Promon Deobfuscation Benchmark

PDB-LLM: Promon Deobfuscation Benchmark for LLM Evaluation A benchmark dataset of 200 C programs for evaluating large language model deobfuscation capabilities against compiler-level obfuscation. Contents:- 150 general-domain programs: string↔number conversion, string manipulation, arithmetic operations, input validation- 50 security-domain programs: jailbreak detection, root detection, anti-debugging, integrity verification, hook detection Design characteristics:- 15-70 lines of code per program (fits LLM context windows)- Generic function naming (function_1, function_2) to prevent memorization- Ground-truth test oracles for functional validation- GCC-compatible (compiled with gcc -std=c99) Intended for research on AI-assisted reverse engineering and obfuscation resilience evaluation. Associated paper: "Evaluating OLLVM Obfuscation Resilience Against AI-Assisted Deobfuscation" (2026)