Actually, computer intelligence could eventually evaluate the relevance of attached records. And events/relationships without supporting evidence could be assigned a rating of "undocumented."

As to spellings, whether misspellings or variants, since even official records contain those, I don't see how they could be used to evaluate the worth of a given tree.

In fact, even trees containing egregious errors often contain a grain of truth. Think of those family stories we have all tried to prove at one time or another, and found that the facts were muddled or distorted over the years, even though the story had a factual origin.

Eventually, a computer generated "universal" tree could reduce or almost eliminate what you call junk trees, since it would represent a weighted summary of all the research that has been done over the years.