Reproducibility
Same code, same data, different result. This should be alarming, and it usually traces to one of a small, well-understood set of non-determinism sources — reproducibility is layered, and the right response is deciding which layer actually matters before paying for the strictest possible guarantee.