Winds pass cut-out speed across a coastal cluster and output collapses within two hours.
- Return period
- 1 in 10 yr
- Peak ramp
- −85 % in 2 h
Winter quarter, tail view
Tempora · Stress-test
Tempora generates rare weather your history does not contain, such as storms, ramps, icing and heat, then runs it through your forecasts and decisions to show the exposure.
Capabilities
Synthetic extremes
Cosmos and Earth-2 world models generate extremes conditioned on each site's climate.
Exposure
Tail risk expressed as imbalance and curtailment cost, not only as weather statistics.
Stress tests
Run Dispatra, Cellix and Fleetra rules through thousands of extremes.
Reports
Exposure analytics in a form risk teams and insurers can use.
Corpus
Generated extremes enrich the training data for the whole platform.
Early warning
Flags when forecasts start to resemble a stressed scenario.
How it works
World models create extremes conditioned on location and climatology.
Skycast's chain turns each scenario into asset-level power.
Settlement rules turn power swings into imbalance and curtailment cost.
Policies are tested and adjusted where the tails hurt most.
Scenario library
Filter by event type. Each card is an illustrative scenario of the kind Tempora generates for a site.
Showing 8 of 8 example scenarios
Winds pass cut-out speed across a coastal cluster and output collapses within two hours.
A cold front crosses a wind cluster earlier than the day-ahead forecast expected.
Freezing fog ices blades on a hill site for most of a day.
High module temperatures and inverter derating cut midday output during a price spike.
A cold, still, overcast week with low wind and solar across a region.
Fast-moving thunderstorms cause sharp swings across a solar and wind mix.
Morning fog clears later than forecast and the solar ramp arrives two hours late.
Freezing rain triggers ice-detection stops across a whole site.
Example scenarios with illustrative values. A real library is generated per asset and climatology.
Specification
| Item | Detail |
|---|---|
| Inputs | Site climatology, Skycast models, settlement rules, decision policies |
| Outputs | Scenario libraries, exposure curves, stress-test reports |
| Generation | Cosmos and Earth-2 (CorrDiff) on DGX or HGX; GB200 at scale |
| Validation | Physical plausibility checks; fully validated extremes are a later goal |
| Statistics | Extreme-value analysis on RAPIDS |
| Delivery | Reports, API and Parquet exports |
Part of the loop
Downscales ensemble AI weather to each asset and outputs calibrated P10, P50 and P90 power.
02 · DecideDispatraTurns those quantiles into bids and setpoints inside your limits, with a reason for each.
03 · StoreCellixSchedules batteries for value and state of health together.
04 · BalanceFleetraNets imbalance and plans the whole portfolio against grid constraints.
05 · See and actTwinoraShows decisions on a live twin and executes them at the edge.
06 · Stress-testTemporaGenerates extreme-weather scenarios to price and harden the tails.
Questions
They are generated to be physically plausible for each location and checked against historical extremes. We state the validation for every library and do not present them as observed events.
Reports are designed to be shared with insurers and risk teams. They describe exposure; they are not an insurance product or an actuarial opinion.
No. Tempora is an add-on. It is most useful once Skycast and Dispatra are running on your sites.
Thousands per site and season, so tail estimates rest on more than a handful of events.
We start with a backtest on your history and show the result side by side with your current approach.