future events. This failsto assess how well a model captures the as-is process, particularly whenprocess behavior changes from train to test period. Thus, it becomes diffi-cult to determine whether poor results stem from an inaccurate model orthe inherent complexity of the data, such as unpredictable drift. Second,the evaluation approach strongly relies on Earth Mover’s Distance-basedmetrics, which can obscure temporal patterns and thus yield misleadingconclusions about simulation quality. To address these issues, we pro-pose a novel framework that evaluates simulation quality based on itsability to generate representative process behavior. Instead of compar-ing simulated logs to future real-world executions, we evaluate whetherpredictive process monitoring models trained on simulated data performcomparably to those trained on real data for downstream analysis tasks.Empirical results show that our framework not only helps identify sourcesof discrepancies but also distinguishes between model accuracy and datacomplexity, offering a more meaningful way to assess BPS quality.
Cited by 0Open access
Manuel F. Schmid, Marco G. Giometto, Gregory A. Lawrence et al. · Geoscientific model development · 2024
Abstract. We present BoundaryLayerDynamics.jl, a new code for turbulence-resolving simulations of atmospheric boundary-layer flows as well as canonical turbulent flows in channel geometries. The code performs direct numerical simulation as well as large-eddy simulation using a hybrid (pseudo)spectral and finite-difference approach with explicit time advancement. Written in Julia, the code strives to be flexible and adaptable without sacrificing performance, and extensive automated tests aim to ensure that the implementation is and remains correct. We show that the simulation results are in agreement with published results and that the performance is on par with an existing Fortran implementation of the same methods.
Cited by 4Open access
Qin, Tianjian · Zenodo (CERN European Organization for Nuclear Research) · 2025
Cited by 0Open access
Shlyk, Darya · Zenodo (CERN European Organization for Nuclear Research) · 2025
Cited by 0Open access
Debora Barros · Zenodo (CERN European Organization for Nuclear Research) · 2026
This release contains the scripts and processing routines used for the analysis of hydrographic and microstructure data presented in the study "Barros et al. 2026" Key contents: Microstructure data processing and ϵ calculation. Stability metrics and dimensionless numbers (Fri, Ri, Reb). Data visualization and figure generation scripts.
Cited by 0Open access
Philipp Eibl, Sadra Sabouri, Souti Chattopadhyay · ArXiv.org · 2025
Recent AI code assistants have significantly improved their ability to process more complex contexts and generate entire codebases based on a textual description, compared to the popular snippet-level generation. These codebase AI assistants (CBAs) can also extend or adapt codebases, allowing users to focus on higher-level design and deployment decisions. While prior work has extensively studied the impact of snippet-level code generation, this new class of codebase generation models is relatively unexplored. Despite initial anecdotal reports of excitement about these agents, they remain less frequently adopted compared to snippet-level code assistants. To utilize CBAs better, we need to understand how developers interact with CBAs, and how and why CBAs fall short of developers' needs. In this paper, we explored these gaps through a counterbalanced user study and interview with (n = 16) students and developers working on coding tasks with CBAs. We found that participants varied the information in their prompts, like problem description (48% of prompts), required functionality (98% of prompts), code structure (48% of prompts), and their prompt writing process. Despite various strategies, the overall satisfaction score with generated codebases remained low (mean = 2.8, median = 3, on a scale of one to five). Participants mentioned functionality as the most common factor for dissatisfaction (77% of instances), alongside poor code quality (42% of instances) and communication issues (25% of instances). We delve deeper into participants' dissatisfaction to identify six underlying challenges that participants faced when using CBAs, and extracted five barriers to incorporating CBAs into their workflows. Finally, we surveyed 21 commercial CBAs to compare their capabilities with participant challenges and present design opportunities for more efficient and useful CBAs.
Cited by 0Open access
Aristidis Vasilopoulos · arXiv (Cornell University) · 2026
LLM-based agentic coding assistants lack persistent memory: they lose coherence across sessions, forget project conventions, and repeat known mistakes. Recent studies characterize how developers configure agents through manifest files, but an open challenge remains how to scale such configurations for large, multi-agent projects. This paper presents a three-component codified context infrastructure developed during construction of a 108,000-line C# distributed system: (1) a hot-memory constitution encoding conventions, retrieval hooks, and orchestration protocols; (2) 19 specialized domain-expert agents; and (3) a cold-memory knowledge base of 34 on-demand specification documents. Quantitative metrics on infrastructure growth and interaction patterns across 283 development sessions are reported alongside four observational case studies illustrating how codified context propagates across sessions to prevent failures and maintain consistency. The framework is published as an open-source companion repository.
