No Past Tense

A news story is not a document. It looks like one. There is a headline, a date, a byline and a fixed web address you can send to a friend, and all of that is the trouble, because underneath it is a live surface and the text on it keeps moving.

We know because we go back and look. Between 19 and 22 July we re-read 374 news articles we had already captured. A hundred and sixty-nine of them had changed since we first saved them. Ninety-nine had changed substantively: new paragraphs, altered claims, a different account of what had happened, not a corrected typo. One BBC news page covering last week's change of government had picked up several fresh paragraphs after we first read it. Same address, same headline, different article.

None of that is dishonest, and it is worth saying so before anything else. A story that breaks at nine in the morning is not finished at nine in the morning. A department returns a call, a second source firms up a number, the picture improves through the day. A newsroom that could not update its own copy would be a good deal worse than one that does. The updating is journalism working.

The problem is not the updating. It is that the update overwrites.

When a page is revised, the earlier version does not move somewhere else with a note beside it. It stops existing. The sentence that was on the screen at eleven, that a few hundred thousand people read and a few thousand repeated, is gone by three, and the page carries no trace of having said it. The record does not contradict you. It has no memory of the argument at all.

That changes what an argument about the press actually is. Someone says an outlet reported X. Someone else opens the link, finds it does not say X, and posts the link. There is no way to settle that. Not because the facts are genuinely in dispute, but because the exhibit has been edited since the disagreement began. Whoever took a screenshot wins. Whoever merely remembers loses, and remembering accurately is no defence, because memory is exactly the thing the other person is entitled to doubt.

The instinctive answer to this is fact-checking, and fact-checking is the wrong tool for it. Fact-checking adjudicates. It arrives after the event, weighs a claim against the world and issues a finding. That is a court, and a court needs its evidence intact. By the time this one sits, the exhibit has been rewritten, so the finding rests on the checker's own account of what the page used to say, which is the same weak currency as everybody else's memory with a logo on it. Adjudication also runs on trust in the adjudicator, and that is the one thing nobody has left to spend.

Archiving does something duller and much stronger. It decides nothing. It takes the photograph and keeps it, with the time on the back.

Dull is the point, and so is cheap. A capture is a fetch, a hash and a timestamp, and the machinery costs less than a phone contract. The expensive part is the discipline: doing it every fifteen minutes, on everything, before anyone knows which page will matter. That last clause is the whole game. You cannot choose to save the important ones, because importance is only visible later, and later is when the page has already moved. So you save the lot, including the overwhelming majority nobody will ever look at again. A gap in the archive cannot be filled in afterwards. There is no supplier of yesterday.

There is a temptation here to reach for a villain, and it should be resisted, because it is both wrong and weaker. Nothing we have captured shows a plan. It shows a system publishing continuously onto a surface with no past tense, which is what happens when the printing press is replaced by a database and nobody stops to ask what the paper edition had quietly been doing for us. The bound volume in the library was not neutral, and it was not accurate either. It simply could not be revised after the fact, and that turns out to have been half of its value.

So the practical rule is unglamorous. Any serious argument about what was said should open with a copy: a capture, a timestamp, a hash somebody else can check. If nobody kept one, the argument cannot be won and is better dropped than shouted. That is a real cost. It means letting go of things you are certain you read.

It is still the better trade. An argument you cannot prove is not an argument, it is a mood. The version you remember reading is not on the internet any more. Somebody has to have kept it, or as far as the record is concerned it never happened.

An opinion of the house. The argument is ours; the record beneath it belongs to no one.