Hypertext before the web: what Xanadu wanted

1965Nelson, who named it

The word hypertext was coined for a paper describing writing that branches, and the project attached to it ran for decades without shipping anything most people could use.

What it wanted is worth listing precisely, because each item is a genuine problem that the system which won did not solve.

Links that both ends know about

A link would be a first-class object rather than a mark inside one document, so a document could be asked what points at it.

The web cannot answer that question. Finding who links to a page requires crawling most of the web, which is why the ability to answer it is one of the assets that made search companies what they are, and why an ordinary author cannot have it.

Links that do not break

Addresses would refer to permanent, versioned storage, so a reference to a passage would continue to resolve to that passage, including after the document around it changed.

The web made the opposite choice and links rot continuously. Every citation to a page is a bet that somebody keeps paying for it, and large-scale studies of legal and academic citations find substantial proportions already unreachable.

Quotation by reference

Text quoted from elsewhere would be included by reference rather than copied, so a reader could see where it came from, follow it back, and the original author would remain attached to it.

Copying, which is what everyone does instead, severs that link immediately. A quotation on the web is an assertion that somebody wrote something, verifiable only if the source still exists and the quoter was honest.

Payment attached to the fragment

The design included payment flowing to authors for the portions of their work actually read through such references.

That problem was not solved by the web either, and the arrangements invented since to work around it, advertising and subscription and platform revenue sharing, are widely disliked by everybody involved.

1990sWhy the weaker design won

The design that spread has one-way links that anybody may create without permission, no guarantee of persistence, and no coordination between the two ends at all.

That is the whole reason. Publishing required nobody's agreement, linking required nobody's cooperation, and a document could be put up by one person in an afternoon. The stronger design required a coordinated infrastructure, and the same asymmetry decided the connection between programs and the shape of the file system elsewhere in this document.

2000s onwardsWhat has been rebuilt since, badly

Archives that copy pages so that citations resolve. Identifier schemes for academic work that add a layer of indirection over addresses. Permanent links maintained by convention. Content-addressed storage where a reference is a hash of the thing referred to. Citation tooling that records what a page said at a moment.

Each is a partial reconstruction of one guarantee, bolted onto a system designed without it, and none of them is universal because universality is precisely what the original design required up front.

1990sThe one guarantee the web did keep

Worth crediting, because this entry is otherwise a list of omissions. Addresses are unowned: anybody may mint one under a name they control, and no registry decides what may be published or linked.

That is the property that produced everything else, and the stronger designs generally gave it up, because guaranteeing persistence and back-links means somebody must be responsible for the guarantee, and responsibility implies control.

present dayWhat the missing back-link cost

The ability to ask what points at a document turned out to be the most commercially valuable fact on the network, and because no document holds it, it can only be assembled by whoever crawls everything.

A design decision that looked like a simplification therefore determined the industrial structure of the medium: a small number of organisations hold the map, because the map was not built into the territory.

1945The oldest version of the idea

The essay usually credited with starting all of this describes a desk that stores documents and, crucially, the trails a reader builds between them, which could be shared with others.

The trail is the part nobody built. What exists is the ability to link, not the ability to publish a path through other people's work as an object in its own right, and reading lists and thread compilations are the impoverished modern substitute.

The lesson that is not about hypertext

The project was right about almost everything and delivered almost nothing for decades, and both halves are the point. A design can be strictly better and lose to something available now, and the loss is not a failure of taste on anybody's part.

Which is worth carrying into any argument that begins with the observation that the thing everybody uses is badly designed. That is frequently true and it is rarely the relevant fact.

What we cannot verify

The published papers and specifications can be read, and implementations of the later project exist. Accounts of why it took so long are contested and often unkind, and we take no position. Figures for how much of the web has decayed vary enormously by corpus and by the definition of decay, and any single percentage should be treated with suspicion.