Why does the evidence keep vanishing?

Picture the citation you saved three years ago. Maybe it sat in a report, a group chat, the footnote of something you wrote and were proud of: the proof, right there, one blue link away. You click it today. Nothing. A blank server error, a domain up for sale, or a company’s new homepage that has never heard of the page you meant. The proof used to be right there. Now the address points at silence, or worse, at something else entirely wearing the old page’s clothes.

This happens more than people assume, and it happens quietly, which is the part that should worry you. A page can die outright: the classic dead link, red text, an apologetic 404. Or a page can do something stranger: keep answering, keep returning a normal looking page at the exact same address, while the words behind it have quietly changed underneath. The second kind should trouble a careful reader most, because a page that still loads gives a careful reader no reason at all to double check it.

Researchers who study this split the failure into two kinds, and the names matter. Link rot is the obvious one: the address serves nothing at all, a flat 404. Reference rot is the sneakier cousin: the link still resolves, the page still loads, and the material actually cited is gone or swapped for something else, all under the same URL. A reader skimming the address bar sees a working link either way. Only the content behind it tells the real story.

Jonathan Zittrain, Kendra Albert, and Lawrence Lessig went looking for this inside the highest court in the United States. They pulled every URL cited in a Supreme Court opinion, 555 links in total, and checked each one by hand. 63.6% still returned a normal 200 OK response, the web’s way of saying everything is fine. But of those working links, only 78.5% actually still held the material the citation pointed to: the rest had been quietly swapped, address intact, page loading, original material gone. Fold the flat 404s back in (24.5% of the total) and the paper’s headline figure lands at roughly half: 49.9% of the citations inside Supreme Court opinions no longer showed what was originally cited. A separate count, inside the Harvard Law Review and two sister journals, found the number climbing past 70%. Two different corpora, two different figures, easy to mix up and worth keeping straight: the difference between half and seventy percent is the difference between a serious problem and an alarming one.

Here is a page of citations, the kind sitting quietly under half the claims you read in a year. Nothing has happened yet. Now move the slider forward and watch what ten years does to evidence nobody thought to protect.

Watch which row survives. Seven links rot on their own timetable, and by the far end of the slider, every one of them has gone quiet: the line struck through, the little marker turned hollow, the exact shape a real citation takes when it dies online. One row never moves. It carries a seal instead of an ordinary marker, because what it points to was signed once and stays signed. Redesign the site, rebrand the company, delete the original page: the seal holds regardless.

None of this is some rare edge case, either. Pew Research pulled a huge sample of pages that existed back in 2013 and checked, a decade later in 2023, how many were still reachable. 38% were gone: no slow fade, no redirect to an apology page, simply gone, the domain lapsed or the content deleted or the whole site rebuilt out from under the address. Narrow the lens and the pattern holds. 23% of news pages carry at least one broken link. 54% of Wikipedia pages, the internet’s own reference desk, link to at least one citation that no longer exists.

And the decay starts early. Pages Pew collected in 2021 lost about one in five of their number within just two years, a startling pace for evidence that only just went online.

This is the entire argument for a different kind of proof. A claim resting on an ordinary link is a claim resting on somebody else’s server, somebody else’s hosting bill, somebody else’s decision about what stays online next year. A signed record is built to skip that risk completely: the testimony is sealed the day it is gathered, and the seal travels with it, immune to whichever of the ordinary, mundane reasons eventually takes the original page down.