Right now we document how to recreate all of the built in analyzers but we don't test those docs and they aren't 100% consistent. The docs for the language analyzers is all in CONSOLE form but, for example, the keyword analyzer just says that it is made of the keyword tokenizer. While true, it isn't consistent with the language analyzers and isn't tested. I mean, the keyword analyzer is trivial the tests aren't going to find anything, but consistency is nice.