Two products, one set of rules
OpenWindCast publishes two kinds of day-ahead forecast. For wind zones it forecasts onshore wind output and scores it against the grid operator's own forecast. For price markets it forecasts the day-ahead electricity price for every hour of the next day and scores it against the price the auction actually clears at. The two are separate products with separate records, and the same three rules hold for both: the target is fixed in advance, every published day is scored, and nothing published is ever rewritten.
Wind: what is forecast
One value per zone per day: onshore wind output at 11:00 local time on the following day, expressed as a load factor, which is generation divided by installed capacity. The forecast is made on the day before delivery, for the day-ahead market, and the operator forecast it is scored against is the one published for that same day.
Load factor rather than megawatts, because fleets grow. A record kept in megawatts would appear to improve on its own every time a wind farm is connected.
Wind: what it is scored against
The benchmark is the transmission system operator's own published day-ahead wind forecast for the same zone and the same 11:00 target: Elia, Energinet, Fingrid, the German transmission system operators, Red Eléctrica, PSE, REN, NESO, EirGrid and SONI, Terna, APG and IPTO. It is the bar to be beaten, and it is never an input.
The benchmark is frozen on first capture. Operators can and do return revised values for a past hour on a later query, and if those revisions were allowed through, the bar would move after the fact and every comparison built on it would be worthless. Metered output is the one exception, because realtime estimates are legitimately replaced by settled values. That is the outcome being measured, not the bar.
The published forecast record is write-once for the same reason. A published forecast is never rewritten after the fact.
Wind: how accuracy is measured
The headline metric is nMAE, the mean absolute error as a share of installed capacity. One sample per day, on the 11:00 target, for both forecasts, over the same set of days. The share of days on which the forecast lands closer to the metered output than the operator's forecast is reported beside it as the win rate.
Resolution is not the same thing as horizon. Averaging error across all 96 quarter-hours of a day produces a considerably smaller and much more flattering number than measuring the single day-ahead value a market participant has to commit to. Only the latter is reported here.
Where a difference between the two forecasts is too small to be meaningful, it is described that way rather than presented as a win.
Live record and back-test
Every headline accuracy figure on this site is measured on the live record only: forecasts that were published for real, one day at a time, from the date each zone went live. A back-test, which is the same method run over earlier days, is useful context and is shown on each zone page, but it is always shown apart and never mixed into a headline number.
Price: what is forecast
For each price market, one forecast per day covering every hour of the next delivery day, in euro per megawatt hour. The hours are the CET auction day, because that is the day the market trades. Each forecast is made the day before delivery and uses only information that was available before the day-ahead auction closes at 12:00 CET.
Price: what it is scored against
The outcome is the price the day-ahead auction actually clears at, hour by hour. The auction clears once and its price is final, so each day is scored the same afternoon the auction result is published, and it stays in the record for good, including on the days the forecast was wrong.
No grid operator publishes a day-ahead price forecast with an error record, so there is no operator bar to beat. Accuracy is therefore also reported against a naive benchmark: a simple rule that forecasts tomorrow's prices from recent prices alone, such as repeating yesterday's or last week's. Each market is compared with the strongest such rule for that market, measured on its own price history, on exactly the same hours.
Price: how accuracy is measured
The headline metric is the mean absolute error in euro per megawatt hour: the average size of the hourly miss, whichever direction it went. Every hour counts, including hours that clear at zero or below zero. The comparison with the naive benchmark is reported as how much lower the forecast's error is than the benchmark's on the same hours.
A price record takes time to become a reliable estimate. The number of scored days is always shown next to every price accuracy figure, and while a record is young those figures are provisional.
Euro errors are not comparable across markets, because some markets are far more volatile than others. Read each market against its own benchmark rather than against the others.
The uncertainty band
An 80% interval is published alongside each wind forecast and is calibrated to cover approximately what it claims to cover. It is context for the point forecast rather than the product. What is measured, published and scored here is the point forecast.
Where to see it applied
Every zone page publishes its full daily record under these rules. Start from the zone list, or compare every operator on the TSO forecast accuracy page. Terms are defined in the glossary.