On June 1, 2026, the team behind the bioRxiv and medRxiv preprint servers announced openRxiv Labs, which they call “a structured program for testing new and ambitious approaches to research communication.” They plan three experiments in 2026. The first, built with a company called Curvenote, tests “a new interactive preprint reading experience.”
That sounds modest, and it is. Nobody is abolishing the PDF. But I think the announcement points at something odd about how we share science. Most research today is born digital, analyzed with code, and read on screens. Then we flatten it into a document laid out like a printed page and treat that file as the finished product.
My view is simple: the PDF is a good archive copy and a poor main event. The paper of the future should probably be a living, linked record, with a stable version sitting underneath it. Getting there without losing what the PDF does well is hard, and openRxiv is right to test it.
What openRxiv is actually testing
This is easy to oversell, so it helps to be precise. The launch post says each project will have “predefined hypotheses, goals, success metrics, and defined durations,” and that results will be published “including when things don’t go as planned.” The stated goal is to experiment “while retaining the stability of our core platforms.”
The first experiment went live on June 11. According to the Curvenote Reader launch post, it lets readers of bioRxiv preprints “explore references, terminology, expanded figures, and related works while staying in the context of the original preprint.” The hypothesis being tested is that preprints are most valuable when readers can easily find the data, code, protocols and cited works behind the story.
Two details stood out to me. First, it runs on the existing archive: the Curvenote team converted the stored article files into an open, web-friendly format “without requiring any additional work on the part of authors” (launch post). Second, progress will be assessed with measures like time on page, clicks and feedback from authors and readers. The team has since shared early numbers: in a July 2026 update, 29% of users had engaged with at least one connected feature, more than the 10 to 15% they expected. That is encouraging, but it came from a relatively small group who found the Reader through the blog or social media, and it shows whether people use the features, not whether anyone understood the science better.
The PDF is a print habit
The PDF made sense for a world of printed journals. It fixes the layout so every page looks the same everywhere. What it does badly is everything the web does well: linking out, zooming into a figure, letting you check a number against the data behind it.
eLife’s editors made this point bluntly in 2022. Science publishing, they wrote, “replaced printed articles with pdfs, but nothing fundamental has changed” [1]. I wrote about eLife’s review model earlier in this series, so I will not repeat it here. What matters here is the shape of their output. Instead of accept or reject, eLife publishes the reviews plus a short eLife assessment of how significant the findings are and how strong the evidence is, “written in language accessible to a non-expert reader” [1].
That assessment is versioned. If authors choose to revise and the paper is sent back for review, “the assessment will be rewritten and a new version of the Reviewed Preprint will be posted” [1]. The paper stops being a frozen object and becomes a record of how the work and its evaluation changed.
Imagine reading a study about a new blood test and seeing a figure that looks too good to be true. In a PDF, you squint at the image and hunt through a supplement for a file name. In a linked version, you click the figure, see the data behind it, read a short plain-language verdict from reviewers, and notice that version 2 fixed an error that version 1 had.
None of that needs exotic technology.
What a better research output could hold
If I had to list what a modern paper should carry, it would be four things. Linked data and code, so claims can be checked. Figures you can explore, so you can zoom, filter or hover rather than trust a single static snapshot. A short explanation written for people outside the field. And a visible version history, with assessments attached to each version.
This is close to my own work. I build web tools for reviewing medical images and automated pipelines for imaging and clinical data, and in that kind of work a static screenshot of an image is rarely enough to judge it; people want to scroll, adjust and compare.
Linking code is not the same as making it work, though, as I wrote earlier in this series about shared research code that fails to run. An interactive figure built on code nobody can rerun is only a prettier PDF.
The honest case for the humble PDF
The strongest argument against all this is permanence. A PDF is one self-contained file that is easy to archive, share and cite, and it will still open in twenty years. A web-based paper depends on servers, scripts and links that someone has to keep alive.
That risk is well documented. Klein and colleagues checked over one million web references in more than 3.5 million science, technology and medicine (STM) articles from arXiv, Elsevier and PubMed Central, published from 1997 to 2012. For recent publication years, they estimated about “one out of five STM articles suffering from reference rot,” meaning the web pages they cited could no longer be revisited as they were [2]. Among articles that cited web resources at all, the share rose to seven out of ten [2]. If plain links decay that fast, interactive figures and live notebooks could decay faster.
Even eLife, the most radical example here, kept a fixed endpoint. Authors can turn their latest version into a final Version of Record, because many databases, such as PubMed, “will only index the final version of a manuscript” [1]. Libraries and indexes are built around stable objects, and they are not wrong to be.
So I concede a lot. The PDF is a good answer to the question of how to preserve a result. My complaint is that we also treat it as the answer to how to communicate one.
Build the living layer, keep the stable core
The sensible path is the one openRxiv describes: keep the archive stable and build richer layers on top of it. The stable version gets cited and preserved. The living layer carries the links, the explorable figures, the plain-language summary and the review history. If the layer breaks, the science survives.
I think the PDF should be the archive copy of a study, not the whole study. Keep a stable version for the record, and put the data, code, explorable figures and updated assessments around it where readers can actually use them.
When a study matters to you, look past the PDF: check for linked data and code, a newer version, or a public review. For researchers, it means treating the data, code and figure files as part of the paper, archived with the same care as the text. And for experiments like openRxiv Labs, I hope the next results they publish measure understanding as well as clicks, because that is the test that matters.
References
[1] M. B. Eisen, A. Akhmanova, T. E. Behrens, J. Diedrichsen, D. M. Harper, M. D. Iordanova, et al., “Peer review without gatekeeping,” eLife, vol. 11, Art. no. e83889, Oct. 2022, doi: 10.7554/eLife.83889.
[2] M. Klein, H. Van de Sompel, R. Sanderson, H. Shankar, L. Balakireva, K. Zhou, et al., “Scholarly context not found: One in five articles suffers from reference rot,” PLoS ONE, vol. 9, no. 12, Art. no. e115253, Dec. 2014, doi: 10.1371/journal.pone.0115253.
Related Articles
