In this case, every one of the results contained the word "error" which we key on. We'll still match on the result if just one of them doesn't contain an error.
In this workunit there's something else. After the error there's a message that LLR is restarting with a new FFT - in other words starting from scratch. The validator has now been updated so if the last full restart comes after the last error, then it's no longer considered an error. A further change means the validator will email me when four results on the same workunit all have errors. That way we'll never get to the point of a task being returned 20 times.
As Mike said, thanks for bring this to our attention.