Current progress and boundaries
Coding assistants and automated analysis tools can accelerate exploration, but researchers still need to confirm that data processing follows the study design and disciplinary standards.
Why it remains difficult
Undocumented data filtering, dependency versions, and multiple comparisons can change conclusions; code that appears to run successfully may still answer the wrong question.
Paths worth exploring
Require analyses to be delivered as executable scripts with locked environments. Record every transformation, exclusion criterion, and statistical decision as a reviewable step.
How to evaluate success
Have an independent person rerun results from the raw data and compare the main conclusions, sensitivity analyses, and environment reproducibility success rate.