[JTBooth]

Very different operating modes are required depending on how many people you're trying to coordinate. The allegory to software development is pretty clear - sometimes you're working on a personal project and zero documentation is tolerable, but sometimes your work is going to be translated into 200 languages and distributed to a hundred million users the day after it ships, and it needs ten times as many lines of tests as lines of code.

It seems to me that our model of "collect the data and analyze it yourself" is a sort of "ten engineer startup" scale process. Now that many fields of science have four or five digits of PhDs collaborating between countries, there's an increasing need for specialization, specifically in creating reusable data. It'll make us less efficient on a small scale, but creating any artifact at all that can be reliably used by ten thousand people is much higher leverage than a fully packaged data+conclusion that's even odds to be nothing but noise.