... well, we do the bounds check at index creation but that allows us to store indicies and use them later without the bounds check overhead.

The first few versions of this had embarassing issues and I'm indebt to Kornel Lesiński for pointing them out.