A Dead Link Is Not Lost Code: Separating Repository Reachability from Deposit Availability in Zenodo Software Citations
This study analyzes 3,837 Zenodo software citations to demonstrate that while most archived files remain accessible even when their source code repositories become unreachable, a significant portion of broken citations stem from dead repository pointers rather than lost software, highlighting a critical distinction between repository availability and deposit integrity.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the modern world of science, a piece of software is no longer just a tool; it is a research product in its own right, deserving the same credit as a journal article. To ensure this credit is fair and lasting, scientists now attach a special digital tag to their code, a unique identifier that acts like a permanent address. This tag points to a specific version of the software, frozen in time, so that anyone reading a paper years later can find exactly the same code the author used. The system relies on a promise: that this address will always lead the reader to the software. But for this promise to hold, two things must happen. First, the address itself must work, leading to a page where the software is stored. Second, the software itself must still be there, waiting to be downloaded. If the address leads to a page that says "file not found," or if the page loads but the code has vanished, the scientific record has a gap.
Researchers have long known that web links in science often break, a problem known as "link rot." However, software citations are different because they are built on a two-layer system. The permanent tag points to a digital archive, which in turn points to the original code repository, often hosted on a platform like GitHub. The archive is managed by an organization dedicated to keeping things safe, while the original code lives on a platform where the owner can rename, privatize, or delete the project at any time. This creates a unique situation where the permanent tag might still work perfectly, even if the original project has disappeared. A researcher set out to measure exactly how often this happens, and more importantly, to see if the software itself is still recoverable when the original path is blocked.
The study focused on thousands of software records from a major digital archive called Zenodo, which is widely used by scientists to preserve their code. The researcher did not just check if the original project pages were still online; they checked three distinct levels of the citation chain. First, they checked if the original project repository still existed. Second, they checked if the specific version of the code mentioned in the citation was still accessible. Finally, for the cases where the original path was broken, they checked if the digital archive still held the actual files. This approach allowed them to separate the failure of a link from the loss of the software itself.
The results revealed a clear pattern of decay over time. As software records got older, the likelihood of their links breaking increased. About 4.7 percent of the original project repositories were no longer reachable. However, the problem was even more common when looking at the specific versions of the code that scientists actually cited. In this stricter check, the failure rate rose to 5.45 percent. This difference is significant because it means that in roughly one out of every six broken citations, the original project is still alive and well, but the specific version of the code the researcher cited has been deleted or moved. This is a crucial distinction: a simple check of the project's main page would miss this failure entirely, making the software appear available when it is not.
Despite these broken links, the digital archive proved to be a remarkably effective safety net. When the researcher followed the broken links to the archive, they found that in 94.4 percent of cases, the archive was still serving the original files. This means that even when the path to the original project was lost, the software itself was usually still there, preserved in the archive. The archive is doing exactly what it was designed to do: keeping the code safe even when the original home is gone. The only time the software was truly lost was in a very small fraction of cases, less than one percent of all citations, where the archive record existed but the files were missing.
The study also highlighted a subtle but important risk. About 6.4 percent of the citations that appeared to work were actually relying on a courtesy redirect. This happens when a project changes its name, and the hosting platform automatically sends the old name to the new one. While this works today, it is not a permanent guarantee; if someone else claims the old name in the future, the link could silently break or point to unrelated code. This suggests that while the archive is holding the software, the way we point to it needs to be more careful. The researcher concluded that scientists should cite the permanent archive identifier rather than the original project link, and that tools used to check citations should look at the archive first before declaring a reference dead. The infrastructure built to protect research software is working, but it requires us to look deeper than the surface link to find the code that remains.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.