Repository navigation
Models
Pre-release
Spot Forecast model de_lu-2026-11
Walk-forward, out of sample: issue days 2026-05-05 until the training cutoff 2026-11-01 (target days before it), monthly refit. Last 3 months = target days from 2026-08-01. Naive = same hour last week; skill = 1 - MAE / naive MAE. Peak hours: actual > 23.95 ct/kWh, bias = forecast - actual. P90 / P10 calibrated with the rolling conformal offset (56 days). Prices in ct/kWh; all = mean over horizons.
| horizon | MAE | MAE last 3 mo | naive MAE | skill | peak MAE | peak bias | P90 exceeded | P10 undercut | neg. hours with P10 < 0 | rows |
|---|---|---|---|---|---|---|---|---|---|---|
| D+1 | 2.04 | 2.36 | 4.61 | 0.56 | 6.39 | -5.89 | 12.5 % | 10.8 % | 89.2 % | 4297 |
| D+2 | 2.16 | 2.50 | 4.60 | 0.53 | 6.76 | -6.22 | 13.6 % | 11.7 % | 91.3 % | 4273 |
| D+3 | 2.37 | 2.76 | 4.59 | 0.48 | 7.18 | -6.79 | 13.6 % | 11.8 % | 88.1 % | 4249 |
| D+4 | 2.57 | 2.99 | 4.55 | 0.43 | 7.61 | -7.16 | 13.2 % | 11.9 % | 84.6 % | 4225 |
| D+5 | 3.10 | 3.72 | 4.55 | 0.32 | 8.84 | -8.54 | 13.4 % | 12.6 % | 83.7 % | 4201 |
| D+6 | 3.24 | 3.86 | 4.57 | 0.29 | 9.03 | -8.86 | 14.3 % | 13.1 % | 81.6 % | 4177 |
| all | 2.58 | 3.03 | 4.58 | 0.44 | 7.64 | -7.24 | 13.4 % | 12.0 % | 86.4 % | 25422 |
Quality gate: passed
Target days in 2026-10, out of sample for both models. candidate skill 0.527 > 0; no previous model to compare (OVERRIDE --accept-incomparable (set by hand): the published model de_lu-2026-10 cannot be compared (de_lu-2026-10.json.gz cannot be used: feature schema 1 != 2), accepted; no other previous model is comparable: skill only).
| model | MAE | naive MAE | skill | rows |
|---|---|---|---|---|
| candidate | 3.61 | 7.63 | 0.53 | 4470 |
| previous | - | - | - | - |
Manual override: this run was started with --accept-incomparable, so the published model was not compared with. the published model de_lu-2026-10 cannot be compared (de_lu-2026-10.json.gz cannot be used: feature schema 1 != 2), accepted; no other previous model is comparable: skill only
Spot Forecast model nl-2026-11
Walk-forward, out of sample: issue days 2026-05-05 until the training cutoff 2026-11-01 (target days before it), monthly refit. Last 3 months = target days from 2026-08-01. Naive = same hour last week; skill = 1 - MAE / naive MAE. Peak hours: actual > 23.55 ct/kWh, bias = forecast - actual. P90 / P10 calibrated with the rolling conformal offset (56 days). Prices in ct/kWh; all = mean over horizons.
| horizon | MAE | MAE last 3 mo | naive MAE | skill | peak MAE | peak bias | P90 exceeded | P10 undercut | neg. hours with P10 < 0 | rows |
|---|---|---|---|---|---|---|---|---|---|---|
| D+1 | 2.31 | 2.66 | 4.31 | 0.47 | 6.62 | -6.14 | 14.0 % | 10.3 % | 91.3 % | 4297 |
| D+2 | 2.45 | 2.86 | 4.30 | 0.43 | 7.24 | -6.80 | 14.0 % | 10.9 % | 87.5 % | 4273 |
| D+3 | 2.68 | 3.18 | 4.29 | 0.38 | 8.12 | -7.83 | 15.6 % | 11.6 % | 87.5 % | 4249 |
| D+4 | 2.84 | 3.34 | 4.25 | 0.33 | 8.00 | -7.75 | 14.7 % | 11.5 % | 83.7 % | 4225 |
| D+5 | 3.07 | 3.62 | 4.26 | 0.28 | 8.82 | -8.50 | 14.7 % | 11.2 % | 81.5 % | 4201 |
| D+6 | 3.22 | 3.86 | 4.28 | 0.25 | 8.82 | -8.57 | 16.2 % | 12.6 % | 75.9 % | 4177 |
| all | 2.76 | 3.25 | 4.28 | 0.35 | 7.94 | -7.60 | 14.9 % | 11.4 % | 84.5 % | 25422 |
Quality gate: passed
Target days in 2026-10, out of sample for both models. candidate skill 0.437 > 0; no previous model to compare (no previous model in the manifest).
