Failure detector mistake rate
A failure detector's mistake rate is the number of false-suspicion episodes per unit of eligible observation time.
Hayashibara et al. distinguish average mistake rate from detection time and plot mistake rate in inverse seconds. It is not the dimensionless heartbeat-tail probability encoded by phi, nor the posterior probability that a process has failed. [1]
An operational report must define an episode, the observer and the independently established alive state, as well as the observation-time denominator. Ten nonoverlapping one-second episodes in 1,000 eligible seconds and ten fifty-second episodes have the same rate of 0.01 episodes per second, but occupy 1% and 50% of that time. These are Hesela's invented accounting examples, not measured traces. [2]
Polling a suspected state repeatedly does not create new episode onsets. Report duration and detection delay separately, and document missing observations or uncertain crash-and-restart labels before interpreting the rate. [2]