See my blog post on substack for details.
The main script is arc_solve/solve.py. This requires a redis server running at port "6381" (for caching), several hours of time, a considerable amount of memory, an openai key, and a bunch of money.
Required dependencies are tqdm, numpy, scipy, skimage, attrs, cattrs, nest_asyncio, redis-py, matplotlib, anthropic (not actually needed), and openai==0.28.1. (I might make an actual requirements.txt file later.)
Data can be loaded from jsons which can be found at this drive link and visualized and viewed using arc_solve/load_and_viz.py. (I may add additional plots and display later.)