| model | MAE | naive MAE | skill | rows |
|---|---|---|---|---|
| candidate | 4.12 | 7.32 | 0.44 | 4470 |
| previous | - | - | - | - |
Spot Forecast model dk1-2026-11
Walk-forward, out of sample: issue days 2026-05-05 until the training cutoff 2026-11-01 (target days before it), monthly refit. Last 3 months = target days from 2026-08-01. Naive = same hour last week; skill = 1 - MAE / naive MAE. Peak hours: actual > 23.23 ct/kWh, bias = forecast - actual. P90 / P10 calibrated with the rolling conformal offset (56 days). Prices in ct/kWh; all = mean over horizons.
Blend: 0.5 x own model + 0.5 x DE-LU model de_lu-2026-11. Walk-forward MAE over the horizons (ct/kWh): own model 2.80, DE-LU model 2.66, blend 2.61.
| horizon | MAE | MAE last 3 mo | naive MAE | skill | peak MAE | peak bias | P90 exceeded | P10 undercut | neg. hours with P10 < 0 | rows |
|---|---|---|---|---|---|---|---|---|---|---|
| D+1 | 2.01 | 2.29 | 4.90 | 0.59 | 6.37 | -5.75 | 12.0 % | 10.5 % | 92.5 % | 4297 |
| D+2 | 2.17 | 2.49 | 4.90 | 0.56 | 6.90 | -6.29 | 13.6 % | 11.8 % | 93.3 % | 4273 |
| D+3 | 2.40 | 2.76 | 4.89 | 0.51 | 7.31 | -6.89 | 13.3 % | 11.9 % | 93.3 % | 4249 |
| D+4 | 2.64 | 3.03 | 4.89 | 0.46 | 7.80 | -7.40 | 12.4 % | 12.1 % | 86.6 % | 4225 |
| D+5 | 3.14 | 3.73 | 4.90 | 0.36 | 9.12 | -8.85 | 13.6 % | 12.6 % | 88.8 % | 4201 |
| D+6 | 3.33 | 3.90 | 4.92 | 0.32 | 9.33 | -9.22 | 13.9 % | 13.1 % | 79.9 % | 4177 |
| all | 2.61 | 3.03 | 4.90 | 0.47 | 7.81 | -7.40 | 13.1 % | 12.0 % | 89.1 % | 25422 |
Quality gate: passed
Target days in 2026-10, out of sample for both models. candidate skill 0.554 > 0; no previous model to compare (no previous model in the manifest).
| model | MAE | naive MAE | skill | rows |
|---|---|---|---|---|
| candidate | 3.40 | 7.62 | 0.55 | 4470 |
| previous | - | - | - | - |
Spot Forecast model dk2-2026-11
Walk-forward, out of sample: issue days 2026-05-05 until the training cutoff 2026-11-01 (target days before it), monthly refit. Last 3 months = target days from 2026-08-01. Naive = same hour last week; skill = 1 - MAE / naive MAE. Peak hours: actual > 24.08 ct/kWh, bias = forecast - actual. P90 / P10 calibrated with the rolling conformal offset (56 days). Prices in ct/kWh; all = mean over horizons.
Blend: 0.5 x own model + 0.5 x DE-LU model de_lu-2026-11. Walk-forward MAE over the horizons (ct/kWh): own model 2.78, DE-LU model 2.72, blend 2.63.
| horizon | MAE | MAE last 3 mo | naive MAE | skill | peak MAE | peak bias | P90 exceeded | P10 undercut | neg. hours with P10 < 0 | rows |
|---|---|---|---|---|---|---|---|---|---|---|
| D+1 | 2.08 | 2.33 | 4.85 | 0.57 | 6.59 | -6.09 | 12.9 % | 10.9 % | 97.9 % | 4297 |
| D+2 | 2.21 | 2.51 | 4.85 | 0.54 | 7.10 | -6.60 | 13.6 % | 11.7 % | 100.0 % | 4273 |
| D+3 | 2.39 | 2.71 | 4.84 | 0.51 | 7.44 | -7.07 | 13.7 % | 11.8 % | 97.9 % | 4249 |
| D+4 | 2.61 | 3.01 | 4.84 | 0.46 | 7.98 | -7.66 | 12.9 % | 11.9 % | 97.9 % | 4225 |
| D+5 | 3.13 | 3.70 | 4.86 | 0.36 | 9.54 | -9.33 | 13.8 % | 12.5 % | 93.8 % | 4201 |
| D+6 | 3.34 | 3.91 | 4.88 | 0.31 | 9.70 | -9.62 | 14.7 % | 13.5 % | 89.6 % | 4177 |
| all | 2.63 | 3.03 | 4.85 | 0.46 | 8.06 | -7.73 | 13.6 % | 12.1 % | 96.2 % | 25422 |
Quality gate: passed
Target days in 2026-10, out of sample for both models. candidate skill 0.545 > 0; no previous model to compare (no previous model in the manifest).
