The backlink sacrifice

We got used to thinking about links as unidirectional.

The Web is a system where a page can declare a relation to other pages so that you can jump from one to another, but not necessarily both ways. Although we can address this issue by indexing pages and creating a graph that facilitates back-and-forth navigation, it is not a feature that people currently expect.

Before the Web, links were commonly perceived as bidirectional, where two entities were linked, and additional information was associated with that relationship.

Bidirectionality and the Web

When Tim Berners Lee designed the Web, he made a design choice: anyone could write a link to any resource, and that link was allowed to break. The 404 error was born. By accepting that links could point to nothing, he removed the need for coordination between the linking party and the linked resource. This is what enabled permissionless publishing. If every link had to be negotiated and maintained by both sides, the Web as we know it could not have emerged like it did.

What about controlled environments?

Many contemporary hypermedia systems brought backlinks back. In tools like Obsidian, where you produce and maintain the documents yourself, backlinks are repaired as content evolves. There is no 404 within the vaults, authorship and linkage live under the same roof. The same can be done for agent hypermedia memory, but it's not common. If backlinks can be maintained, retrieval by navigation has less friction, you can see everything that references a particular document. On the other hand, agents can retrieve context directly.