Context Navigation

Changes between Initial Version and Version 1 of Ticket #19875, comment 10

-                      initial
+                      v1
 > I'm also a big fan of pastly's notion of tests -- it might take quite a bit of testing infrastructure to test this one well.
 Typically, we implement tests with random outcomes by working out what the probability of a random bit flip is, and then run the test enough times that the failure rate is less than the bit flip rate.
+Typically, we implement tests with random outcomes by working out what the probability of a hardware bit error is, and then run the test enough times that the failure rate is less than the hardware error rate.
 (Hmm, we should standardise this, see #20617.)
 …
 The new code is expected to choose A or B, uniformly at random.
 To keep the failure rate below `2^2048` (which is what crypto_rand_check_failure_mode_zero uses), run the test 2048 times. If it comes up A each time, it is almost certain that the code is broken.
+Edit: clarify hardware errors