VOLTSTACKREFERENCE GUIDE
DATED SNAPSHOT
REFERENCE · WEATHER FORECAST SKILL

ECMWF ensemble forecast skill for wind, solar and temperature

Scorecard as of , trailing 30 days and since . Published .

A DATED SNAPSHOT, NOT A LIVE FEED

Scorecard as of 6 Nov 2026, trailing 30 days. This is a dated snapshot, not a live feed: these are the figures as Voltstack's scorecard stood at , scored through the 00 UTC run of 6 Nov 2026, and this page does not update. Figures: ECMWF open data, CC BY 4.0, Voltstack computation; the full ECMWF credit is under the tables.

THE SHORT ANSWER

Scored on the next UTC day for seven European bidding zones, the ECMWF ensemble's average forecast landed close to what followed, but the spread of its 50 members was the wrong width. From 13 May 2026, its p10 to p90 band caught 2 m temperature only 54.5% (France) to 67.6% (Belgium) of the time, against the 80% a well-sized band catches: too narrow in every zone. For 100 m wind it caught 80.5% (Belgium) to 94.7% (GB): too wide, most of all in GB and France.

For a desk, the middle of the forecast holds up. Mean absolute error on the zone-mean temperature was 0.25 to 0.41 °C, and every zone's temperature bias sat between −0.05 and +0.03 °C. The band is what needs adjusting: a temperature band taken straight from the members has missed 32% to 46% of the time, not one time in five, so the cold tail behind an HDD position is wider than the members say.

Wind and temperature are scored against ECMWF's own analysis and solar against ECMWF's later short-range runs. Both share the model's physics, so this is the forecast's best case: against station measurements we would expect the band to miss more often.

DATED SNAPSHOTSCORECARD AS OF: 6 NOV 2026PUBLISHED: 15 NOV 2026AUTHOR: Voltstack Intelligence
Temperature band, outcomes inside
54.5% to 67.6%
seven zones, 13 May 2026 to 6 Nov 2026, to 28 Oct 2026 for Germany and Denmark West · target 80% and 1.00 · as of 6 Nov 2026, not live
Wind band, outcomes inside
80.5% to 94.7%
seven zones, 13 May 2026 to 6 Nov 2026, to 28 Oct 2026 for Germany and Denmark West · target 80% and 1.00 · as of 6 Nov 2026, not live
Temperature spread-skill
0.65 to 0.80
seven zones, 13 May 2026 to 6 Nov 2026, to 28 Oct 2026 for Germany and Denmark West · target 80% and 1.00 · as of 6 Nov 2026, not live
Wind spread-skill
1.08 to 1.70
seven zones, 13 May 2026 to 6 Nov 2026, to 28 Oct 2026 for Germany and Denmark West · target 80% and 1.00 · as of 6 Nov 2026, not live
Voltstack Analytics

The same scorecard, rescored after every run

This page stops at 6 Nov 2026. In Voltstack the scorecard is rescored after every ECMWF ensemble run Voltstack stores, next to the overnight run change by gas day.

Start a 14-day free trialRequest early access →

The button opens a 14-day free trial sign-up. Sign-up needs a work email.

The button opens the early-access request on our home page.

01 · Band coverage by zone

How often the p10 to p90 band held the outcome, as of 6 Nov 2026

Across 50 members, the band from the 10th to the 90th percentile should hold the outcome about 80% of the time. Red is a band too narrow for its errors, blue one too wide. Long window: 13 May 2026 to 6 Nov 2026, or to 28 Oct 2026 for Germany and Denmark West.

