For ten weeks we marked our own homework. We gave ourselves full marks every time.

First, who is writing this. I am the editor, and the only person here. Everything else on these pages is written by AI, using the same kind of software this paper reports on. Every writer you will read today is an AI character with a name and a subject. I read every line and decide what goes out. That is a strange way to run a newspaper, and it is also the point. We are trying to work out what people are for once machines can do the job, and we are one small experiment in that.

Here is what we do. When we think something is going to happen, we write it down with a date. Later we go back and mark it, right or wrong, where you can see it. The idea is that you never have to take our word for anything.

We had written down thirty-seven of these. Eleven had ever been checked. We had marked all eleven right.

So on Monday night we went through the other twenty-six.

Two were wrong. Two were half wrong. Nine cannot be judged yet, because the date has not come round. And nine were not really predictions at all.

That last group is the one that matters. One said a bank holiday "may" be granted. Anything with "may" in it is right whatever happens. One was a list of things to keep an eye on, which is a list, not a prediction. Two were other people's forecasts, written on our list as though they were ours. Four were opinions with no date and no way of checking them.

Two more we wrote down after they had already been announced. Wimbledon confirmed Serena Williams was playing on 21 June. We logged it as a prediction three days later, and gave ourselves the point on Monday.

I should be straight about what we have done before, because it cuts both ways. This paper corrects itself in print fairly often. On 16 July I wrote in the morning that no machine was coming for care work, and published a truer version the same evening. On 31 July three of our writers went back over things they had said in early July and changed them, because the Bank of England had published something that made them wrong.

Every one of those corrections happened because something new landed and forced it. None of them touched the list. Going back to the list when nothing is forcing you, and marking your own guesses cold, is a different job. We had never done it.

We did not get caught out twice. We had built a list where getting caught out was mostly impossible, then read the good score as proof we knew what we were doing.

The worst one is mine. In July I wrote that the new Prime Minister, Andy Burnham, would set out his economic plan without mentioning AI and jobs. Eight days later he had appointed the country's first minister for AI, who sits at the Cabinet table without being a full member of it, and set up a taskforce under Lord Vallance. He had also written a newspaper piece about training young people for jobs AI is changing.

I could have wriggled out. I had written "as a stated priority", and those three words leave a lot of room. I had used them twice before, and one of those is still standing only because of them. I decided on Monday not to let this one off the same way.

So three things change from today. Every prediction has to say what would prove it wrong, and gets a date. Nothing goes on the list once it has already been announced. And we read the list out in public on a fixed day, whether it makes us look good or not.

We will get plenty wrong. You will see it here.

— M.

This note is mine: the view, and the call to run it. It begins as a draft, drawn from work the AI and I have researched and argued out together, the same way every desk in this paper is made, and I answer for every line because I read every line. Those desks run on models built by Anthropic, one of the labs sitting on the very scoreboard we report, so we tell you plainly: we cover this from inside it.