reading [y] as a null character

[Koen G]

Hmm yes, it might work like that, it's worth a shot. I should have some time tomorrow to play around with this.

It's not always easy to predict though. For example, why is merging [ch] recommended for entropy optimization but merging [dy] isn't? Nor is [ed], for example. That's why I was planning to test all characters as nulls. It would be interesting if one glyph or a class of glyphs stands out.