2 m temperature: share of next-day outcomes inside the p10 to p90 band
2 m temperature: ECMWF ensemble p10 to p90 band coverage by zone, long window, against the 80% targetFrance 54.5% (13 May 2026 to 6 Nov 2026); Denmark East 64.6% (13 May 2026 to 6 Nov 2026); GB 64.7% (13 May 2026 to 6 Nov 2026); Germany 65.6% (13 May 2026 to 28 Oct 2026); Denmark West 66.9% (13 May 2026 to 28 Oct 2026); Netherlands 67.3% (13 May 2026 to 6 Nov 2026); Belgium 67.6% (13 May 2026 to 6 Nov 2026). A well-sized band catches 80%.50%60%70%90%100%80% targetFrance54.5%Denmark East64.6%GB64.7%Germany65.6%Denmark West66.9%Netherlands67.3%Belgium67.6%
100 m wind speed: share of next-day outcomes inside the p10 to p90 band
100 m wind speed: ECMWF ensemble p10 to p90 band coverage by zone, long window, against the 80% targetBelgium 80.5% (13 May 2026 to 6 Nov 2026); Denmark West 81.0% (13 May 2026 to 28 Oct 2026); Denmark East 82.9% (13 May 2026 to 6 Nov 2026); Netherlands 83.9% (13 May 2026 to 6 Nov 2026); Germany 84.4% (13 May 2026 to 28 Oct 2026); France 91.1% (13 May 2026 to 6 Nov 2026); GB 94.7% (13 May 2026 to 6 Nov 2026). A well-sized band catches 80%.50%60%70%90%100%80% targetBelgium80.5%Denmark West81.0%Denmark East82.9%Netherlands83.9%Germany84.4%France91.1%GB94.7%
Solar radiation (GHI): share of next-day outcomes inside the p10 to p90 band
Solar radiation (GHI): ECMWF ensemble p10 to p90 band coverage by zone, long window, against the 80% targetFrance 83.4% (13 May 2026 to 6 Nov 2026); Belgium 84.2% (13 May 2026 to 6 Nov 2026); Netherlands 85.1% (13 May 2026 to 6 Nov 2026); Denmark East 85.5% (13 May 2026 to 6 Nov 2026); Denmark West 86.9% (13 May 2026 to 28 Oct 2026); Germany 87.2% (13 May 2026 to 28 Oct 2026); GB 94.0% (13 May 2026 to 6 Nov 2026). A well-sized band catches 80%.50%60%70%90%100%80% targetFrance83.4%Belgium84.2%Netherlands85.1%Denmark East85.5%Denmark West86.9%Germany87.2%GB94.0%

Solar sits on the wide side too, at 83.4% (France) to 94.0% (GB), and the forecast ran brighter than ECMWF's later runs in every zone, by +4.2 to +9.4 W/m2 on average over daylight slots.

02 · The scorecard, as of 6 Nov 2026

Bias, MAE, RMSE, band coverage and spread-skill by zone

Two windows per zone. The long window runs from 13 May 2026. Germany and Denmark West changed outline in Voltstack's zone map on 28 Oct 2026, so their long window is the previous outline's, 13 May 2026 to 28 Oct 2026, and their trailing 30 days hold only the 9 days of the new outline. A window on fewer than 10 days is not settled; those rows are grey and marked *.

2 m temperature (°C), scored against ECMWF's own analysis

ZoneWindowDaysBias (°C)MAE (°C)RMSE (°C)Inside p10 to p90Against 80%Spread-skill
Germany and LuxembourgLong, previous outline
13 May 2026 to 28 Oct 2026
169−0.030.280.3765.6%−14.4 pp0.77
Trailing 30 days
29 Oct 2026 to 6 Nov 2026
9 *−0.110.300.3759.4%−20.6 pp0.78
Denmark West (DK1)Long, previous outline
13 May 2026 to 28 Oct 2026
169−0.040.270.3566.9%−13.1 pp0.78
Trailing 30 days
29 Oct 2026 to 6 Nov 2026
9 *+0.010.280.3570.3%−9.7 pp0.82
NetherlandsLong
13 May 2026 to 6 Nov 2026
178−0.030.340.4667.3%−12.7 pp0.78
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
30−0.020.240.3282.7%+2.7 pp1.12
BelgiumLong
13 May 2026 to 6 Nov 2026
178+0.030.410.5767.6%−12.4 pp0.80
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
300.000.310.4177.8%−2.2 pp1.01
Denmark East (DK2)Long
13 May 2026 to 6 Nov 2026
178+0.010.330.4364.6%−15.4 pp0.79
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
30+0.100.310.4165.6%−14.4 pp0.80
Great BritainLong
13 May 2026 to 6 Nov 2026
178−0.050.250.3364.7%−15.3 pp0.76
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
30+0.020.200.2673.5%−6.5 pp0.99
FranceLong
13 May 2026 to 6 Nov 2026
178−0.020.330.4254.5%−25.5 pp0.65
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
30−0.010.220.2970.3%−9.7 pp0.89

