As this has obtained the required number of community votes, consider this an official PhysicsOverflow policy.
I see that this is a good point. After all, even the ArXiV has lecture notes, etc., which are not really "original" in nature.
When reviewing things like blog articles, lecture notes, etc., there is no "originality" in them. This is not necessarily a problem since the formula for overall score is
\(x= \mathfrak{S} \exp\left( \sqrt[3]{\frac{\mathfrak{\times}}{5}} \right) \)
So, even if the originality is 0, which it usually is, the overall score will just be the accuracy score.
I don't think newspapers should get reviewed, for obvious reasons.
Not even blog articles. It is difficult to make an objective criteria between a real physics blog and a troll one. Some are extremely debatable. Also, many blogs are devoted to reviewing themselves.
Great blogs, such as TRF, are not necessarily devoted only to physics. TRF for example is also about politics, and so on. Now, the bot doesn't know that Barrack Obama is a politician, or that Joseph Polchinski is a physicist.
So, in conclusion, it is safe to import research papers, review papers, conference papers (from their proceedings), and seminars. The sources can be ArXiV, ViXrA, journal databases, proceedings, wherever. Direct uploads aren't supported yet.