Counting ArXiv Delays

Frustrated by my arXiv submission being moved into the "on hold" queue, I did what any reasonable person would do in this situation: an analysis of arXiv submission delay times since 2015.

• • •

Recently, I submitted a preprint to the arXiv, expecting to have it up the following day as previous submissions. But even a few days later, I hadn't gotten mail about the submission, so I checked back to find that the submission is "on hold". This must be an effect of some of the recent changes to the moderation policies (such as requiring literature reviews and position papers in computer science to be accepted in a peer-reviewed venue before indexing on arXiv). But none of the moderation policy changes applied to my paper, so I thought, then the moderation queue is longer than usual because of a general onslaught of fake vibe-researched submissions. I tried finding average times of delay from submission to announcement / indexing. If there is a recent increase in delays, that would explain the situation. But neither the cool arXiv statistics dashboard nor some third-party offers that.

So let's scrape the data ourselves and find the reason for the delay.

We want papers' submission and announcement dates, over time and in different categories. To get papers over time, the search feature lets us find papers from a specified time frame and exposes the submission date, but weirdly enough the announcement date seems not to be exposed anywhere (even though, oddly, search results can be ordered by announcement date). The announcement pages themselves at arXiv.org/list/<category>/new and /recent only go back a few days, so they don't help much. The https://arxiv.org/catchup/<category> page seems to go the furthest, 90 days.

Fortunately the Internet Archive has backed up a decent number of announcement pages, so we can use those every 3 months. We can then query the paper pages themselves to get the submission dates from arXiv itself, and now have paired datasets. We sample from 2015-01-28 to 2026-04-04. For the last 90 days, we can directly use the catch up page. Overall we get 63,847 papers.

To then calculate the delay, we simply subtract the time between submission and announcement, not counting weekends and holidays (using just the arXiv 2026 holidays), to get the delay in number of working days.

Here's a histogram of the delays in working days from submission to announcement.

The distribution seems to be roughly a zeta distribution / a power law; notice how the height of the histogram bars roughly declines linearly on the log-log axes.

Here's the delays over time, aggregated by quarter for the earlier sampling window and weekly for the most recent 90 days.

Not only is there no recent increase in overall delays, but there never has been an increase for a sustained period of time. So an overall increase in delay times is probably not what is happening. Fair enough, if there would be a widespread enough delay I would surely have heard about it from friends.

Let's look at the delays by subcategory. The arXiv is divided as follows. The topmost division is by group (physics, math, computer science, quantitative biology, quantitative finance, statistics, electrical engineering and systems science, as well as economics). The next division down is the "archive", but all groups except physics only have one archive. The subject category is a much finer division, with 149 unique subjects in total (plus 6 subjects that just alias to other ones). The archive and subject is what actually shows up in the shorthand you see as an author. For example, cs.AI means the AI subject category within the computer science archive.

For each subject category, here's the mean (a dot), median (a triangle) and 0.1 and 0.9 quantiles (whiskers), colored by archive. Note that some subjects have very low sample sizes. For the purposes of this, we don't count cross-listings, only the primary listing subject.

Show full chart

And here we found the culprit. The paper I was submitting belongs under cs.CY, Computers and Society, as I now know after reviewing all the categories more carefully. This subject has one of the longest delay times, maybe because there are so many people like me who don't know all the subjects. Some other subjects with similar delay times are cs.OH, math.GM, physics.gen-ph - Other Computer Science, General Mathematics, and General Physics. I suppose for these, many submissions must also not have fit into the original category the author proposed and were reclassified, or conversely, the submissions under this category need to be read and classified into a more specific subject category if appropriate. Another fun and high-delay subject category I didn't know about is physics.pop-ph, popular physics.

So if you're submitting to a new arXiv subject, don't rely on the time to announcement being the same as you're used to. Find a handy somewhat searchable table below: