RESEARCH SYSTEM / 03

Research System

The core design keeps each layer responsible for one class of question instead of placing every decision in one script.

ARCHITECTURE

Six layers with explicit contracts

Data

Membership, OHLCV, suspension, ST, listing dates, and cache

Research

Factors, forward returns, IC, grouped returns, and metrics

Signals

Ranking, selection, rebalance cadence, lag, and weight_df

Execution

RQAlpha accounts, orders, costs, slippage, and retries

Evidence

Configuration, NAV, positions, trades, summaries, and release notes

Presentation

README files, update reports, and this bilingual portfolio

REPOSITORY MAP

How the code maps to the workflow

data/ + cache/
research/
  momentum_20d_forward_5d.py
  download_hs300_dynamic_20y.py

src/archer_factor_analyzer/
  calculate_factors.py
  calculate_returns.py
  signal.py
  backtest/
    run_backtest_rqalpha.py
    rqalpha_weight_df_executor.py
    rqalpha_result_saver.py

result/ + docs/ + website/

Research ≠ execution

Factor analysis tests information content; the executor determines whether orders can fill.

Target ≠ position

weight_df expresses intent; account positions record execution reality.

Traceable data boundaries

Eligibility conditions are stored separately before the final mask is composed.

Evidence over screenshots

Configuration and structured outputs are preserved with every formal run.