Logical Consistency testing 101
I used to do validation of this sort of data all the time, there is a boatload of tools and the methods are well published (though it seems not well known). I've seen a great slide illustrating spatial data rejected by a validation frameworks, pity I can't find a version on the web but you can tell *what* sorts of errors are the most popular (this particular one is in the top 5). For instance, getting your coordinates out by a decimal place gets you a miniature dataset flowing in the sea off the coast of west Africa near the Greenwich meridian (hint: if it's continental data, it's probably wrong). Once you have a feel for how it can go wrong, you can look out for it. (Cities should be on a land mass unless the name is "Atlantis")
Same for attribute data. You have pH values? Domain better be between 0 and 14. Temperature? Minimum should be less than maximum. You get the idea. If you use data from external sources, you have to be paranoid about QA.