Masterarbeit

This thesis investigates how agentic AI can autonomously assess the computational reproducibility of scientific studies. The system will use specialized agents to analyse papers, repositories, notebooks, dependencies, datasets and execution environments and collect evidence for reproducibility. The agents should be able to plan investigations, execute computational artifacts, diagnose failures and adapt their strategy based on the observed results. The thesis will evaluate whether an agentic approach can identify reproducibility barriers more accurately and with less human effort than conventional automated or LLM-based approaches. The resulting system could build on existing reproducibility concepts such as ReproScore while extending them from assessment toward autonomous investigation.