The way you handle inconsistent failures IRL is the same way you handle it in software: Run a hundred instances of the test. Measure the failure rate. Make the

The way you handle inconsistent failures IRL is the same way you handle it in software: Run a hundred instances of the test. Measure the failure rate. Make the change. Re-measure. In bash it might be a for loop running the test a hundred times, in person it might getting a hundred phones.