Our fleet against its P50, year by year

Eighty-four assets, fourteen years, no smoothing. Two of them have never made their case and we explain why.

Most published fleet performance is presented as a rolling average against a rebased expectation. Both of those choices flatter the result. A rolling average hides the years that hurt, and rebasing an expectation after the asset underperforms turns a miss into a match. We have done neither here. Every year is against the P50 the asset was underwritten on, at the time it was underwritten.

The headline

• Across the fleet and the full period, actual generation is 97.4% of the original P50.
• Nine of fourteen years came in below P50. Five came in above.
• The dispersion is wider than the average suggests: individual asset-years range from 71% to 118%.
• Two assets have never reached their P50 in any year of operation.

A fleet at 97.4% of P50 is a fleet that is, on the whole, working. It is not a fleet that hit its case. A P50 is by definition the level you should exceed half the time, and we have exceeded it in five years out of fourteen. That gap is small, persistent, and worth more attention than the headline.

Where the 2.6 points went

We can account for most of it and not all of it.

• Curtailment, 1.1 points. Not in the original cases at the levels that arrived, and concentrated in four assets on two nodes.
• Availability against contracted, 0.6 points. Two long outages, one gearbox campaign, and one substation fault we caused ourselves.
• Wind resource against long-term, 0.4 points. The period contains two of the lowest wind years in the reference series for our main markets.
• Unexplained, 0.5 points. We have not closed this and we are not going to invent a category for it.

Half a point of a fourteen-year record is unexplained. We would rather publish that line than distribute it across the other four until it disappears.

The two that have never made it

Asset one is a 42 MW onshore project energised in 2016. It has run between 78% and 89% of P50 every year. The wind resource is close to the long-term series; the machines are available; nothing is broken. The P50 was wrong. The assessment used a two-year on-site campaign correlated to a reference station 61 km away, across a ridge, and the correlation did not hold. We have re-derived the P50 with eight years of operating data and it is 14% below the original. The asset performs against the new number.

Asset two is a 28 MW project where the resource assessment was sound and the constraint regime was not. It sits behind a node that has been constrained every year since 2019, at levels between 9% and 17%. The original case assumed 2%. This one is not a measurement failure; it is a failure to price a network risk that was, in hindsight, visible in the network operator's own ten-year statement.

What we changed

• Every P50 we now underwrite carries a stated correlation quality and a stated distance to reference. Below a threshold we either extend the campaign or discount the number explicitly.
• We re-derive P50 from operating data at year five and publish the revision internally against the original, so that a wrong assessment is visible as a wrong assessment rather than as an underperforming asset.
• Constraint went from an assumption in the yield model to a separate line with its own evidence base.

Why we publish it in full

Because a fleet record with the bad years removed is not evidence about anything, and because the two assets that never made their case are the two that taught us most. The first cost us a resource methodology. The second cost us the belief that a network operator's published constraint outlook is a floor.

The full series, asset by asset and year by year, is available on request under NDA. It is the same file we give our own investment committee, with nothing taken out.