Model
AI-Chosen Signal
An LLM-designed baseline trained on gasoline futures volatility and short-term returns.
Decision surface
The model's buy probability across its two most influential features. Every point was computed by calling model.predict_proba on the actual .joblib. Drag to rotate.
What the model pays attention to
Approach
Given the team-member brief, the LLM chose RBOB gasoline futures (RB=F) from Yahoo Finance as its data source — its stated reasoning was that gasoline futures are the most direct upstream proxy for retail gas prices with a publicly accessible API. It engineered four features: the latest closing price, the 5-day log return, the 20-day realized volatility, and the ratio of current volume to the 30-day average. It chose a regularized logistic regression for interpretability and to avoid overfitting on a small feature set, and noted that adding more features did not help in cross-validation. The resulting model is intentionally simple — the LLM's own writeup says it 'wanted a disciplined baseline to compare the human-built models against.'
Features
Built by
Built by an LLM (Claude, Anthropic) following the exact same team-member brief as the human contributors. The model picked its own data source, engineered its own features, and trained its own scikit-learn classifier — delivering the same four files as any other teammate. Included to compare human-chosen vs. AI-chosen approaches to the same problem.
Trained on
2023-02-14 to 2026-04-14