100 m wind speed (m/s), scored against ECMWF's own analysis

ZoneWindowDaysBias (m/s)MAE (m/s)RMSE (m/s)Inside p10 to p90Against 80%Spread-skill
Germany and LuxembourgLong, previous outline
13 May 2026 to 28 Oct 2026
169+0.050.270.3684.4%+4.4 pp1.19
Trailing 30 days
29 Oct 2026 to 6 Nov 2026
9 *+0.040.190.2485.9%+5.9 pp1.37
Denmark West (DK1)Long, previous outline
13 May 2026 to 28 Oct 2026
169+0.100.430.5681.0%+1.0 pp1.08
Trailing 30 days
29 Oct 2026 to 6 Nov 2026
9 *+0.240.350.4373.4%−6.6 pp1.05
NetherlandsLong
13 May 2026 to 6 Nov 2026
178+0.010.380.5183.9%+3.9 pp1.18
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
30+0.120.410.5590.4%+10.4 pp1.41
BelgiumLong
13 May 2026 to 6 Nov 2026
178−0.020.400.5580.5%+0.5 pp1.08
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
30+0.140.400.5287.2%+7.2 pp1.45
Denmark East (DK2)Long
13 May 2026 to 6 Nov 2026
1780.000.430.5782.9%+2.9 pp1.19
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
30−0.010.430.5782.3%+2.3 pp1.29
Great BritainLong
13 May 2026 to 6 Nov 2026
178+0.140.230.3094.7%+14.7 pp1.70
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
30+0.210.290.3796.6%+16.6 pp1.74
FranceLong
13 May 2026 to 6 Nov 2026
178−0.050.180.2491.1%+11.1 pp1.47
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
30+0.020.160.2197.6%+17.6 pp2.14

Solar radiation (GHI) (W/m2), scored against ECMWF's later short-range forecasts

ZoneWindowDaysBias (W/m2)MAE (W/m2)RMSE (W/m2)Inside p10 to p90Against 80%Spread-skill
Germany and LuxembourgLong, previous outline
13 May 2026 to 28 Oct 2026
168+4.910.016.787.2%+7.2 pp1.43
Trailing 30 days
29 Oct 2026 to 6 Nov 2026
8 *+4.88.111.581.5%+1.5 pp1.28
Denmark West (DK1)Long, previous outline
13 May 2026 to 28 Oct 2026
168+6.614.324.086.9%+6.9 pp1.29
Trailing 30 days
29 Oct 2026 to 6 Nov 2026
8 *+10.613.116.967.4%−12.6 pp0.86
NetherlandsLong
13 May 2026 to 6 Nov 2026
177+5.814.626.685.1%+5.1 pp1.23
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
29+3.88.914.387.1%+7.1 pp1.40
BelgiumLong
13 May 2026 to 6 Nov 2026
177+7.516.329.484.2%+4.2 pp1.20
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
29+5.411.518.783.8%+3.8 pp1.19
Denmark East (DK2)Long
13 May 2026 to 6 Nov 2026
177+9.417.332.085.5%+5.5 pp1.16
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
29+5.513.319.979.4%−0.6 pp1.15
Great BritainLong
13 May 2026 to 6 Nov 2026
177+4.29.314.394.0%+14.0 pp1.76
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
29+0.94.57.797.2%+17.2 pp1.79
FranceLong
13 May 2026 to 6 Nov 2026
177+5.08.113.483.4%+3.4 pp1.42
Trailing 30 days
8 Oct 2026 to 6 Nov 2026
29+5.37.611.490.9%+10.9 pp1.42
  • Bias is the forecast minus the reference, so a plus sign means the forecast ran high. “Against 80%” is the band coverage minus 80, in percentage points.
  • Each figure pools the four runs started the day before (00, 06, 12 and 18 UTC) scored on the next UTC day, so leads run from 6 to 42 hours for wind and temperature and from 6 to 45 hours for solar. The scorecard does not split them by lead time.
  • The forecast is the ensemble mean of the zone mean over the 0.25 degree cells inside each zone, and the band is the zone mean's 10th and 90th percentile across the 50 perturbed members.
  • Wind and temperature are compared at 00, 06, 12 and 18 UTC with the zone mean at step 0 of the run started at that hour, the starting state ECMWF builds from its analysis. Solar radiation is compared in 3-hour slots with the same slot from the newest run started at or before it; slots where both sides are below 5 W/m2 are dropped as night.