| model | MAE | naive MAE | skill | rows |
|---|---|---|---|---|
| candidate | 3.47 | 7.63 | 0.54 | 4470 |
| previous | - | - | - | - |
Spot Forecast model no2-2026-11
Walk-forward, out of sample: issue days 2026-05-05 until the training cutoff 2026-11-01 (target days before it), monthly refit. Last 3 months = target days from 2026-08-01. Naive = same hour last week; skill = 1 - MAE / naive MAE. Peak hours: actual > 17.81 ct/kWh, bias = forecast - actual. P90 / P10 calibrated with the rolling conformal offset (56 days). Prices in ct/kWh; all = mean over horizons.
| horizon | MAE | MAE last 3 mo | naive MAE | skill | peak MAE | peak bias | P90 exceeded | P10 undercut | neg. hours with P10 < 0 | rows |
|---|---|---|---|---|---|---|---|---|---|---|
| D+1 | 1.79 | 2.06 | 3.13 | 0.43 | 3.47 | -3.47 | 13.7 % | 8.8 % | 57.1 % | 4297 |
| D+2 | 1.93 | 2.26 | 3.13 | 0.38 | 4.08 | -4.07 | 14.9 % | 9.7 % | 50.0 % | 4273 |
| D+3 | 2.10 | 2.54 | 3.11 | 0.33 | 4.42 | -4.41 | 15.6 % | 10.3 % | 53.6 % | 4249 |
| D+4 | 2.15 | 2.59 | 3.11 | 0.31 | 4.69 | -4.69 | 15.6 % | 10.2 % | 57.1 % | 4225 |
| D+5 | 2.23 | 2.72 | 3.11 | 0.28 | 4.76 | -4.73 | 15.1 % | 11.1 % | 55.4 % | 4201 |
| D+6 | 2.33 | 2.88 | 3.12 | 0.25 | 4.77 | -4.75 | 15.3 % | 11.8 % | 53.6 % | 4177 |
| all | 2.09 | 2.51 | 3.12 | 0.33 | 4.36 | -4.35 | 15.0 % | 10.4 % | 54.5 % | 25422 |
Quality gate: passed
Target days in 2026-10, out of sample for both models. candidate skill 0.441 > 0; no previous model to compare (no previous model in the manifest).
| model | MAE | naive MAE | skill | rows |
|---|---|---|---|---|
| candidate | 2.37 | 4.24 | 0.44 | 4470 |
| previous | - | - | - | - |
Spot Forecast model de_lu-2026-10
Walk-forward, out of sample: issue days 2025-04-01 until the training cutoff 2026-10-01 (target days before it), monthly refit. Last 3 months = target days from 2026-07-01. Naive = same hour last week; skill = 1 - MAE / naive MAE. Peak hours: actual > 21.19 ct/kWh, bias = forecast - actual. P90 / P10 calibrated with the rolling conformal offset (56 days). Prices in ct/kWh; all = mean over horizons.
| horizon | MAE | MAE last 3 mo | naive MAE | skill | peak MAE | peak bias | P90 exceeded | P10 undercut | neg. hours with P10 < 0 | rows |
|---|---|---|---|---|---|---|---|---|---|---|
| D+1 | 1.79 | 2.36 | 3.39 | 0.47 | 8.60 | -8.42 | 12.0 % | 10.3 % | 84.8 % | 13128 |
| D+2 | 1.88 | 2.49 | 3.39 | 0.45 | 8.97 | -8.79 | 12.0 % | 10.4 % | 82.1 % | 13104 |
| D+3 | 1.96 | 2.64 | 3.39 | 0.42 | 9.15 | -8.95 | 12.2 % | 10.7 % | 81.7 % | 13080 |
| D+4 | 2.12 | 2.86 | 3.39 | 0.38 | 9.91 | -9.64 | 12.3 % | 10.4 % | 81.0 % | 13056 |
| D+5 | 2.36 | 3.18 | 3.39 | 0.30 | 10.48 | -10.26 | 12.7 % | 10.4 % | 77.9 % | 13032 |
| D+6 | 2.48 | 3.20 | 3.39 | 0.27 | 10.45 | -10.28 | 12.4 % | 10.9 % | 81.5 % | 13008 |
| all | 2.10 | 2.79 | 3.39 | 0.38 | 9.59 | -9.39 | 12.3 % | 10.5 % | 81.5 % | 78408 |