Note that the original article doesn't disavow all unit tests; nor does it disavow all testing. Sounds to me like the kind of tests you're describing aren't ne

Note that the original article doesn't disavow all unit tests; nor does it disavow all testing. Sounds to me like the kind of tests you're describing aren't necessarily unit tests (system-level race conditions aren't typically discoverable with a simple unit test); where the tests were truly unit-level, the proposed alternative (assertions) may have been even more informative. Finally - a few real unit tests for known correct behavior are advised where in essence the algorithm can be described independently from its outcome.