CERES
DashboardRegionsMapSub-nationalMethodologyAPIAboutTrack RecordValidationImpactDataChangelogSign In
Sign In →
Accuracy Metrics
0 predictions \u00B7 first grades expected Aug\u2013Oct 2026
Brier Score
Pending
Target < 0.10 Expected Aug\u2013Oct 2026
SI Coverage (90%)
Pending
Target > 88% Expected Aug\u2013Oct 2026
Brier Skill Score
Pending
Target > 0 Expected Aug\u2013Oct 2026
Total Predictions
0
Target ≥ 43/week ✗ Missed

Note: 0 predictions issued; the first T+90 grading windows opened June 2026. Observed IPC/FEWS NET classifications publish on a 24 month lag, so the earliest grades land AugOct 2026. Grading is automated and pre-registered: the Brier decomposition computes the moment 10 outcomes are published, with no manual intervention.

Pending Verification

Predictions Awaiting Grading Window

Loading…
Loading verification ledger\u2026
Public Prediction Ledger

Graded Predictions: Forward Validation

Loading…
FIRST GRADES EXPECTED AUG–OCT 2026

The first T+90 grading windows opened June 2026. Grades are written automatically once IPC/FEWS NET publish the observed Current-Situation classification for each region, which lags the target window by 24 months. This ledger updates every Monday and nothing is graded against projected data.

Calibration · Awaiting Prospective Data

87 IPC Records · 31 Countries · 4 Back-validation Cases

Model initialised against 87 IPC transition records (2011–2023, 31 countries). 4 data-complete back-validation cases.

Reliability Diagram \u00b7 Predicted vs. Observed Probability
Perfect calibration lies on the diagonal. Points above = underconfident; below = overconfident.
0%0%20%20%40%40%60%60%80%80%100%100%Predicted ProbabilityPerfect
Well-calibrated (±10%)
Outside tolerance
Perfect calibration
Calibration by Predicted Probability Bin

Grey = ideal calibration · Amber = CERES observed rate · (n) = predictions in bin

Validation Dataset Breakdown
IPC transition records87 country-seasons
Countries represented31
Time period2011–2023
Phase 4–5 events18
Back-validation cases4 (data-complete only)
Perturbation drawsn=2,000 per prediction
Interval typeInput-perturbation 90%
Pre-Registered Calibration Protocol

What We Commit to Measuring

Table 1 from the CERES preprint. These metrics were pre-registered before any prospective outcome data was collected. No metrics will be selectively reported: all graded predictions remain permanently visible. Minimum sample sizes are fixed; targets cannot be revised retroactively.

MetricDefinitionMin. NTarget dateStatus
Brier ScoreMean (P̂₃ − O₃)²100 predictionsJun 2026 Pending
Brier Skill Score1 − BS / BS_climatology100 predictionsJun 2026 Pending
TIER-1 PrecisionTrue TIER-1 / all TIER-1 issued30 TIER-1 alertsSep 2026 Pending
TIER-1 RecallTrue TIER-1 / all Phase 4+ events10 Phase 4+ eventsSep 2026 Pending
Sensitivity interval coverageFraction outcomes in 90% interval200 predictionsSep 2026 Pending
CRPS (ordered categorical)Full distribution vs IPC phase500 predictionsMar 2027 Pending
Reliability diagramForecast prob. vs empirical frequency500 predictionsMar 2027 Pending

Pre-registered in Pedersen (2026), Table 1. Protocol locked prior to accumulation of prospective outcome data.

The CERES Transparency Commitment

Every prediction CERES issues is permanently recorded in this ledger with a timestamp, probability estimate, 90% sensitivity interval, and T+90 day grading date. We do not remove predictions that prove incorrect. We analyse and publish the reasons for forecast errors. The accuracy record here is the complete record: there is no curated subset. This is the foundation of institutional trust.