A forecast that lets itself be measured.

Forecast quality is usually reported as a percentage error. For anyone exposed to imbalance settlement, that number does not say what matters: it does not indicate what the error cost, in which periods, or what would have happened with a different forecast.

The result is a market where people buy for years without knowing whether the supplier is any good.

• Probability distribution — not just the central value, because it is the uncertainty that sizes the risk
• Published calibration — periodic verification that the P90 is exceeded as often as it should be, by horizon and by regime
• The error in euros — the settlement cost associated with the month’s error, not a percentage
• Decomposition — how much of the imbalance was weather, how much was model, how much was dispatch

Calibration is published in the bad months too. That is what makes it a verification rather than an assertion.

A forecast supplier delivers the number and never learns what it cost. We operate assets in the market: we see the settlement, period by period, and can tie the error to its real cost.

There is also a technical difference that cannot be replicated from outside: an external forecaster cannot tell a cloud from an instructed curtailment. It trains on series in which dispatch periods look like drops in output — and the bias stays in the model. Whoever steers the asset knows the difference.

The first piece of work we propose is not selling a forecast: it is processing the asset’s history, separating the weather effect from the dispatch effect and returning a clean series.

It is cheap to do, it is verifiable, and it serves those who stay with us as much as those who do not.

The service is aimed at asset owners who nominate themselves or through third parties, at consumers with material exposure, and at anyone wanting an independent check on their current supplier’s performance.

We process the plant’s history, separating the weather effect from the dispatch effect, and quantify the result.