Grid operators (TSOs) publish their own day-ahead forecast of wind and solar output. Augur forecasts the same quantities and is measured two honest ways — against the TSO at the same information cutoff, not against a stale number.
Day-ahead, same cutoff (the real test): we beat every TSO's wind forecast by 7–11%, and beat Romania's solar forecast by 64% — because Transelectrica's is genuinely weak. Where a TSO's solar forecast is already excellent (Germany, Belgium) we roughly match it. Our edge scales with how weak the incumbent is.
Intraday (re-forecast every 15 min): using fresh measurements we beat the TSO's own continuously-updated forecast by 40–78% at 15 minutes ahead, fading to a tie by a few hours out. Strict temporal split, public data only, no lookahead. Where we lose, the table shows it.
Live, not just backtested: since June 2026 Augur submits hourly probabilistic forecasts (P10/P50/P90, 15-min resolution) to Predico, the collaborative forecasting platform used by Elia, Belgium's grid operator — every gate, around the clock, unattended — and self-scores each submission against the operator's published forecast once actuals settle.
Production models — trained through the latest available data, refreshed every 15 min. (Backtest panels below use the frozen 2026-01-01-cutoff models for honest out-of-sample skill.)
Forecast issued once at the 08:00 UTC gate for 16–38h ahead, scored vs the TSO's own day-ahead forecast for the same slots — same information, same task. Skill tracks how weak the incumbent is: we beat every TSO on wind, beat a weak solar forecaster (RO) big, and can't beat an excellent one (DE/BE solar). nMAE shown so markets are comparable.
| market | target | n | TSO DA nMAE | Augur nMAE | RMSE skill | MAE skill |
|---|
| target | horizon | n test | TSO DA RMSE | model RMSE | skill vs DA | skill vs TSO intraday |
|---|
Intraday re-forecasts every 15 min using fresh actuals. "Skill vs TSO intraday" is the fair fight — vs the TSO's own continuously-updated forecast (the best it had by delivery). We win at short horizons, fade to losses by 4h as the TSO's near-delivery info wins out. RO has no published intraday forecast, so only vs-DA is shown there.
| market | model MAE | persistence MAE | MAE skill | directional | P80 cov |
|---|
Chart: DE-LU, last 7 test days (actual vs Q50, Q10–Q90 band). Same model across all 5 markets — residual-load-driven, quantiles conformal-calibrated. Skill tracks how renewable-driven each market's price is: high for DE/NL/BE, moderate for nuclear FR, lowest for hydro-driven AT. No TSO price forecast exists; baselines are persistence/seasonal. MAE + directional matter more than RMSE — price is near-random in level.
| horizon | model MAE | persistence MAE | MAE skill | premium-direction acc | P80 cov |
|---|
Imbalance price (what BRPs pay to deviate) is extremely volatile. We beat persistence for ≥30 min (it's frozen quarter-to-quarter at 15 min) and the day-ahead anchor ≥1 h. Honest limit: predicting the long/short *direction* adds nothing over the current imbalance state. Intervals conformal-calibrated.
Backtest results against each transmission system operator's own published forecast, identical delivery slots, temporal split at 2026-01-01, public data only. Negative skill means the TSO wins — shown either way. Since June 2026 the same models also submit live hourly P10/P50/P90 forecasts to Predico, the collaborative forecasting platform used by Elia (Belgium's TSO).
| market | Augur | TSO DA | RMSE skill |
|---|---|---|---|
| DE solar | 1338 MW | 1264 MW | -5.8% |
| DE wind onshore | 1799 MW | 1996 MW | +9.9% |
| DE wind offshore | 773 MW | 788 MW | +1.9% |
| RO solar | 108 MW | 302 MW | +64.0% |
| RO wind onshore | 198 MW | 236 MW | +16.1% |
| BE solar | 395 MW | 368 MW | -7.4% |
| BE wind offshore | 225 MW | 269 MW | +16.2% |
| market | Augur | TSO intraday | RMSE skill |
|---|---|---|---|
| DE solar | 271 MW | 756 MW | +64.1% |
| DE wind onshore | 597 MW | 1028 MW | +41.9% |
| DE wind offshore | 149 MW | 505 MW | +70.6% |
| BE solar | 46 MW | 209 MW | +78.1% |
| BE wind offshore | 59 MW | 215 MW | +72.5% |