Or why we start demanding the 26 eyes principle …

The assumed quality of human manual identity verification is 99%, giving an error rate of 1%. That was the main reason behind the 4 eyes principle rule at CAcert, trying to have at least 2 assurers verify the identity of a person. Statistically, this should reduce the error rate to 1:10,000 . In practice, we had a lot of cases already, where the second assurer noticed small problems, that the first assurer oversaw, which resulted in correcting the error. (Most errors are numerical errors with the birthday, or smaller mistakes in the name). So the system seemed to work fine that way. Until an assurer noticed a couple of days ago, that there are 2 small errors in the name in an account, and that there were already 12 other assurers that assured the person, without noticing it.
Statistically, the possibility for 12 assurers not noticing it should be 1:1,000,000,000,000,000,000,000,000 (according to the statistical model)
Thanks to the 13th Assurer for noticing it!
Your CAcert Quality Assurance Department