ZEPHYRTRADE / METHODOLOGY
What the numbers mean
A browser research desk for inspecting synthetic wind-market assumptions and hypothetical energy offers.
Data provenance and scope
The supplied research pipeline generates synthetic wind, weather and price tables. The browser ships a validated observation snapshot rather than a live data feed. It contains 3,888 retained hourly test observations, nine strategies and one 30 MW synthetic Ronne Coastal Proxy site. The snapshot covers 20 July–31 December 2024 UTC.
There are 53 absent hours within that test span. The original pipeline removed 319 negative-price hours across its full generated source period. The snapshot is therefore not an unfiltered calendar backtest. Supporting source-file hashes and provenance are available in the app’s Method & evidence view and public repository.
Forecasts, offers and revenue
The application compares forecast-based and direct-offer strategies. Overview presents the supplied backtest; Model arena compares the nine strategies; Data explorer makes retained observations inspectable. Strategy lab calculates a hypothetical offer from editable scenario assumptions using a weighted-quantile method.
The portable browser engine runs locally in JavaScript. An optional local Python service adds a separate SciPy linear-program cross-check. The public website does not provide that Python service, and changing filters does not train a new model.
Interpreting comparisons
Perfect foresight knows actual production and serves as an analytical benchmark, not an operational forecast. Forecast issuance, gate-closure availability, current market rules, transaction costs and real settlement data are not validated. Synthetic capture percentages do not establish live trading performance.
The scenario calculator permits negative prices under its stated price-order assumptions. That does not restore observations excluded from the historical snapshot. Read the visible units and assumptions before comparing results.
Reproducibility
The quality report records verification and limits; CI runs show source checks. The browser export records its source revision and runtime SHA-256 in app.json. Passing software checks is not validation of a real trading service.
Save and export
Named scenarios use browser-local storage. Export JSON backups before clearing browser data or moving devices. Storage is specific to this origin; scenarios saved on MKLab or another host are not automatically transferred here. CSV exports support inspection of the bundled data.