http://mathoverflow.net/questions/21014/isotropic-deformation-retract-of-weinstein-manifolds http://mathoverflow.net/questions/21236/weight-filtration-for-smooth-analytic-manifolds

I have one question: will something like '1001.1234' be picked up without the 'arXiv:' prefix?

Yes. Anything matching the following regex will be picked up: [^\w.]\d\d[01]\d.\d{4}(?:[v\W]|$) . Based on five months of data, this doesn't seem to have picked up any false positives. If we do get false positives, we'll refine our search. Feel free to use this thread to imagine situations where somebody might reasonably match this regex without meaning to refer to an arXiv article.

And yes, at least for users with javascript enabled (almost everyone), we'll see in Anton's google analytics statistics how many people arrive via arxiv trackbacks. Remind us again in a month if we haven't reported on this --- I'm really interested to find out how effective this will be.

We've submitted all the "historical" trackbacks from earlier posts, and these have been incorporated into the appropriate pages. There were ~1000 of these. We have a cron job running on the same server that hosts meta, which periodically looks for new arxiv links in recently edited questions, and submits those. (That is, we're doing this entirely without StackExchange support.)

We locate references to arxiv papers very aggressively -- it's not even necessary to actually have a link. We've implemented a bunch of tricks to locate arxiv identifiers, and checked the output carefully for the historical links, but won't be reviewing things routinely in future. If you happen to see an incorrect trackback, please let us know, as we'd like to play nice for the arxiv folks (who've been extremely helpful and accommodating!). If you suspect that something should have generated a trackback, make sure it wasn't the result of a recent edit (there's at most a 24 hour delay before you'll see things on the arxiv, it seems) and then let us know here or elsewhere.

Even though we try to find other arxiv references, I'd strongly encourage everyone to use proper links to the arxiv, with [link text](link url) style links. This makes it easier for readers than just bare arxiv identifiers, and means that the links will be detected by the less error prone parts of our program!

Go out there and discuss some arxiv papers! You can also use http://arxiv.org/tb/recent to see recent trackbacks on the arxiv -- I expect that mathoverflow will dominate this listing for now.

(Interestingly, in the database dump I already find ~650 links to the arxiv.)

I think the technically easiest solution is to run a cron job that watches for links to the arxiv from mathoverflow, then constructs the appropriate request to the arxiv. If someone wants to get thinking about a little command-line program that finds recent links from mathoverflow to the arxiv, please go ahead and report back here. For now, just something that outputs pairs (mathoverflow URL, arxiv URL) would be great.