Ledger // 3 claims on record

Predictions,
graded

Anyone can sound clever about the future. The test is whether you were right, at the confidence you claimed, by the date you named. Each row links back to the edition where the claim was made. Resolved claims are never deleted.

Calibration // 0 resolved · 3 open

Each point is a group of claims made at the same confidence; the number inside is how many. A point on the diagonal means the claims came true exactly as often as promised. Below the line: overconfident. Above: underconfident. Hollow marks on the baseline are claims not yet due. With fewer than twenty resolved claims this is a sketch, not a verdict.

Every claim

Ed.	Claim	Confidence	Resolve by	Status
001	By the end of 2028 at least one of the five largest English-language news publishers will charge model trainers more per year for archive access than it earns from print subscriptions.	55%	2028-12-31	open
001	By mid-2027 a major jurisdiction will have created a legal category for trained model weights that is explicitly neither a 'copy' nor a 'derivative work'.	40%	2027-06-30	open
001	Per-copy royalty income for recorded music will be lower in real terms in 2030 than in 2025, despite higher total industry revenue.	70%	2030-12-31	open

The annual review

Once a year, one edition grades the previous twelve months: every claim that came due, the calibration chart read out loud, and the framings that did not survive. The first is due in January 2027.

How grading works

A claim is right if it happened as stated by the resolve-by date, wrong if it plainly did not, and partial if a reasonable reader could argue either way — in which case the argument is written out under the claim. Confidence is the probability assigned at the time of writing. Over many claims, the ones marked 70% should come true about seven times in ten. If they come true nine times in ten, the writer is underconfident; if five, overconfident. Both are failures, and both get reported here once a year.