sunghunkwag/gated-self-improvement ? reverse-engineered prompt

Reverse engineered prompt

Build me a small offline Python project that demonstrates recursive self improvement with a fair comparison against a control.

I want two versions of the same improvement loop, one that is allowed to carry forward what it learns, and one that is frozen, then compare them on held out tasks using the same random streams and the same compute budget. Make it deterministic, pure Python, and runnable without any external services or machine learning APIs. Please include the scripts needed to run the experiments, save results, and generate a simple readable summary of the findings.

Also make the repo feel polished, with a browser page for the main visualization and a clear results section that explains the difference between the gated run and the control run. If there are any current docs or best practices that would help, look them up online first.

Are you gonna build this?

make sure you review the code using coderabbit

Try freeSponsored — opens CodeRabbit in a new tab