lightsighter

I agree with you, but if I were an admissions officer I would also like to see the SAT/ACT be two different numbers: one being your score when you took it as a freshman and a second from late in your junior year so I can see the delta. If you're a great test taker and both scores are equally high I can see that, but I can also see the case where someone with a very low initial score might have improved it greatly to an average score and I can infer that they might not have the best IQ, but that they are at least working hard at improving, which is also a valuable signal from an admissions perspective. Maybe we should standardize when the tests are given and that there be multiple scores recorded over time.

Just as counter example, all the assignments for my OS class at top-20 CS university were graded using a set of a few hundred tests that were hidden from us. For example, the first assignment was to implement a thread scheduler and over the years the course had developed a few hundred tests on which our grade was based. We were allowed to submit our solution three times, and we would be told how many tests we passed, but not why we failed any specific tests or what those tests were, with the highest of our submissions being our grade. The incentive of this process was two fold: first, it forced us to learn to write our own good tests and not rely on the TAs to spoon feed us exactly what we needed to do (in fact one student wrote a test that broke the reference solution and was added to the test suite for the next semester), and second it is completely intractable to expect a few TAs to scan through several thousand lines of code for each of a hundred students to discover how "close" to being correct their implementation is. Bugs are hard enough to find in code you wrote, let alone code written by an undergraduate just learning how to write systems code for the first time. In my experience TAing for both math and CS classes, finding bugs in students math assignments is significantly easier than finding bugs in code.

Actually in a lot of new HPC codes for doing physical simulation we are now starting to incorporate Uncertainty Quantification (UQ)[1] directly into the code such that we can determine where the instabilities and bifurcation points exist in non-linear systems. UQ is really expensive to compute, but it at least allows us to know when our simulations are accurate or are sensitive to perturbations.

[1] https://en.wikipedia.org/wiki/Uncertainty_quantification

To be fair, as someone with Celiac disease, living in that world might be a little easier for me than living in present day America (minus the obvious implications for the end of society as we know it).

Still in development (paid for by LANL, done by PGI):

https://developer.arm.com/-/media/developer/developers/hpc/f...