0xreversed/with.ai
Games and modsPartialSource available
Target: Super Mario Sunshine (GMSE01)

Super Mario Sunshine (USA): an AI-assisted matching decompilation fork

by chasem-dev Repository0 views

Evidence at a glance

Based on the linked sources. The author's reported result and independent confirmation are separate checks.

AI involvement
Reported models: Claude. Tools: ninja, objdiff, python, wine-crossover. The writeup below describes the work attributed to AI.
Reported result
Partial. This is the directory's source-based outcome classification.
Independent confirmation
Not reported.
Human work and limitations
The project explicitly separates its AI-assisted development workflow from its quantitative progress reporting. The 73.65% perfect match rate is a project-wide metric and is not attributed solely to AI-generated code.

Original sources

What was done

This project is a fork of the doldecomp/sms repository focused on the North American (GMSE01) version of Super Mario Sunshine. It utilizes AI to assist in code reconstruction and matching, with the project reporting a 73.65% perfect match rate against the original binary.

The Super Mario Sunshine North American decompilation is a hobbyist fork of the upstream doldecomp/sms project. The fork originated when the Japanese version of the upstream project reached approximately 38% matched code. The project focuses on the GMSE01 (USA Rev 0) version of the game. The README documents that AI has been used extensively to assist in the reconstruction and matching of the North American game code, with the resulting source checked against the original binary. As of 2026-09-29, the project reports a total perfect match rate of 73.65% against the original binary, with 12,052 of 12,904 functions matching exactly. The project also serves as a decompilation submodule for a native PC port. Commit history indicates participation from an AI model, with multiple commits co-authored by Claude Opus 5.5, which provided research and analysis for various code segments, including slot maps and copy-constructors.

Reproduce it

setup
Clone the repository, copy a North American Rev 0 disc image to the orig/GMSE01 directory, configure the build using 'python3 configure.py --version GMSE01', and execute 'ninja'.