Notes
What makes a prediction checkable.
A closed set of short pieces, not a blog. Each one exists because the archive can answer it with a number, and every number is counted when the page is built rather than written into the text.
- People who write predictions get scored. People who say them don't.
Six of six for Jakob Nielsen. One of seven for Elon Musk. The difference isn't rigour, it's the medium.
August 29, 2026
- What makes a prediction testable
Four things have to be present. Most public predictions have two or three, which is why so few of them ever get marked right or wrong.
August 29, 2026
- The goalposts problem
Everyone knows to screenshot the claim. Almost nobody writes down what would settle it, which is the half that gets quietly rewritten.
August 29, 2026
- What an accuracy score actually measures
Change which claims are eligible and the same forecaster moves twenty points. Nobody has to lie for that to happen.
August 29, 2026
- How a track record lies without a single false statement
Selection is the whole trick. You never have to fabricate anything — you just have to pick.
August 29, 2026
- How to check whether someone actually predicted what they say they did
Five steps, and four of them happen before the outcome is known. Most disputes die at the second one.
August 29, 2026
- Sometimes the honest verdict is that nobody can tell
The pressure is always to produce a verdict. Refusing is the harder discipline and the one that keeps the rest of the numbers worth anything.
August 29, 2026
- Everyone predicts in January and finds out in December
The clustering is severe enough to be a structural fact about the genre, not a quirk of one archive.
August 29, 2026
- Bitcoin price predictions almost never carry a date
A price with no deadline is not a prediction. It is a mood with a figure attached.
October 3, 2026
- Election predictions are the only kind everyone dates
It is not rigour. It is that polling day is already in the diary.
October 3, 2026
- How far ahead people actually predict
One year is not a forecasting horizon. It is the length of a publishing cycle that happens to be called a year.
October 3, 2026
- AGI timeline predictions mostly cannot be settled
The date is the easy half. Nobody agrees what has to happen on it.
October 3, 2026
- Will AI take your job? What the claims actually say
Almost every claim about AI and jobs is about a whole economy — the hardest kind to check and the easiest kind to repeat.
October 3, 2026
- A prediction, a commitment and a position are not the same thing
Saying what will happen, promising what you will do, and stating what you believe are three different risks.
October 3, 2026
- The further out a prediction looks, the less it usually says
A six-year claim is the one nobody will remember. For the person making it, that is a feature.
October 3, 2026
- The three ways a claim gets settled
The weakest link in any verdict is who decided it. These are ordered by how little that matters.
October 3, 2026
- What happens when a record turns out to be wrong
An archive you can edit is an archive somebody can edit. The fix has to leave a mark.
October 3, 2026
- Having the most claims on the record means nothing at all
Appearing often is a fact about publishing habits. It is not evidence of anything else.
October 3, 2026