Seven CPU baselines
Persistence, seasonal naive, moving average, linear, ridge, random forest and gradient boosting.
REPRODUCIBLE FORECASTING BENCHMARK
Strong baselines before model spectacle.
HeatAQ Nexus evaluates environmental forecasts across multiple horizons with chronological splits, rolling origins, event metrics, conformal intervals and versioned research artifacts.
Leakage controls
Lagged inputs, shifted rolling features and strictly chronological evaluation.
Uncertainty
Split-conformal intervals with coverage, width and interval-score reporting.
Rolling origins
Selected models are repeatedly evaluated using only earlier observations.
Research artifacts
Leaderboard, model cards, configuration and experiment manifest.
No paid AI required
The scientific benchmark runs entirely with open Python tooling.
Metrics
MAE, RMSE, bias, R², sMAPE
Precision, recall, F1, Brier
Coverage, width, interval score
Rolling-origin evaluation