> "no non-deterministic tests"

These tests can be made determenistic. Shrinking to the minimal reproducing case, is always deterministic. Also you can [re]start the tests with the specific seed - you typically log the seed, so you can replay the exact failure deterministically - prop-based tests are stochastic by design but reproducible by capture.

correct, I've done this at past places to verify rules engine output for variety of inputs.