Academic peer review works reasonably well for traditional research papers, but breaks down when applied to software-heavy research. Reviewers lack the time and incentive to thoroughly examine poorly-written research code, yet simply running the software isn't sufficient since most scientific code is simulation-based and could theoretically fake results. The challenge is compounded by the reality that research code is typically written by domain experts without formal software engineering training, and requiring researchers to become expert programmers isn't feasible given the already lengthy path to becoming a scientist.
Sort: