> For the complete documentation index, see [llms.txt](https://concept.mosaic-program.org/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://concept.mosaic-program.org/appendix/appendix-a-anchor-evidence-for-networked-research.md). # Appendix A: Anchor Evidence for Networked Research #### Networked research outcomes on platforms not designed for it. * COVID Moonshot (von Delft et. al, 2021). In March 2020, a researcher posted a viral crystal structure on Twitter. Within weeks, 150+ scientists had self-organized into a global, open drug-discovery collaboration that produced a preclinical antiviral candidate. The collaboration was an unprecedented mode of scientific work, running on social-media coordination and open data sharing without an institutional sponsor. As the researchers observed: “It is a way of working that none of us realized was possible.” * The Polymath Project. Beginning in 2009, mathematicians Timothy Gowers, Terence Tao, and others demonstrated that difficult mathematical problems could be solved through massively collaborative online effort. Gowers and Michael Nielsen published in Nature on the lessons for collaborative science. In 2023, Polymath contributors translated a proof of the Polynomial Freiman-Ruzsa conjecture into Lean, where machine-verification caught a minor error humans had missed. * Science Twitter. In 2022, researcher Andy Matuschak wrote that "Twitter radically improved the quality of \[his] creative and intellectual conversations" — and asked, "What happens if you build a pivot Twitter along this axis?" () Many researchers have reported similar experiences - for a broader review, see [Twitter changed science — what happens now it’s in turmoil?](https://www.nature.com/articles/d41586-022-04506-6) * OpenPath (Huang et. al, 2023) demonstrates the AI-for-modular-science thesis directly: a Stanford team observed that pathologists were already producing microcontributions in everything but name on medical Twitter: image-plus-diagnosis annotations. They scraped roughly 200,000 of them, trained a pathology foundation model, and beat the prior state-of-the-art. * Informal peer review on social media. A longitudinal digital ethnography by Patel & Chan (2026) documented 100+ cases of "informal peer review": substantive evaluation of published research occurring on Bluesky, Twitter/X, LinkedIn, PubPeer, and personal blogs. The study covers a diverse ensemble of credentialed scholars, hobbyists, and methodologists who would often be passed over for formal peer review. Three cases the authors studied in depth show this work producing real institutional consequences: a viral BMJ Nutrition paper on apple cider vinegar was retracted after informal reviewers exposed statistical flaws; a high-profile metascience study in Nature Human Behavior was retracted following critique from informal reviewers like Joe Bak-Coleman; and informal reviewers' fraud allegations against Cassava Sciences triggered FDA reports and university investigations. The authors cite Hilda Bastian ([2014](https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1001772)) observing that “Substantive discussion in journal clubs, in email lists, in social media, and... conferences are not distilled into a concise, permanent, accessible record.” . **EleutherAI.** Adjacent open-source AI research shows the same dynamics at more developed scale. Founded as a Discord server in July 2020 to build open-source alternatives to GPT-3, EleutherAI's volunteer community produced The Pile (an 886 GB training dataset later adopted by Microsoft for Megatron-Turing) and the GPT-Neo/GPT-J/GPT-NeoX models (the largest open-source GPT-3-style models in the world at release). In early 2023, it incorporated as a non-profit research institute backed by Stability AI, Hugging Face, and Canva. The trajectory – open chat-platform community → significant scientific outputs → institutional formalization with mainstream funder backing – is the arc MOSAIC envisions for networked research at large. Even this case emerged largely by accident: through exceptional persistence, ad-hoc compute arrangements, and years of volunteer labor. With intentional infrastructure, this trajectory could become routine rather than rare. See [this Less Wrong post](https://archive.ph/C9PTB) for an in-depth survey. **Scientists migrating to Bluesky/ATProtocol.** Following changes to X, researchers have been migrating to Bluesky. Researchers describe it as "a place of joy" offering more control over content and engagement (Mallapaty, Nov 2024). A population-scale study tracking 276,431 academics on Twitter/X (January 2023 to December 2024) found that "one in five academics adopted Bluesky," roughly 50,000 researchers (arXiv:2505.24801). A separate 2025 analysis found over 2.6 million Bluesky posts referencing 532,302 scholarly articles from January 2023 to July 2025, with a sharp increase in scholarly activity from November 2024 to January 2025 (Zheng et al., arXiv:2507.18840). In the months following October 2024, Bluesky reached 39% of academic mentions for Nature and Science, 48% for PNAS, and exceeded X in some scholarly communities — 65% for JASIST and 56% for Quantitative Science Studies (Arroyo-Machado et al., 2025). Zheng et al. conclude that Bluesky has emerged as a credible platform for science communication. **The ATProto Science ecosystem.** Co-founded by Ronen in 2025, the ecosystem now contains more than 20 independent, interoperating research projects, most built by volunteer or bootstrapped teams. Ronen also co-founded the Cosmik R\&D lab, which is building Semble, a social research curation tool on ATProto. Other projects include annotation tools, research feeds, a science crowdfunding platform (Catalyze), independent research portals (research.space), and a researcher-focused social client (Lea). See for a longer list. The pace shows that researchers and developers are eager to build on shared open protocols once they exist. **Discourse Graphs deployments**. Active deployments across 10+ research labs (see list below), with 100+ weekly active users generating 6,000+ structured discourse nodes (claims, evidence, questions, hypotheses) in approximately one year. From one pilot user: "Even in the limited time I've used Discourse Graphs, I've found huge improvements in my thinking and doing of science." The same user said using discourse graphs is "more aligned with what the actual process \[of doing science] is like" than the current publication system. A key design insight from these deployments: when tools scaffold how research actually works rather than how it's presented after the fact, adoption becomes natural rather than burdensome. Discourse Graph pilot research labs: * Allen Institute for Cell Science * McGill University * NYU * Quantum Biology Institute * Stowers Institute * University of Toronto * University of Washington * Washington University in St. Louis * University of Maryland
--- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://concept.mosaic-program.org/appendix/appendix-a-anchor-evidence-for-networked-research.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.