Source: ECMWF open data, IFS ensemble, CC BY 4.0, as computed by Voltstack (Voltstack computation), as Voltstack's scorecard stood at 12:00 UTC on 6 Nov 2026. Full credit and the changes Voltstack made: credits.

03 · Reading ensemble skill

What each measure tells a desk

MeasureWhat it isWhat it means on a desk
BiasThe average of forecast minus outcome, in the variable's unit.A lean you can subtract. A bias that drifts away from zero over the trailing 30 days is a reason to trust a zone less.
MAEMean absolute error: the size of an ordinary miss.The miss to expect on a normal day, in m/s, W/m2 or °C.
RMSERoot mean square error, which weights large misses more.When RMSE sits well above MAE, the misses come in occasional large ones rather than many small ones.
Band coverageThe share of outcomes inside the members' p10 to p90 range. About 80% when the band is the right width.The measure for risk. Below 80%, a limit or hedge sized on the band is sized on too little uncertainty; above 80%, on too much.
Spread-skill ratioRoot mean square member spread over the RMSE of the ensemble mean. About 1.00 when spread matches error.Below 1.00 the members agree more than the forecast deserves; above 1.00 they disagree more than it misses.

Why band coverage matters more than the average miss

The ordinary day the MAE describes rarely hurts a desk. The day outside the range it planned for does. If the temperature band catches the outcome 54.5% of the time, as it did in France, the band misses far more often than the one day in five its p10 and p90 imply, and a cold tail priced off the members is priced too thin.

A first adjustment

A simple correction scales each member's distance from the ensemble mean by one over the spread-skill ratio. For France temperature, at 0.65, that widens the band by about 1.53 times. It assumes the errors are roughly normal and stable, so check the coverage again after applying it, and use the trailing 30 days to see whether the ratio is moving. In the Netherlands the trailing 30 days to 6 Nov 2026 held temperature 82.7% of the time against 67.3% over the long window, and the GB wind band 96.6%: one season's ratio is not the next one's.

Wind speed is not wind power

The scorecard scores 100 m wind speed. A fleet turns a 1 m/s miss into very different megawatts depending on where it sits on the power curve, so a small speed error near cut-in can be a large generation error. Generation is a separate layer and is not scored here.

04 · Do it yourself

How to get ECMWF open data and score it

ECMWF publishes a subset of its real-time IFS and AIFS forecasts free of charge under the Creative Commons CC BY 4.0 licence and its Terms of Use, which allow redistribution and commercial use with attribution (ECMWF open data). The ensemble (ENS) is 51 forecasts at about 9 km: one control and 50 members run with slightly altered starting conditions and model physics (ECMWF medium-range forecasts). ECMWF publishes its IFS open data on a 0.25 degree grid and, from 20 October 2026, also on a 0.1 degree grid, about two hours later (ECMWF open data). This scorecard uses the 0.25 degree data.

  • Runs and steps. Four runs a day, at 00, 06, 12 and 18 UTC. The 00 and 12 UTC ensemble runs go to 360 hours (3-hourly to 144, then 6-hourly); the 06 and 18 UTC runs stop at 144 hours.
  • Fields. The parameters behind this scorecard are 100u and 100v (100 m wind components), 2t (2 m temperature) and ssrd (accumulated surface solar radiation downwards), from the perturbed members, stream enfo and type pf.
  • Where. data.ecmwf.int/forecasts, with copies on AWS, Microsoft Azure and Google Cloud. ECMWF limits the portal to between 500 and 2,000 simultaneous connections and suggests the cloud copies when the portal is hard to reach.
  • How long it stays. Only the most recent 12 runs, about two to three days. To score a forecast, store every run as it arrives; ECMWF offers its full archive under its service agreements.

