Historical May 2026 prototype; current scores are unvalidated. This article describes a dated design and sample output. It does not rank which country forecast is most trustworthy today.
What the prototype combined
The proposed score was a hand-weighted sum of interval width (0.30), evidence age (0.20), calibration drift (0.20), sample size (0.15), and cross-model disagreement (0.15). It was a heuristic, not a probability that a forecast would be correct. The May implementation used forecast, multi-horizon, and classifier outputs with different targets.
The May writeup ranked Pakistan, Turkey, and Iran using then-available interval, ACI, and drift fields. Those rankings and underlying values are historical and unvalidated; they cannot be carried forward as current country scores. No eligible cohort currently supports publishing the 7-day forecast interval or empirical calibration claims.
Current interpretation
The public 7-day forecast withholds interval_90 while interval
coverage remains unverified. Any uncertainty endpoint or saved score must
be checked for fresh input dates and outcome eligibility before use.
A high historical score never proved a forecast wrong; a low score never
proved it reliable.
To restore a daily ranking, rebuild the score from current, eligible inputs, validate the calibration component against settled outcomes, and publish the scoring date, cohort size, and missing-component behavior. Until then, the May figures are research notes only.