Confidence stops being vibes when it has receipts.
Every public claim gets a probability, a threshold, and a date. Wins count. Losses count. Unresolved claims stay visible until the evidence arrives.
- 7
- Visible claims
- 7
- Pending
- 0
- Resolved
- 10
- Sealed time-capsule
The grade, above the fold.
A page can sound smart and still hide from measurement. This one puts the score where you can't miss it.
Brier = mean (prior − outcome)². Lower means cleaner judgment. Calibration lights up as predictions resolve — zero is the honest starting point.
No resolved predictions yet. The first horizons arrive December 2026; the score appears the day they do, whichever way they land.
Calibration
predicted prob → realized rate
- 0–10%n=0
- 10–25%n=0
- 25–50%n=0
- 50–75%n=0
- 75–100%n=0
Calibration plot lights up once predictions resolve. Zero is the honest starting point.
Four moves, every claim.
Prior
A probability before the result is known. On the record, in public.
Threshold
The exact evidence that changes the call. Written down in advance.
Resolve
A date the claim has to answer to. No quiet extensions.
Grade
Brier score moves with every outcome. Lower means cleaner judgment.
The audit ledger.
By the end of 2026, a normal smartphone will come with AI writing or summarizing turned on by default.
By the end of 2026, at least one major streaming service will raise the price of its ad-free plan again.
By the end of 2026, a bad AI search answer will become a mainstream consumer trust story.
By summer 2027, more schools will stop pretending AI homework does not exist and will write rules for it.
By the end of 2027, grocery apps will use AI to make coupons feel more personal and harder to ignore.
By the end of 2027, more people will admit their saved lists are bigger than their finished reading.
10 long-horizon claims are locked until May 6, 2031.
The claims, priors, and thresholds were committed before the outcome window. When the seal opens, they join the same public ledger and get graded the same way.