agbcc: SLaDe-AGBCC: Neural Matching Decompilation

A research project called SLaDe-AGBCC is testing whether large language models can perform matching decompilation of Game Boy Advance agbcc assembly, optimizing for C code that recompiles to byte-identical output rather than readable structure. The project, which credits Claude Fable 5 for co-authorship on implementation and diagnostic tasks, treats C recovery as a folding problem inspired by AlphaFold 2 and uses a compiler-in-the-loop recycling signal, with recent v4 recycling experiments showing progress in loop-based matching. The author states the core hypothesis — that transformers can defeat the agbcc optimizer to achieve byte-exactness — is being tested rather than assumed, drawing on a curated corpus of GBA decompilation projects and decomp.me data.

Evidence at a glance Loading source evidence… What was done SLaDe-AGBCC is a research project using LLMs to perform matching decompilation on GBA agbcc assembly. It treats C recovery as a folding problem, aiming for byte-identical output via compiler-in-the-loop recycling. SLaDe-AGBCC is a research project investigating neural matching decompilation for the Game Boy Advance. Unlike traditional decompilers, it does not prioritize readable structure or name recovery; instead, it optimizes for producing C code that recompiles to byte-identical assembly when processed by the agbcc compiler. The project draws inspiration from AlphaFold 2, treating C recovery as a folding problem and utilizing a compiler-in-the-loop recycling signal to refine predictions. The author reports that the core hypothesis—that transformers can defeat the agbcc optimizer to achieve byte-exactness—is currently being tested rather than assumed. Development involves an automated pipeline where disassembled objects are converted to C and iteratively refined. The project relies on a curated corpus of GBA decompilation projects and decomp.me data. Recent experiments include v4 recycling, which demonstrated progress in loop-based matching, though the author notes that recycling must be explicitly trained. The project documentation credits Claude Fable 5 for co-authorship on various implementation and diagnostic tasks. Sources: Source 1 https://raw.githubusercontent.com/jrickey/slade-agbcc/HEAD/README.md Reproduce it uv venv && uv pip install -e . && cd vendor/agbcc && ./build.sh && cd - && python scripts/measure corpus.py Community Ask a question, share what worked, or follow the next milestone. Your library https://reversedwithai.com/library