Europe's grid increasingly runs on weather. Every fifteen minutes the power market asks the same question: how much will wind and solar produce a few hours from now? Get it wrong and the error settles in cash: imbalance charges, redispatch, reserves called that didn't need to be. Forecasting is not a dashboard nicety here. It is the price signal.
We built Augur, a system that forecasts wind and solar production for Germany, Belgium and Romania, to do the one thing most forecasting vendors quietly avoid: grade itself against the strongest available baseline, in public, including where it loses.
The baseline you beat is the whole story
When someone tells you their forecast is "accurate," the only useful follow-up is: accurate compared to what? Accuracy in a vacuum is a number with no meaning. A forecast is only as good as the baseline it improves on.
The most common baseline in vendor decks is persistence — the assumption that the next hour looks like the last one. Beating persistence is trivial. Weather has structure; any model that learns a little of it clears that bar. So "we beat persistence" is not a result. It is a red flag. It tells you the vendor picked the weakest possible opponent and won.
The honest baseline is the one the market actually uses: the grid operator's own forecast. Each transmission system operator (TSO) publishes its forecast for wind and solar, and updates it continuously as new weather data arrives. That is a strong, professionally maintained, moving target. If you want to claim you forecast the grid well, that is the thing you have to beat.
What Augur does
Augur is an unattended system. It pulls weather and production data, produces probabilistic forecasts, not just a single number but a distribution with calibrated uncertainty, and submits them hourly to Predico, the collaborative forecasting platform run for Elia, Belgium's grid operator. Then it scores every submission it makes against the operator's own published forecast, and records the result whether it won or lost.
That last part is the point. A system that only reports its wins is marketing. A system that reports where it loses, and defers there, is an instrument you can actually trust.
The results, honestly stated
Short-range is where the value concentrates, because that is where imbalance is priced. At fifteen minutes ahead, Augur beats the operator's continuously updated forecast by 40–78%, and reaches parity with it around three hours out. Past that horizon the operator's forecast is excellent and Augur does not claim to beat it — it says so and defers.
Day-ahead, where planning happens: Augur's wind forecast beats every covered operator by 7–11% RMSE at the same cutoff, and its Romania solar day-ahead forecast runs 64% better than Transelectrica's. Its prediction intervals are calibrated: roughly 80% of actual outcomes land inside the 80% interval, which means the uncertainty it reports is the uncertainty you actually get, not decoration.
It has run live since June 2026, hourly, with no missed submission gates. Numbers will move as the system evolves; the discipline of measuring them against the right baseline will not.
What this says about how we build now
Augur was built in days by a small senior team working with AI agents on top of 25 years of real-time energy-systems engineering. That combination is the whole thesis. The AI agents move fast; the energy experience keeps the speed pointed at the right problems, and the honesty discipline keeps the claims defensible.
We spent a decade running energy infrastructure at grid scale, dispatching assets, balancing markets, holding TSO-grade reliability. Augur is what it looks like to put modern AI on top of that foundation instead of starting from a demo. It is unglamorous in exactly the right way: it forecasts, it scores itself, and it tells you where it is wrong.
You can watch it run — including the losses — at augur.eloquentix.com, and read the longer story in our case studies.