In 1960 — thirty-one years before Tim Berners-Lee published the first web page — a Harvard sociologist named Ted Nelson coined the word "hypertext" and began designing a system called Xanadu. It was meant to be the ultimate document architecture: a universal, versioned, interconnected library of all human writing. It was not a prototype or a thought experiment. Nelson spent decades building it, secured funding, hired engineers, and published detailed specifications. And yet the World Wide Web, a system Nelson has called "a universal platform for thin gruel," shipped first and won.

What Xanadu proposed was radical and, in hindsight, prescient. Its core features included:

• Bidirectional links. If document A linked to document B, document B automatically knew about the link. No broken links, ever. The web gave us one-way links and 404 errors.
• Transclusion. Instead of copying text, you could include a live reference to the original. Changes to the source propagated everywhere. The web gave us copy-paste and content rot.
• Version history on everything. Every edit to every document was preserved permanently. The web required third-party tools like the Wayback Machine to approximate this, badly.
• Micropayments for quotation. If you transcluded someone's paragraph, they got paid automatically. Nelson saw the content monetization crisis of the 2010s coming from 1960.
• Parallel document comparison. Xanadu could render two documents side by side with visible connections between shared passages — what Nelson called "intercomparison." Plagiarism detection and source tracing, built into the infrastructure.

So why did it fail? The short answer is scope. Nelson wanted to solve every problem of digital text simultaneously, and 1960s–1980s hardware simply could not support a global, persistent, versioned, transclusive document store. The project cycled through multiple implementations. Autodesk funded a team from 1988 to 1992, spending roughly $5 million before pulling out. The engineers, led by Roger Gregory and Mark Miller, built working prototypes in Smalltalk and later C++, but the complete vision required distributed storage, content-addressable retrieval, and micropayment infrastructure that didn't exist. When Berners-Lee launched the web in 1991 with deliberately simple one-way links and no versioning, it shipped fast precisely because it avoided Xanadu's hard problems.

Nelson was bitter about this — publicly, repeatedly, and not without justification. He called HTML "an imitation of paper" and argued that the web had trapped humanity in a flat, fragile, link-rotting document model. He continued working. A partial implementation called XanaduSpace launched in 2007. OpenXanadu, a browser-based demo, appeared in 2014. Neither gained traction.

Here is the case for Xanadu in 2026: every hard technical prerequisite now exists. Content-addressable storage is production-ready (IPFS, Git's object model). CRDTs enable real-time collaborative versioning without central servers. Blockchain and payment-channel networks like Lightning provide micropayment rails. Git already proved that permanent version history is not only feasible but beloved by developers. The missing piece was never the idea — it was the infrastructure. Nelson needed distributed hash tables, Merkle DAGs, and sub-cent transaction fees. We have all three.

Bidirectional linking is arguably the most consequential missing feature of the modern web. The Webmention W3C standard attempts it, but adoption is negligible. A protocol-level implementation of Xanadu's link model would eliminate broken links, enable automatic citation tracking, and make plagiarism structurally visible. Transclusion would replace the grotesque practice of screenshotting tweets and pasting them into articles. Micropayments would offer an alternative to the surveillance-advertising model that funds most online publishing.

Ted Nelson turned 89 in 2026. He has spent sixty-six years on this idea. The tragedy is not that Xanadu was wrong — it is that the web shipped a rough draft and we never went back to finish the job.