The Correction Column: Why a Record You Cannot Quietly Fix Is a Different Kind of Trust

Print was never honest by temperament; it was honest by physics, and digital publishing dissolved the physics while keeping the pride.

Somewhere on the second page of a newspaper, in small type, the paper tells you it was wrong. A name misspelled, a figure inverted. The correction column is the least read feature in journalism and among the most important, because it exists for a reason that has nothing to do with virtue. Once a hundred thousand copies have gone out on the vans, the publisher cannot reach them. The wrong version is in kitchens, on trains, in the bound volumes of a public library. The only remedy left is to say so in the next edition, in public.

That is the whole argument. Print was not honest by temperament; it was honest by physics. The impossibility of quiet retrieval forced accountability into the open, and the institutions journalism is proudest of grew up around that impossibility, not around anyone's good character. Digital publishing dissolved the physics and kept the pride. An article is now a row in a database, the database has an owner, and the owner can revise the row in place, at which point yesterday's version does not become obsolete: it stops existing. The distinction that matters is not print against digital. It is whether the alteration of the record is itself part of the record.

Corrections practice hardened into institution within living memory. The New York Times began gathering its corrections in one fixed place, page A2, in 1972, setting one of the precedents that standardised the practice. In 1997 the Guardian appointed Ian Mayes its first readers' editor, an in-house ombudsman fielding, by his own account, at least sixty complaints a day. Mayes made it a question of consistency: news organisations that "constantly call on others to be accountable" cannot exempt themselves. Notice the assumption underneath: the erroneous version is beyond recall. A correction is a message addressed to a record its publisher no longer controls.

What put the record beyond recall was redundancy, some of it legislated. England's Licensing Act of 1662, a censorship statute forbidding publication without prior licence, carried inside it the first formal library deposit requirement, in favour of Oxford, Cambridge and the king's library. The licensing apparatus is long dead; the depositing outlived it. Redundancy makes alteration expensive. When Lavrentiy Beria was executed in 1953, his entry in the Great Soviet Encyclopedia had already gone to subscribers, and the state's recourse was to mail them replacement pages with instructions to cut out the article and paste in expanded adjacent entries, among them the Bering Sea. To edit a distributed record you must persuade every holder to pick up the scissors.

The romance of print immutability should not survive contact with the archives. Paper decays, and its custodians have not always been gentle: Nicholson Baker's Double Fold (2001) charged libraries with discarding bound newspaper volumes once microfilm existed. Nor was the printed record ever a strict transcript. The Congressional Record is only a substantially verbatim account, in the Congressional Research Service's phrase, and House members may ask leave to revise and extend their remarks. But look how the concession is handled: matter inserted rather than spoken is set in a different type style in the House and marked with a bullet in the Senate. The record permits alteration and discloses it in the same gesture. That convention is the entire principle, worked out on paper long before anyone needed a hash.

What digital publishing changed can be measured. The Pew Research Center's 2024 study of link rot found that 38 percent of webpages that existed in 2013 were gone by October 2023. Across the full decade it sampled, a quarter of all pages had become inaccessible, and the revealing figure is how that quarter divides: 16 percent had disappeared individually from root domains still serving content perfectly well, against 9 percent lost because an entire domain had died. The site is fine. The page is not.

The response has been to build watchers from outside. NewsDiffs, assembled at a Mozilla, Knight and MIT hackathon in June 2012 over thirty-eight hours, archived successive versions of articles at outlets including the New York Times and CNN so readers could see what had quietly changed. That it had to be built by outsiders, in a weekend, is the confession. The profession's own argument was about deletion. Kathy English's 2009 survey for the Associated Press Managing Editors' Online Credibility Project, covering 110 North American news organisations, found 50.8 percent with no policy at all on unpublishing. The Canadian Association of Journalists set the norm out plainly the following year: "News organizations do not rewrite history or make news disappear."

A norm is a promise, and a promise is exactly the trust the correction column was invented to make unnecessary. The most interesting settlement of the past two decades left the record alone. In Google Spain, decided by the Court of Justice of the European Union on 13 May 2014, a Spanish man objected that searches for his name surfaced pages La Vanguardia had printed in 1998 about the auction of his property to recover social security debts. The judgment left the newspaper's pages standing and reached for the index instead, requiring the search engine to delist. Findability was governed separately from the record: closer to the layered, polycentric arrangements of Part V than to either deletion or archival absolutism.

No design escapes the problem, and the better ones say so. Wikipedia is heavily audited: every edit carries a timestamp, an author name and an optional summary, and the whole history is open to anyone. And yet, as of August 2026, forty-five accounts on the English Wikipedia hold the suppression permission, which expunges a revision from ordinary access even by administrators. That layer exists for real reasons, including personal information and libel. But there is always a floor of privileged access. The honest question is never whether anyone can alter the record. It is how thin that floor is, who stands on it, and whether the rest of us can see the shape of what was done.

So this register carries sequence numbers and content hashes, corrections append in public with the superseded hash retained, and nothing here is silently edited. Not because software confers virtue: Part VI was blunt that protocols concentrate like everything else when nobody is watching. Print had a useful property by accident of its physics, journalism built its best institutions on that accident, and the accident is gone. What Part V called cheap verification, the capacity of any member to check the record without asking permission, must now be engineered on purpose or it will not exist. A database you are asked to trust and a record you are equipped to audit look identical on a screen. They are not the same object, and the difference shows only on the day someone would prefer the past to have been different.