Indeed, I was often frustrated when a "full text" has turned out to be only a title page or an excerpt. However, RG's algorithms cannot distinguish between a real full text and an excerpt. It is the responsibility of the author to distinguish between both forms of texts. I never upload excerpts as full texts, but I use "Linked data" - see, e.g., https://www.researchgate.net/publication/320130779.
sometimes it says download, when it is not rh final version of the article or work. lets say it might be the second to last version. Or maybe when there is just images. I think in these cases, the almost final draft is better
In such cases, you cannot even request for full text (though many authors do not oblige). This is a real problem. The best practice seems to be to maintain a private full text unless RG software is able to recognize the difference between cover/first page and full text.