Settings
API keys, model configuration, notifications
API Keys
ANTHROPIC_API_KEYClaude Sonnet — LLM scoring for all three strategies. 200K context window handles full 10-K + earnings call in one pass.
ALPACA_API_KEYPrice data (OHLCV, adjusted closes, real-time quotes) for Technical and Entropy strategies.
ALPACA_SECRET_KEYAlpaca Markets secret. Use paper API URL for testing.
ALPHA_VANTAGE_API_KEYQuarterly income statements for Fundamental strategy only (revenue, operating income, net income, margins).
Model Configuration — ML Weights
Locked from Table 1, Cohen et al. (2025) — do not modify without explicit reason
| Strategy | Frequency | ML Weight (w) | LLM Weight | Sharpe |
|---|---|---|---|---|
| technical | monthly | 1.00 | 0.00 | 0.6934 |
| entropy | monthly | 0.70 | 0.30 | 0.4207 |
| fundamental | monthly | 0.15 | 0.85 | 0.5001 |
| technical | quarterly | 0.45 | 0.55 | 1.2967 |
| entropy | quarterly | 0.40 | 0.60 | 0.6048 |
| fundamental | quarterly | 0.00 | 1.00 | 0.4899 |
LLM Fallback Behavior
If the Claude API fails, the system automatically falls back to w=1.0 (pure ML) for all strategies. A warning banner is shown in the UI and all score rows are logged with llm_provider="none". Restore the API key and re-run scores to include semantic scoring.
Data Source Failure Isolation
What happens when each source is unavailable
Technical + Entropy blocked
Fundamental can still run
Fundamental blocked
Technical + Entropy can still run
No filing context for Claude
All strategies fall back to w=1.0 (pure ML)
No LLM semantic scores
All strategies use w=1.0; warning banner shown