Reverse-Engineering Constraint Models from CNFs: An Agentic Approach (Supplementary Material: code, benchmarks, and results)

Supplementary material for "Reverse-Engineering Constraint Models from CNFs: An Agentic Approach" (PoS 2026). Source code, benchmark instances, and experimental results for an LLM agent that reverse-engineers a semantically equivalent PySAT encoding from a DIMACS CNF file. Includes the agent runners, the recon analysis library, the SAT-based equivalence checker, 48 cross-encoder instances over 12 problems (PySAT/Picat/Sugar/BEE) plus scaling instances, and the full reconstruction results. See README.md for details.