Model documentation
Atlas & Falcon: model design and evaluation
Compare the stated design intent of Wealthoak’s two engines and inspect the model documentation needed to evaluate their behaviour, risk and limitations.
| Characteristic | Atlas | Falcon |
|---|---|---|
| Approach | Patient trend following | Opportunistic momentum |
| Intended pace | Lower turnover | Faster decision cadence |
| Described horizon | Weeks to months | Days to weeks |
| Measured risk and return | No verified report published here | No verified report published here |
Automation is not a description of the model
Automation executes a process without requiring a person to repeat each step. A rules-based strategy can be automated without machine learning. A machine-learning model requires its own explanation of inputs, training, validation and inference. The label “AI” alone does not establish either predictive quality or a particular implementation.
This library does not publish a neural-network architecture, training dataset or independent AI capability assessment for Atlas or Falcon. Evaluate each engine by its versioned documentation and evidence, not by the label.
What a model card must disclose
- Identity: version, release date, responsible publisher and change history.
- Scope: instruments, exchanges, sessions, required data and minimum history.
- Decision process: signal inputs, decision timing, sizing assumptions and any machine-learning component.
- Evaluation: selection procedure, held-out data, realistic costs, stability tests and sensitivity to missing observations.
- Operating limits: losing regimes, liquidity limits, gaps, leverage, correlated exposures and dependence on broker availability.
- Controls: which limits are configured, tested and actually enabled for the account.
Read characteristics as hypotheses
A slower strategy can still suffer sudden losses. A momentum strategy can experience repeated losses in a choppy market. Mixing engines does not establish diversification unless their joint behaviour has been measured using compatible return series.
Read the testing protocol and download the model-card template. Confirm the exact engine version and enabled controls before using execution tools.