How cetirzine/elucidator works
Elucidator is a minimal, auditable A-share news-factor research scaffold, not a completed trading system. Its implemented path covers point-in-time news contracts and optional LLM event extraction; the README and execution plan describe a broader research workflow whose modeling, ranking, backtesting, and paper-trading components are not present in the sampled runtime source. Edges for that broader workflow reflect documented design, not inspected implementations.
Main entry points: llm.py News extractor · EXECUTION_PLAN.zh-CN.md OOF features · contracts.py Event factors · settings.py Runtime settings
Components 17
Inputs and contracts
- Market data
docs/EXECUTION_PLAN.zh-CN.md - News data
docs/EXECUTION_PLAN.zh-CN.md - Timing contract
src/elucidator/contracts.py - Research config
configs/research.yaml - Runtime settings
src/elucidator/settings.py
News extraction
- News extractor
src/elucidator/llm.py - DeepSeek API
- Event factors
src/elucidator/contracts.py
Model research
- Prophet baseline
docs/EXECUTION_PLAN.zh-CN.md - Shared LSTM
docs/EXECUTION_PLAN.zh-CN.md - OOF features
docs/EXECUTION_PLAN.zh-CN.md - XGBoost ranker
docs/EXECUTION_PLAN.zh-CN.md
Evaluation and controls
- Cost backtest
docs/EXECUTION_PLAN.zh-CN.md - Robustness audit
docs/EXECUTION_PLAN.zh-CN.md - Paper acceptance
docs/EXECUTION_PLAN.zh-CN.md - Environment preflight
src/elucidator/preflight.py
Other components
- Researcher
Connections 17
- Researcher → Research config: configures
README.md - Market data → Timing contract: checks timing
README.md - News data → Timing contract: checks timing
README.md - News data → News extractor: provides articles
README.md - Runtime settings → News extractor: configures
src/elucidator/llm.py - News extractor → DeepSeek API: requests extraction
src/elucidator/llm.py - News extractor → Event factors: validates output
src/elucidator/llm.py - Market data → Prophet baseline: informs baseline
README.md - Market data → Shared LSTM: informs representation
README.md - Prophet baseline → OOF features: contributes predictions
README.md - Shared LSTM → OOF features: contributes predictions
README.md - Event factors → OOF features: joins features
README.md - OOF features → XGBoost ranker: ranks factors
README.md - XGBoost ranker → Cost backtest: evaluates strategy
README.md - Cost backtest → Robustness audit: feeds audit
README.md - Robustness audit → Paper acceptance: supports acceptance
README.md - Researcher → Environment preflight: checks environment
README.md