Sherlock4Py MaxSAT-Based Misbehaviour Verification and Localisation Framework for Python
- Coordinator
- Agencia Estatal Consejo Superior De Investigaciones Cientificas, Spain
- Runs
- 1 April 2026 to 31 March 2028
- Status
- Running
- EU funding
- €194k
- Total cost
- —
- Partners
- 1 organisations in 1 countries
- Type
- Marie Skłodowska-Curie Action
- Grant agreement
- 101269051
What the project is doing
In the project's own words, from its summary on CORDIS.
Debugging remains one of the most costly and time-consuming phases of software development. Formula-based fault localisation (FBFL), which leverages bounded model checking and logical reasoning through Maximum Satisfiability (MaxSAT), to identify faulty program statements, has shown strong potential for languages such as C, but remains under-explored for Python, which is the dominant language in artificial intelligence, data science, and education.
The aim of this project is to design and implement Sherlock4Py, a MaxSAT-based misbehaviour verification and localisation framework for Python. First, Sherlock4Py will implement FBFL to Python by incorporating bounded model checking with ESBMC-Python, and will develop scalable MaxSAT algorithms tailored to program-derived formulas. These algorithms will exploit structural features of Python, such as loop iterations and branching constructs, to achieve greater scalability and more precise diagnoses than existing approaches.
Second, Sherlock4Py will demonstrate the synergy between FBFL and Large Language Models (LLMs) by embedding MaxSAT-based bug localisation into counterexample-guided inductive synthesis (CEGIS) loops, thereby enabling LLMs to synthesise and repair faulty Python programs more reliably, with greater accuracy and efficiency. By bridging Automated Reasoning and Machine Learning, Sherlock4Py will advance the state of the art in both MaxSAT-based fault localisation and trustworthy AI.
The project is ambitious in delivering the first MaxSAT-based toolchain for Python, advancing MaxSAT algorithms with program-specific heuristics, and integrating symbolic verification with LLM-driven code synthesis. The expected outcome is a step change in software reliability, with direct benefits for AI research, software engineering practice, and computer science education.
Topics
Who is involved
| Organisation | City | Country | Role |
|---|---|---|---|
| Agencia Estatal Consejo Superior De Investigaciones Cientificas | Madrid | Spain | Coordinator |
Source: CORDIS, European Commission. Data of 5 Nov 2026, reused under the Commission's reuse policy. View on CORDIS · Grant DOI