I did notice that the parser seems to concentrate most of its effort on well-formed queries. That is, I did not see a lot of code detecting bad state transitions in the parsing routine. While I did resubmit a few of the bad queries and received the "Cannot parse fragment:" error I decided to toss a few more bad queries at the API. Many of these queries returned search results instead of generating an error. This might be due to my lack of understanding about the search engine and potential query cleanup optimizations.
If the assumption for the classic API interface is that the client is generating well-formed queries (possibly programmatically) then the concern I raise here is not important.
Though the idea that I may submit a poorly-formed query and get results has me wondering about how to interpret the results.
As a few trivial examples I hacked a few of the sample queries above:
via @DavidLFielding :