Definition
Reproducibility is the property that other researchers can recreate your results from your code, data, and procedure. In modern ML it’s under quiet but constant threat from undisclosed data, closed-weight models, and runs that nobody is going to rerun on a thousand GPUs.