Cited by 0Open access
Aristidis Vasilopoulos · Zenodo (CERN European Organization for Nuclear Research) · 2026
LLM-based agentic coding assistants lack persistent memory: they lose coherence across sessions, forget project conventions, and repeat known mistakes. Recent studies characterize how developers configure agents through manifest files, but an open challenge remains how to scale such configurations for large, multi-agent projects. This paper presents a three-component codified context infrastructure developed during construction of a 108,000-line C# distributed system: (1) a hot-memory constitution encoding conventions, retrieval hooks, and orchestration protocols; (2) 19 specialized domain-expert agents; and (3) a cold-memory knowledge base of 34 on-demand specification documents. Quantitative metrics on infrastructure growth and interaction patterns across 283 development sessions are reported alongside four observational case studies illustrating how codified context propagates across sessions to prevent failures and maintain consistency. The framework is published as an open-source companion repository.
Cited by 0Open access
Suman Patel, Kanchan Bala, Aknan Mohammad · Zenodo (CERN European Organization for Nuclear Research) · 2025
Manually reviewing large and complicated codes is time-consuming and labor-intensive. It takes a lot of time and effort to read huge and complex code projects and understand them. Such an effort is a time-consuming and not efficient process. Modern software contains thousands of files; hence, it is very difficult to understand its architecture, features, and key components easily. To address these challenges, there is a need for an automated system that can summarize, clone, and analyze GitHub repositories. Such a system should be capable of handling large and complex repositories while reducing the amount of human effort required. This paper presents an automated solution using a multi-agent system powered by Crew AI to summarize and analyze GitHub repositories. This system employs specialized agents that collaboratively clone repositories, parse key files (e.g., README, source code), and extract relevant functional and architectural insights. The backend is implemented in Python, using GitPython for repository management and large language models via OpenAI for semantic analysis. These agents operate in parallel to perform source code inspection, document analysis, and report generation. The output is human-readable, easily understandable, and structured, presented in multiple formats, such as PDF, Markdown, or JSON. This framework significantly reduces manual efforts and enables the creation of a scalable codebase understanding, making it a practical tool for developers, researchers, and automated documentation pipelines.
Cited by 0Open access
Aristidis Vasilopoulos · Zenodo (CERN European Organization for Nuclear Research) · 2026
LLM-based agentic coding assistants lack persistent memory: they lose coherence across sessions, forget project conventions, and repeat known mistakes. Recent studies characterize how developers configure agents through manifest files, but an open challenge remains how to scale such configurations for large, multi-agent projects. This paper presents a three-component codified context infrastructure developed during construction of a 108,000-line C# distributed system: (1) a hot-memory constitution encoding conventions, retrieval hooks, and orchestration protocols; (2) 19 specialized domain-expert agents; and (3) a cold-memory knowledge base of 34 on-demand specification documents. Quantitative metrics on infrastructure growth and interaction patterns across 283 development sessions are reported alongside four observational case studies illustrating how codified context propagates across sessions to prevent failures and maintain consistency. The framework is published as an open-source companion repository.
Cited by 0Open access
Grandin, Noel · TIB KMO / FLOWWORKS GmbH · 2019
Clang plugins in LibreOffice - global analyses across a large codebase In the old days, we expended considerable effort to make our global analyses more efficient, especially when dealing with large codebases like LibreOffice. However, with the advent of modern hardware and software tools this has become largely necessary. This talk will focus on some pragmatic design choices that lead to being able to run global analysis passes across the LibreOffice codebase, without breaking the bank on hardware or waiting days for results. Also, I discuss gradual approaches to developing such plugins that do require language-lawyer level knowledge of the C++ AST.
Cited by 0Open access
Aristidis Vasilopoulos · Zenodo (CERN European Organization for Nuclear Research) · 2026
Initial release accompanying paper upload LLM-based agentic coding assistants lack persistent memory: they lose coherence across sessions, forget project conventions, and repeat known mistakes. Recent studies characterize how developers configure agents through manifest files, but an open challenge remains how to scale such configurations for large, multi-agent projects. This paper presents a three-component codified context infrastructure developed during construction of a 108,000-line C# distributed system: (1) a hot-memory constitution encoding conventions, retrieval hooks, and orchestration protocols; (2) 19 specialized domain-expert agents; and (3) a cold-memory knowledge base of 34 on-demand specification documents. Quantitative metrics on infrastructure growth and interaction patterns across 283 development sessions are reported alongside four observational case studies illustrating how codified context propagates across sessions to prevent failures and maintain consistency. The framework is published as an open-source companion repository. Full Changelog: https://github.com/arisvas4/codified-context-infrastructure/commits/v1.0.0
Cited by 0Open access
Gerard, Leon, Scherbela, Michael, Grohs, Philipp · Figshare · 2023
The following checkpoints contain the model weights corresponding to the GitHub DeepErwin (https://github.com/mdsunivie/deeperwin). They contain pre-trained weights for the following: <br> 1. A PhisNet reimplementation to generate orbital descriptors for a neural wavefunction. 2. A pre-trained neural wavefunction on a dataset of 18 compounds with Hartree-Fock orbital descriptors. 3. A pre-trained neural wavefunction on a dataset of 98 compounds with PhisNet orbital descriptors. <br> The file "tinymol_v10_2023-02-26_18x20_500k.zip" accompanies the publication "Towards a Foundation Model for Neural Network Wavefunctions", Scherbela, M., Gerard, L. and Grohs, P. (https://arxiv.org/abs/2303.09949).
