The public operational record of CERES: every alert issued and every prediction logged for grading, in numbers, with sources, updated automatically from live data. Forward accuracy results are forthcoming as the first T+90 grading windows resolve.
Standard humanitarian early warning systems issue effective alerts 30–45 days before crisis thresholds. Pre-positioning food aid and mobilising emergency logistics requires 60–90 days minimum. CERES is designed to close that gap.
CERES issues 90-day horizon predictions, each carrying a calibrated probability, a 90% sensitivity interval, and named driver causes. Every prediction is timestamped on issue and graded against IPC Phase 3+ outcomes when the T+90 date arrives.
Prospective hit rate and lead time metrics will populate automatically as graded predictions accumulate. The first T+90 grading windows opened June 2026, and the first graded outcomes are expected Aug–Oct 2026. Performance targets: Brier score <0.10 · Tier I precision >80% · SI coverage >88%.
Every CERES prediction carries a T+90 grading date. When that date arrives, we compare our probabilistic forecast against the IPC phase classification published by OCHA, FEWS NET, or the IPC Global Platform for that region.
Predictions are graded on: (1) whether our predicted tier matches the observed IPC phase, (2) whether the actual outcome fell within our 90% sensitivity interval, and (3) our Brier score, the standard probabilistic calibration metric used in meteorological forecasting.
This ledger is public, permanent, and updated automatically. We do not remove incorrect predictions. View the full ledger →
All CERES predictions are available as open data under CC BY 4.0. Download the full archive for independent validation, academic research, or integration into existing EWS workflows.