I'm mainly interested in the level of fact-checking of computer science conference papers. Do reviewers normally read the entire paper and exhaustively check the correctness of proofs, algorithms and claims?
I have read papers at top conferences where implementation details are sketchy and conceptual claims are inaccurate.
My guess is that an exhaustive level of fact-checking is often not done particularly in cases where the reviewer might not be familiar with all the concepts in the paper. Is this true? If this is so, then it seems unfair that the merit of a paper is essentially just based on how well the paper is pitched in the first one or two pages.