Researchers built an AI system that recovers lost simulation code from research papers — 40% success rate, 10X faster than humans

A team created an automated system that reads control systems research papers and rebuilds the executable code that researchers originally used to generate their results. The system works by having an AI read the paper, write code, run it, compare outputs to published figures, and iterate until the simulation matches — cutting the time to verify published research from weeks of manual work to days of automated checking.

Right now, when someone wants to verify a published control systems paper or build on it, they often hit a wall: the paper doesn't include enough detail about parameters, settings, or implementation choices to reconstruct the original simulation. Researchers spend weeks reverse-engineering the code by hand. This system automates that reverse-engineering and succeeds 40% of the time, which means the bottleneck of replicating research just got much faster. The real implication is that published papers become more verifiable and more reusable without waiting for the original authors to release code they often never saved.