Thank you very much for clarification of "tokenization"!
When searching for "Running is a sport", the real query is expanded to "run, run_is, is, is_a, a, a_sport, sport".
I did not find in the solr6 tokenization doc, that "is_a" or "a_sport" has also to be seen as a token. I expected that only different words are tokens, but not all two word combinations behind each other. (Just to be sure: The underscore of your example does mean a single space, doesn't it?)
So you are getting all the results including that tokens.
Does this mean, that every token you mentioned has to appear in every result document? But the order of the found tokens is not necessary? Therefore also documents are found with the following content: 'Is sport a running game'. No documents are found with this content: "Is this game a sport". Is this correct?
BTW If this is true, I don't understand why this search is called "phrase" search. Normally a phrase search implicits a certain order. It's more like a "set search"...
However, when using ="Running is a sport", the query returns the fields that include exactly that terms in the order specified "Running, is, a, sport".
I am glad that I interpreted this syntax correctly. Is it possible to use it as a JSON query without problems? I could not integrate the equal sign immediately into the following syntax:
"query":"cm:content:('*Running is a sport*')"
IMO the equal sign does not harmonize with cm:content. But perhaps I should omit cm:content and replace it with TEXT?