I was pretty excited when I heard that Microsoft was getting into scientific computing. As the world's biggest desktop software company, I figured they might understand that scientific computing and high-performance computing are not automatically the same thing, and that reliability and reproducibility are more important than peak performance. Turns out I was wrong: the workshop I attended last September was dominated by discussion of topics like GPU programming and computational grids that are still bleeding-edge computer science, rather than the nuts and bolts that would actually help most scientists be productive day-to-day, Microsoft's new HPC++ Computational Finance lab's site has a lot on speed but nothing on correctness, et cetera. So where should they be spending their time? If I ran the world, they'd start by reading Buckheit and Donoho on reproducible research, double back to Jon Claerbout's notes on the same, check out the Madagascar project, and then try to figure out how to scale up those ideas to hundreds of thousands of scientists and publications in as diverse a range of fields as possible. It won't give the senator something to stand beside on opening day, but it'll do science a lot more good.