That's precisely the point. As I said in the OP, the scalars in question contain arbitrary binary data, not text in any form of encoding, hence my use of use bytes, which works admirably well for index, rindex and anything else that doesn't use the regex engine.

However, the search criteria I need to use lends itself nicely to using a regex, and under most circumstances of data and search values, works perfectly. But every now and again I was getting mismatches, and spent ages investigating both how the data and the search terms were constructed, before the realisation dawned.