OCR word low confidence but all characters max high confidence?

I am noticing that sometimes the OCR engine will return a 1 for the high bit of a character's confidence property, which the documentation states as being "uncertain" yet each character in the word are all 0 for the remaining bits, which the documentation states as being the highest of confidence.  If each component character in the word are full confidence what is causing the word confidence to be low?

The documentation states that the converse can happen: "In some cases a word may have some or all characters that are individually suspicious but the characters are not be marked suspicious in the word bit."  This makes sense but there is no mention of what would cause the behavior I've described above.