How Atlas Core Generates Forecasts
From raw data to calibrated, transparent predictions in five steps
Atlas Core is the forecasting engine that powers all Atlas products. It transforms structured market and macro data into calibrated, explainable forecasts with full transparency into what's driving each output.
The Forecasting Pipeline
Data Ingestion
Atlas Core ingests price data, macro indicators, sentiment metrics, and other features from reputable third-party providers. Pipelines run automated quality checks and normalisation before downstream processing.(Data sources and availability vary by product and asset class.)
Signal Engineering
Derived features—momentum measures, volatility signals, economic regime classifications, and more—are computed using systematic, rule-based, and statistical transformations. Every transformation is versioned for reproducibility.
Ensemble Forecasting
Multiple model families are trained and blended using walk-forward evaluation. Model weights adjust over time to reflect evolving conditions, subject to constraints that help prevent overfitting.
Confidence & Drivers
Each forecast is accompanied by a calibrated confidence indicator and a 'driver view'—providing transparency into which factors (momentum, macro regime, sentiment, etc.) are influencing the output.
Delivery
Forecasts are published to the Atlas dashboard and, where applicable, made available via REST API. Standard delivery is end-of-day; other cadences may be agreed for specific use cases.(Scope per Order Form.)
Data Ingestion
Atlas Core ingests price data, macro indicators, sentiment metrics, and other features from reputable third-party providers. Pipelines run automated quality checks and normalisation before downstream processing. (Data sources and availability vary by product and asset class.)
Signal Engineering
Derived features—momentum measures, volatility signals, economic regime classifications, and more—are computed using systematic, rule-based, and statistical transformations. Every transformation is versioned for reproducibility.
Ensemble Forecasting
Multiple model families are trained and blended using walk-forward evaluation. Model weights adjust over time to reflect evolving conditions, subject to constraints that help prevent overfitting.
Confidence & Drivers
Each forecast is accompanied by a calibrated confidence indicator and a 'driver view'—providing transparency into which factors (momentum, macro regime, sentiment, etc.) are influencing the output.
Delivery
Forecasts are published to the Atlas dashboard and, where applicable, made available via REST API. Standard delivery is end-of-day; other cadences may be agreed for specific use cases. (Scope per Order Form.)
How We Validate
Rigorous testing ensures reliable forecasts
Walk-Forward Testing
All models evaluated using walk-forward methodology—training on past data, testing on unseen future data—to simulate real-world forecasting conditions.
Continuous Re-Calibration
Models are retrained on a defined cadence to adapt to changing market conditions while applying regularisation and constraints to reduce overfit risk.
Performance Documentation
Pilot partners receive access to historical accuracy metrics and backtest documentation so they can evaluate fit-for-purpose before production use.
What You See in the Output
Every forecast includes these key metrics
Direction
Expected bias over the horizon
Expected Move
Forecasted change magnitude
Confidence
Calibrated reliability indicator
Model Agreement
When models align or diverge
Key Drivers
What's influencing the forecast
Built on Research, Tested in Markets
Our methodology combines established quantitative techniques with modern machine learning, validated through rigorous walk-forward testing on real market data.
Want to See the Methodology in Action?
Request pilot access to explore forecasts and documentation.
