You're right on the "@Q" part. Dumb mistake. With that corrected, the mapping makes a lot more sense.

As for the right interpretation of the spec, I guess you'll have to ask the OP. I believe, (with the "deliberate" error:) and constant 9th bit excepted), I coded a direct implementation of the spec. If the spec is wrong, or my interpretation of it is wrong, then the OP will have to correct it.

The main point was to answer the OPs question of the best way to perform the mapping efficiently, and that is tr/// with the appropriate translation strings. On a gig of data, doing 10 x 100MB tr///s, (or even 1 x 1GB if he has the memory), is going to beat performing half a dozens boolean operations on every byte, hands down.