So, the problem with invalid chars existing in the test data still persists. I've tried a number of ways to get past this and it looks like it's on the test side and not mine. To give you an idea, at one point I was able to pass one of the real world tests by appending 3 nulls to the end of the string, which matches what I found in the infix passed to me. The next test failed with nulls mixed in at other positions.

This is really frustrating. I've spent more time debugging problems with the test code than with my own.