Perhaps this could work well with results based on publicly available data sets (like the Protein Data Bank at www.rcsb.org), but I'd be surprised if this isn't

Perhaps this could work well with results based on publicly available data sets (like the Protein Data Bank at www.rcsb.org), but I'd be surprised if this isn't done already to a significant degree. Labs post programs (or their biological equivalents, protocols) on their websites all the time. Don't forget that data can be polished as well - for a hypothetical cheater, you'd simply be moving the "messaging" work from the finished project to the source files you mention above. There's a lot more to trusting a paper then just trusting the figures; you're trusting the scientific process that a lab is undertaking, and it's almost always impractical to require proof of every stage of that process. Another downside is that not every scientist is ok with giving access to all of their hard-earned data - that opens up analyses and potential scoops to other groups. This is especially true in biomedical/clinical studies, where data is very expensive to gather and can take years to analyze. Not saying raw data is never shared, but there is a balance there. Finally, scientists greatly prefers testing a published claim by pursuing a follow-up hypothesis rather then redoing an existing analyses. That way you kill two birds with one stone. It's inevitable that some irreproducible results slip in with this system, but if the claims are significant enough sooner or later someone who's trying to build on the claims (and failing) will go looking for more details and the fraud is exposed. The fear of being exposed in that way is what keeps the system honest, and to a large extent this system works.