[dchftcs]

The goals before the 4 years were probably based on results claimed, not making a product. Research management is difficult and efficiently running a lab requires some degree of trust that researchers don't lie.

I think the bigger part of management's fault here is that they should have spent some time on verifying the research before dedicating more resources.

[lumost]

Also setting up mechanisms to verify results and share results. Even a basic mechanism of “check in your code” would have served the purpose of identifying fake work. A code review mechanism can help stop irresponsible work.

[fshbbdssbbgdd]

“Check in your code” can identify certain kinds of blatant fraud, but it only does so much. For example, you can p-hack by evaluating your result in many different ways and then commit the version that gives the best number. It takes a lot of discipline to prevent this at an organization level. You can even have an emergent group form of this, where people collectively p-hack by individually evaluating their work and randomly finding lottery tickets. I’ve seen this phenomenon extend from research all the way to shipped product changes.