> Replace libraries, languages and products with statistics, hypotheses and figures and you have described scientific research.
Well, yes - and the inability to build on previous research code because people don't publish it results in a colossal waste of (usually) public money. Nonreproducibility in science, and Google's abandoning of products, are phenomena that resemble one another somewhat.
Replace libraries, languages and products with statistics, hypotheses and figures and you have described scientific research.
The point is that each iteration learns from the last, in principle, even if they aren’t backwards or forwards compatible.