Cited by 0Open access
Gerard, Leon, Scherbela, Michael, Grohs, Philipp · Figshare · 2023
The following checkpoints contain the model weights corresponding to the GitHub DeepErwin (https://github.com/mdsunivie/deeperwin). They contain pre-trained weights for the following: <br> 1. A PhisNet reimplementation to generate orbital descriptors for a neural wavefunction. 2. A pre-trained neural wavefunction on a dataset of 18 compounds with Hartree-Fock orbital descriptors. 3. A pre-trained neural wavefunction on a dataset of 98 compounds with PhisNet orbital descriptors. <br> The file "tinymol_v10_2023-02-26_18x20_500k.zip" accompanies the publication "Towards a Foundation Model for Neural Network Wavefunctions", Scherbela, M., Gerard, L. and Grohs, P. (https://arxiv.org/abs/2303.09949).
Cited by 0Open access
Drammis, Sabrina, Jiajia Zhao, Qzhang21 · Zenodo (CERN European Organization for Nuclear Research) · 2020
Striosomes Mediate Valence Learning Disrupted by Age and Huntington's Disease Model
Cited by 0Open access
Timothy Ho · Zenodo (CERN European Organization for Nuclear Research) · 2025
This release contains the complete codebase, dataset, and analysis materials for the Master's Thesis titled "Integrating Cognitive Control Paradigms Into a Unified Experimental Design Space: A Parametric Analysis" by Timothy Ho, submitted to the Institute of Cognitive Science, Universität Osnabrück. This version (v1.0.0) represents the exact state of the project as submitted on September 29, 2025. It is intended to be a permanent, static archive to ensure the reproducibility of the methods, analyses, and findings presented in the thesis. Contents This repository includes: Literature-coded dataset (data/super_experiment_design_space.csv) containing cognitive control experiments systematically coded along derived dimensions Python data pipeline (convert.py, analysis_utils.py) for translating high-level derived dimensions into absolute timing parameters Interactive web viewer (index.html, viewer.js) for visualizing and simulating experimental paradigms PCA and MOFA analyses (Jupyter notebooks) exploring the latent structure of the design space Conceptual documentation (content/ Obsidian vault) defining the theoretical framework, coding guide, and derived dimensions Test suite (viewer.test.js, test_convert.py, test_utils.py) ensuring correctness of the operational layer Citation If you use this work, please cite: Ho, T. (2025). Integrating Cognitive Control Paradigms Into a Unified Experimental Design Space: A Parametric Analysis. Master's Thesis, Institute of Cognitive Science, Universität Osnabrück. For ongoing development and the most recent version of this project, please see the main branch of this repository.
Cited by 0Open access
Manuel F. Schmid, Marco G. Giometto, Gregory A. Lawrence et al. · 2023
Abstract. We present BoundaryLayerDynamics.jl, a new code for turbulence-resolving simulations of atmospheric boundary-layer flows as well as canonical turbulent flows in channel geometries. The code performs direct numerical simulation as well as large-eddy simulation using a hybrid (pseudo)spectral and finite-difference approach with explicit time advancement. Written in Julia, the code strives to be flexible and adaptable without sacrificing performance, and extensive automated tests aim to ensure that the implementation is and remains correct. We show that the simulation results are in agreement with published results and that the performance is on par with an existing Fortran implementation of the same methods.
Cited by 0Open access
Andersen, Nicklas Sindlev · Zenodo (CERN European Organization for Nuclear Research) · 2022
A codebase containing algorithm implementations for solving the dynamic origin-destination estimation problem.
Cited by 0Open access
Shlyk, Darya · Zenodo (CERN European Organization for Nuclear Research) · 2025
Cited by 0Open access
Talvik, Adeline · DSpace repository (University of Tartu) · 2025
Over the years, the Java programming language has been updated with many code constructs which make programming easier and reduce boilerplate code, such as record patterns and pattern matching for switches. The goal of this Bachelor's thesis was to modernise the codebase for the Java-based course “Automata, Languages and Compilers”. This included replacing the Visitor design pattern with alternatives, converting classes to records, converting abstract classes to sealed interfaces and updating methods. By the end of the modernisation process, the line count of the code repository was reduced by approximately 4000, despite the added files and documentation.
Cited by 0Open access