The Library of Congress resumed choosing from above, late — the early web became readable another way: its authors, linking for their own reasons, left a ranking behind without ever meaning to make one
R814’s keepers chose at last, and chose from the top: themes and events picked in advance by the institution that held the pile. The web of the 1990s faced the same drowning in a different medium. Pages multiplied faster than any directory could list them, and the directory that tried longest — Yahoo!’s, begun as a human-compiled catalogue whose editors evaluated and categorised each admitted site — could not scale with the growth it was cataloguing; the backlog lengthened as the web did. Keyword engines crawled everything, but a page that repeated a word oftenest was not the page a reader wanted, and the words on a page were the one thing its author could inflate at will.
The ranking that worked was already lying in the pile, left there by people who were not ranking anything. Every author who put a link in a page did it to send their own reader somewhere, to cite a source, to point at the thing they were talking about — in the act of using the web, not in the act of organising it. In their 1998 paper on the anatomy of a large-scale search engine, Sergey Brin and Larry Page treated those links the way scholarship had long treated citations: a link is a vote, but votes weigh differently — a link from a much-linked page counts for more than a link from an obscure one, and a page that links out to many places spends its vote thinly across all of them. Run that rule over the whole graph, again and again, and a single score settles out for every page: its PageRank, in their phrase an objective measure of citation importance that corresponded well with people’s subjective ideas of importance. Their prototype had applied it to about 24 million pages.
The idea was older than the web, and that matters. Eugene Garfield proposed citation indexing for science in 1955, and the Science Citation Index that followed in the early 1960s let a reader follow citation trails through a literature too large to shelve by subject — and, as a by-product, discover what the field itself regarded as important, because a highly cited paper is a marker other scientists left in passing, while arguing their own arguments. PageRank ported exactly that: the ranking is nobody’s editorial decision, and no one link was made in order to produce it. It is the users’ collective judgement, recovered afterwards from the traces of their ordinary work.
Hold the contrast exactly. The directory’s choosing was deliberate, prior, and scarce — an editor read a site before anyone could find it through them. PageRank’s choosing is incidental, posterior, and total — nobody read for the ranking at all, and every link counts, including the careless ones. That is also its known weakness, visible from the start: a ranking left as a by-product can be farmed by making the traces without the judgement — links manufactured to look like votes — and much of the engine’s later life has been spent telling real passing from performed passing. A pile ranked by its users is readable in a way no keeper’s list matched, and corruptible in a way no keeper’s list was.
Six pages, no editor. A schematic toy, stated plainly. Six pages, A–F, with fixed links between them — A→B, A→C, B→C, C→A, D→C, E→C, E→F, F→C — made, in the fiction of the box, for the authors’ own reasons. Nobody has ranked anything. Each press runs one round of the vote-counting: every page divides its current score equally among the pages it links to, damping 0.85, the remainder spread evenly. Watch who rises: not the page with the most words, but the one the passing traffic kept pointing at.
The Library’s archive stayed unread behind a keeper who would not choose; the web’s pile got a ranking because its makers could not help choosing, one link at a time, while doing something else entirely. Choosing does not have to be an office or an occasion. Sometimes it is just what a crowd leaves in the dust of having passed through — and the art is in noticing that the dust, read back carefully, is a map.