> By increasing the shard integrity check rate, we potentially moved failures that were going to be found in the future into Q3. While discovering potential pro

> By increasing the shard integrity check rate, we potentially moved failures that were going to be found in the future into Q3. While discovering potential problems earlier is a good thing, it is possible that the hard drive failures recorded in Q3 could then be artificially high as future failures were dragged forward into the quarter. Given that our Annualized Failure Rate calculation is based on Drive Days and Drive Failures, potentially moving up some number of failures into Q3 could cause an artificial spike in the Q3 Annualized Failure Rates. This is what we will be monitoring over the coming quarters. Wouldn't survival analysis on interval-censored data handle this problem automatically? All of your observations of failure presumably are actually interval data, where all you know is that the drive failed sometime in between the last good check and the first bad check. Then it doesn't matter if some time periods have large intervals and others have small intervals, that just affects the precision of estimates.