Predictions

Can forecasts be made checkable, and then actually checked?

What it is

Predictions is the lab's world-forecasting desk. Every forecast is written in plain language and given a probability, a deadline, and a named source that will settle it. Then it is locked: no vague verses, no retrofitting after the fact, no silent edits.

Forecasts are published in registers and batches across politics, geopolitics, the economy, climate, space, technology, health, and culture. When a deadline passes, the desk grades itself in public, hits and misses alike, with a running scoreboard.

How the desk works

Locked at publication and never edited. Settled by the named source. Scored publicly with a Brier score and a calibration line, so confidence is graded as well as correctness.

Inside the case site

  • The register
  • Batches
  • Election 2026
  • Scoreboard
  • Method
  • Lesson zero

Latest from Predictions

PredictionsLab log

Predictions publishes Edition 5

Edition 5 of the forecasting desk added new batches to the register and updated the public scoreboard, where settled forecasts are graded hits and misses alike.

Read the original ↗
PredictionsLab log

Predictions locks its founding register

The forecasting desk locked its founding register of world predictions for 2027 and a near-term batch that settles between December 2026 and April 2027. Each forecast carries a probability, a deadline, and the named source that settles it, and none can be edited after locking.

Read the original ↗