Yeah, the HAL itself does not help much for hardware issues. But if tooling was built for automated on-device tests (to test your HAL and app logic), then you have a much better chance of having a system which can help you reproduce spurious hardware failures. For instance using generative techniques like fuzzing on the software side, or just a predictable test-patterns which can be observed with scope/logic-analyzer. On the testbench side use temperature stressing, under/overclocking, under/overvolting, adding capacitance or inductance to signal or power lines, modifying grounding schemes, can be done to try to provoke the situation more reliably and (maybe) get a better understand of the problem.