ECMWF's Python client, ecmwf-opendata (pip install ecmwf-opendata), fetches the members of one run:

from ecmwf.opendata import Client client = Client(source="ecmwf") client.retrieve( time=0, stream="enfo", type="pf", step=[24, 27, 30], param=["100u", "100v", "2t", "ssrd"], target="ens-00z.grib2", )

Building a scorecard of your own

  • Average each member over your zone or sites, then keep the ensemble mean, p10 and p90 and the member standard deviation for every step.
  • Pick a reference and say which. ECMWF's analysis is free with the data but shares the model's physics; station or metered data is harder to get and stricter.
  • Pair each forecast step with the reference at the same valid time, never across a change in how the zone is drawn, and count bias, absolute and squared errors and whether the reference fell inside the band.
  • Keep the day count beside every number and do not act on a window of a few days. This scorecard waits for 10.
  • Credit ECMWF in the form its Terms of Use give: a copyright statement, the source, the licence, the disclaimer and what you changed (ECMWF Terms of Use for open data).
05 · ECMWF and GFS

ECMWF against GFS: what can be said from public sources

This page scores the ECMWF ensemble only, against ECMWF's own reference, so it cannot say which model is more skilful for these zones. ECMWF publishes headline scores for its forecasts, verified against analyses and station observations (ECMWF forecast quality).

  • NOAA's Global Ensemble Forecast System (GEFS) runs four times a day with forecasts out to 16 days (NOAA GEFS on AWS).
  • NOAA's NCEI describes the 0.25 degree GFS on NOAA's open data programme as a trailing 30-day window (NCEI, GFS). ECMWF's open data keeps about two to three days, so a like-for-like test of the two starts with storing both.
  • Google's WeatherNext models are a separate family again. This page publishes no WeatherNext figures and does not compare it.
06 · What needs the platform

What this page cannot give you

This page is one read, frozen at 12:00 UTC on 6 Nov 2026. The current figures are recomputed with every ECMWF run Voltstack stores and served inside the signed-in Voltstack product and to API keys licensed for them.

  • The scorecard, kept current. The Forecast Skill Scorecard is rescored after each run Voltstack stores (00, 06, 12 and 18 UTC), so the trailing 30 days move forward every day and the long window keeps growing. Same seven zones, three variables and two windows, with the day count beside every number and a CSV export that carries the ECMWF credit.
  • Run-to-run change. The Forecast Risk Monitor ranks tomorrow's run-to-run revision and member spread for Germany and Denmark West against same-hour runs. Weather Run Change sets the newest run against the previous one or the run 24 hours older for any of the seven zones, gas day by gas day out to D+5, in °C and heating degree days, and ranks each change against 30 days of same-hour runs.
  • In the app, and to licensed API keys. These views are inside the signed-in Voltstack product, in the app and its CSV export. Outside the product, the scorecard and the forecast risk figures are served only to Voltstack API keys licensed for the ECMWF weather bundle, with ECMWF's credit.

The platform does not offer alerts on these figures, and neither it nor this page splits skill by lead time.

07 · Questions

Frequently asked questions

How accurate is the ECMWF ensemble forecast for wind, solar and temperature?

In Voltstack's scorecard as of 6 Nov 2026, scored on the next UTC day for seven European bidding zones from 13 May 2026 (to 28 Oct for Germany and Denmark West), the ensemble mean was close: mean absolute error on the zone-mean 2 m temperature was 0.25 to 0.41 °C, and every zone's temperature bias sat between −0.05 and +0.03 °C. The band was the weak part. The p10 to p90 range caught temperature 54.5% to 67.6% of the time against the 80% it should, and 100 m wind 80.5% to 94.7% of the time. Wind and temperature are scored against ECMWF's own analysis, solar against its later short-range runs. This is a dated snapshot, not a live feed.

Is the ECMWF ensemble overconfident?

For 2 m temperature it has been, and for 100 m wind it has been the opposite. From 13 May 2026, the p10 to p90 band caught next-day zone-mean temperature only 54.5% of the time in France and 67.6% in Belgium, the best of seven zones, against a target of 80%; its spread-skill ratio ran 0.65 to 0.80. For wind the band caught 80.5% to 94.7%, with spread-skill ratios of 1.08 to 1.70, so the members disagreed by more than the forecast missed. Germany and Denmark West are measured to 28 Oct 2026, the others to 6 Nov 2026.

What is a spread-skill ratio?

The root mean square spread of the ensemble members divided by the root mean square error of the ensemble mean. Near 1, the members disagree about as much as the forecast actually misses. Below 1 the ensemble is too confident and its band too narrow; above 1 it is too cautious and its band too wide. It reads together with band coverage: a well-sized p10 to p90 band holds the outcome about 80% of the time.

