The Links That Would Never Break

The world it was born into

In 1960 a computer was a room-sized machine owned by a university, a corporation or a government. People submitted programs on punched cards and waited hours for results. There was no network connecting machines across institutions, and the idea of reading on a screen was close to science fiction. Writing meant paper, and paper meant sequence: one page after another, references at the back, quotations copied by hand from one document into the next.

Ted Nelson, a philosophy graduate studying sociology at Harvard, looked at this world and saw a problem older than computers. Ideas are not sequential. They connect in many directions. A thoughtful text quotes, comments on, contradicts and builds on other texts, yet paper forces those connections into footnotes and bibliographies that cut the reader off from the source. Vannevar Bush had imagined in 1945 a desk-sized machine, the memex, that would let a reader build associative trails through documents. Nelson wanted to make that real, at the scale of all human writing.

The bet

By the mid-1960s Nelson had a word for what he wanted, hypertext, which he presented at an ACM conference in 1965. Over the next two decades the design grew into Xanadu: a single, global “docuverse” in which every document could be linked to every other, permanently.

Its features read today like a list of everything the web lacks. Links went both ways, so every document knew what pointed to it, and links could not break because the system managed them. Quotation worked by transclusion: rather than copying a passage, a document included it by reference, so the quotation always stayed connected to its original context and author. Every version of every document was kept, and readers could compare versions side by side. And every time someone read or quoted a piece of writing, a small payment flowed automatically to its author, a scheme Nelson called transcopyright.

The bet was that a literature built this way would be more honest, more navigable and fairer to authors than anything paper or a simple network could offer. It was a bet on a complete system: the features depended on one another, and all depended on a coordinated infrastructure that kept track of every document, link and version.

Why it made sense

Nelson’s diagnosis was correct, and much of it has been confirmed by what happened instead. The web’s one-way links break constantly; studies of academic citations and news archives regularly find large fractions of links dead within a few years. Quotations on the web lose their context, which makes misquotation and manipulation easy. Versions disappear, so that a page can be changed silently. And the problem of paying creators for what they publish online has never been solved well; advertising became the default, with consequences discussed throughout this site.

Within its own logic, Xanadu’s design was coherent. If you want links that never break, someone has to manage them. If you want quotations to stay connected, you need transclusion. If you want authors paid, you need to know who read what. These are not arbitrary features. They follow from taking the problem seriously.

The project also attracted unusually capable people over the years, and serious backing: in 1988 Autodesk, then one of the most successful software companies, invested in turning Xanadu into a product.

How it lost

Xanadu never shipped as a widely used system. A combination of causes, technical and organisational, kept it in development for decades. The coordinated infrastructure it required was hard to build on the computers of the 1970s and 1980s, and harder still to deploy across institutions that had no shared network. Each partial version could be judged against the full design and found incomplete. Autodesk withdrew around 1992.

Meanwhile, in 1989, Tim Berners-Lee at CERN wrote a proposal for a much more modest system. His proposal set out to solve a local problem, information getting lost in a large organisation with high turnover, and made design choices that Xanadu’s would have rejected. Links pointed one way. Documents lived on independent servers. Anyone could publish without asking permission from a central authority, and the proposal stated plainly that the system must allow existing systems to be linked together without requiring central control. Links could break; the web simply returned an error.

The web shipped in 1990–91, was released royalty-free by CERN in 1993, and spread with the Mosaic browser. Within a few years it connected millions of people. Every one of its weaknesses was a Xanadu strength, and none of them mattered as much as the fact that it existed, worked and was easy to join.

Explaining both the same way

The usual story says the web won because it was simple and Xanadu lost because it was too ambitious. That is true but incomplete, and it flatters the winner. A symmetrical account asks what each system optimised for, and who decided what mattered.

The web optimised for adoption by people who had computers on networks in the early 1990s: physicists, programmers, universities and soon businesses. For them, the ability to publish today without coordination outweighed the guarantee that links would work forever. Xanadu optimised for the integrity of a literature over decades: authors, scholars and readers who cared about context and attribution. Those people mattered less to early adoption, and the problems they cared about became visible only after the web was everywhere.

So the web did not win because broken links are better than working ones. It won because, for the people choosing in 1993, the cost of guarantees was immediate and the cost of their absence lay in the future. That is a perfectly reasonable choice for those people. It is not evidence that Xanadu’s goals were wrong.

What the loss teaches

Xanadu is the purest example on this site of a vision beaten by good enough. Its lesson is not “keep it simple”, which is too easy. It is something more specific: in a network, the cost of joining is a design parameter as important as any feature. Every guarantee a system offers has to be paid for by someone, usually in coordination, and coordination slows adoption. The web made joining almost free and let the costs fall later, on everyone.

There is a second lesson, about time. Much of the last thirty years of web development can be read as the slow return of Xanadu’s ideas: backlinks in wikis and note-taking tools, version history in documents and code, embedding that keeps a post connected to its source, recurring efforts to pay creators directly. The ideas were not wrong. They were early, and bundled together in a form that could not ship.

Sources and limitations

The history of Xanadu is contested, and much of the published record comes either from Nelson himself or from critics. This page relies on Nelson’s own published writings for the design, on the official Xanadu site, and on Berners-Lee’s proposal, quoted from CERN’s copy hosted by the W3C. Gary Wolf’s 1995 Wired article is cited only for the dates of the Autodesk period, since Nelson disputed its characterisation of the project. Statements about the scale of link rot are general; specific studies vary by domain and method. The judgement that Xanadu was “beaten by good enough” is the author’s interpretation, offered with the symmetrical caveats above.