Architecture

Public Entry Point

The public dry run does not open a database or download data:

python scripts/run_pipeline.py --list-stages

For users with equivalent, legally obtained data, the ordered command map is:

python src/init_database.py --db data/db/huatai_quant.db
python src/replicate_tradable_universe.py --skip-pull --raw-dir data/raw --processed-dir data/processed
python src/construct_term_structure_factors.py --db data/db/huatai_quant.db
python src/construct_momentum_factors.py --db data/db/huatai_quant.db
python src/construct_inventory_factors.py --db data/db/huatai_quant.db --write-db
python src/construct_position_factors.py --db data/db/huatai_quant.db
python src/reproduce_figures_8_10.py --db data/db/huatai_quant.db
python src/reproduce_figures_56_58.py --db data/db/huatai_quant.db
python src/reproduce_figures_59_63.py --db data/db/huatai_quant.db

The remaining factor families and report figures use the matching scripts in src/. Run scripts/run_pipeline.py --list-stages for the complete stage-to-script map.

Data Flow

Licensed futures / spot / holdings inputs
                 |
                 v
          SQLite contract tables
                 |
                 v
  Dynamic tradable universe (daily, Friday snapshots)
                 |
                 v
 Term structure | momentum | inventory | position | volatility factors
                 |
                 v
 Friday cross-sectional quintiles and next-period daily returns
                 |
                 v
  IC diagnostics, industry-neutral checks, robustness, multi-factor portfolios
                 |
                 v
    Aggregate public evidence, supplementary report, static site

Boundaries

data/ is intentionally empty except for its contract documentation. Scripts default to a local SQLite path but no database is committed. The results bundle contains only portfolio-level summaries, tables, and author-generated figures; it contains no contract-level observations or licensed research text.