Where can I get ECMWF open data?

From ECMWF's open data service at https://data.ecmwf.int/forecasts/, mirrored on AWS, Microsoft Azure and Google Cloud, or with ECMWF's ecmwf-opendata Python client. The IFS ensemble is published four times a day (00, 06, 12 and 18 UTC) on a 0.25 degree grid under CC BY 4.0 plus ECMWF's Terms of Use, which ask for a copyright statement, the source, the licence, a disclaimer and a note of any changes. ECMWF has also published its IFS open data at 0.1 degree since 20 October 2026. Only the most recent 12 runs are kept, so to score the forecast you have to store each run yourself.

Is ECMWF better than GFS?

This page does not score GFS, so it cannot rank the two for these zones. ECMWF publishes its own headline scores at https://www.ecmwf.int/en/forecasts/quality-our-forecasts. A practical difference for anyone building their own record: NOAA's NCEI describes the 0.25 degree GFS on NOAA's open data programme as a trailing 30-day window, while ECMWF's open data keeps only its most recent 12 runs, about two to three days.

Are these figures live?

No. They are a snapshot, as Voltstack's scorecard stood at 12:00 UTC on 6 Nov 2026, through the 00 UTC run of 6 Nov, and the page does not update. Inside the Voltstack platform the scorecard is rescored after every ECMWF run Voltstack stores.

Voltstack Analytics

Know how far to trust this morning's run

The figures above are a read from 6 Nov 2026. In Voltstack the scorecard, the run-to-run revision and the run change by gas day sit side by side and refresh with every ECMWF run Voltstack stores.

Start a 14-day free trialRequest early access →

The button opens a 14-day free trial sign-up. Sign-up needs a work email.

The button opens the early-access request on our home page.

Credits

ECMWF credit, licence and the changes Voltstack made

Every figure on this page is a Voltstack computation from ECMWF open data, credited with the six lines Voltstack's platform prints beside the scorecard, in the same order. The copyright statement, the licence statement and the disclaimer are quoted word for word from ECMWF's Terms of Use for open data (apps.ecmwf.int/datasets/licences/general, read on 15 Nov 2026). ECMWF writes the source line without the colon, and the two notes of changes are Voltstack's own wording.

COPYRIGHT STATEMENT
This service is based on data and products of the European Centre for Medium-Range Weather Forecasts (ECMWF)
SOURCE
Source: www.ecmwf.int
LICENCE STATEMENT
This ECMWF data is published under a Creative Commons Attribution 4.0 International (CC BY 4.0). https://creativecommons.org/licenses/by/4.0/
DISCLAIMER
ECMWF does not accept any liability whatsoever for any error or omission in the data, their availability, or for any loss or damage arising from their use.
CHANGES MADE BY VOLTSTACK
Voltstack has modified the data: zone means of the 0.25 degree cells, 100 m wind speed from the u and v components, GHI de-accumulated from ssrd and floored at zero, temperature in degrees C, and statistics across the 50 perturbed members of the IFS ensemble.
Voltstack has further modified the data for forecast skill: it compared each run's next-day zone means with ECMWF's own analysis and, for solar radiation, with ECMWF's later short-range forecasts, and summarised the differences.
Related Voltstack Intelligence

More reference pages

REFERENCE
Point-in-Time Data: Gas and Power Data Corrections
Why a record of what each series said, and when, matters for backtests and morning checks.
BUILDER GUIDE
ENTSO-E Transparency Platform API Guide
A security token, working requests for load, generation and outages, and the traps we met.
RATINGS · 2027 EDITION
Best European Energy Data APIs (2027)
APIs rated for builders, weather data included, with a licence column for AI use.

Disclaimer: This reference is produced by Voltstack Intelligence for information only and is not investment advice. The scorecard describes how the ECMWF ensemble compared with ECMWF's own later view over the windows shown, as read at 12:00 UTC on 6 Nov 2026; it says nothing about any other model or period. Corrections to research@voltstack.energy.

VOLTSTACK INTELLIGENCE · REFERENCE · SCORECARD AS OF 6 NOV 2026 · NOT LIVEvoltstack.energy