Drift

How much the dataset moves from one research run to the next, and how confident it is.

This page measures the pipeline, not the world. Every Monday the research pipeline re-scores every country from scratch, and a stability gate lets a score change through only with new evidence or when it repeats on the next run (see the methodology). The figures below show what came through that gate on each run: how many countries changed and on which dimension, how large the moves were, the confidence the pipeline assigned, and how closely a run agrees with a hand-checked gold set. A busy week here can mean regulation moved or that the model re-read the same evidence differently; the weekly digest says which, country by country. Nothing on this page is a measure of how well any country governs AI.