J-Quants: Start with Real Local Data

Codex, Claude or Cursor can help prepare your local environment. Use the setup prompt and connection guides in Get started to install the unpublished main candidate in a dedicated virtual environment. Validate and freeze your own data in the steps below.

Run these commands in a main source checkout with Python 3.12+. This candidate is unpublished.

python -m pip install -e ".[gui]"
kabuforge doctor
python -m framework_v2.workbench_qt

Daily-bar CSV or Parquet needs date, code, open, high, low, close and volume. J-Quants Date/Code/Open/High/Low/Close/Volume and O/H/L/C/Vo aliases are recognized. Review duplicates and missing values in validation, then freeze the input before research.

In Data Center → Local market cache, select your own CSV, Parquet or completed manifest; review codes, dates and price basis, then validate and freeze. In My Strategies, select Price momentum and the frozen input, run preflight, then backtest. Open the existing five GUI courses · Integration evidence and limits

This is a short contract excerpt. Place the selected manifest, its referenced files and a recipe JSON inside the dedicated workspace; replace paths with your actual files.

kabuforge --locale en_US research --workspace /absolute/path/to/research_workspace price --manifest inputs/manifest.json --recipe inputs/recipe.json

Minimal momentum recipe (cash in JPY; fee in percent):

{
  "count": 2,
  "lookback": 20,
  "cash": 2000000,
  "fee": 0.1,
  "frequency": "monthly",
  "signal_template": "price_momentum"
}

On Windows use C:/path/to/research_workspace. No market bars are bundled. GUI connection tests/downloads are separate explicit actions; the reference course does not verify live requests. Results remain RESEARCH-ONLY with pit_guarantee=false.

Inspect the output directory, receipts and logs returned by the CLI. In the GUI, use Backtest Results and Run History to check the input SHA, date range, fees, fills and skip reasons. Read results with the existing course