A good article will restate the results again in plain language and state whether they are consistent with the expected findings.
Do the authors overstate their findings? That is, if the results were inconsistent or small effect sizes, are the authors making too much of them? Do the authors downplay or ignore non-significant results?
The discussion should address how the findings fit within existing literature.
There should be a limitations section that should either justify why the limitations weren't so bad, or suggest directly how they can be corrected in future research.
Can you think of any confounds or limitations to the article that the authors have not discussed?
The broader theoretical and practical implications of the findings should be discussed.
No new statistics or findings should be presented in a discussion.
The article, overall, should have an hourglass shape; beginning broadly and getting narrower, with the discussion starting narrow and getting broader.