PsychoHAistoryResearch Initiative

Journal · No. 07

What Makes a Prediction Scientifically Useful?

Stated in advance, falsifiable, calibrated and scored.

A useful prediction is made before the outcome is known, and it is specific enough to be wrong. “The project will probably be fine” cannot fail; “80% likely to finish within 30–42 working days” can.

Scoring

Weather forecasters have long scored probabilistic forecasts; the Brier score dates from 1950. Proper scoring rules reward honesty: you cannot improve your score by exaggerating confidence.

The ledger

This is why the proposed architecture has a Prediction Ledger. Every prediction will be stored with its data window, model version and assumptions, then compared with what actually happened. A model that cannot keep a ledger cannot claim to predict.

This is a concept note. It reports no experiments and no results.