Tanaka-san — USD/JPY Signal Desk (Python)

Tanaka-san is the USD/JPY desk from the Synthetic Capital / Charting the Rubicon AI Agents Fund — isolated into a single standalone Python script you can run yourself. It reproduces the exact decision the fund's Tanaka-san desk makes.

What it does

Tanaka-san runs a disciplined RSI(2) pullback with a trend filter on dollar-yen: he only goes long when the trend is up (price above its 200-period average) and RSI(2) has pulled back to a deep oversold reading, with a secondary "panic" entry when price breaks the lower Bollinger Band. Each time you run it, the script reads real USD/JPY candles and prints the desk's current call — LONG, SHORT, or FLAT — with entry price, an ATR-based stop, and a 0–1 conviction score, as of the last completed candle.

Signal only. The script never places an order and never touches your account. What you do with the signal is entirely up to you.

What's included

• tanaka_san_jpy.py — the self-contained desk script (only dependencies: requests, pandas, numpy)
• SETUP_GUIDE — a plain-English, step-by-step setup guide written for non-coders

Requirements

• Python 3 and pip install requests pandas numpy
• Your own free OANDA practice API token (read-only pricing)

Note: the fund runs this desk at 3× notional; the script's signal itself is leverage-independent (you choose your own sizing).

Historical backtest — zero-fee paper model on real prices

• Archetype: RSI(2) pullback · 15-minute · 540-day OANDA history
• Net return +13.4% (1×) · Sharpe 2.57 · Max drawdown −2.4% · Win rate 66.1% · 3 of 4 positive walk-forward folds

Important: These figures are a zero-fee, zero-slippage paper model on historical data. They are not a forecast and not a promise of profit — live spread, commissions, and slippage will reduce results. This is a research tool, not financial advice. Past performance does not indicate future results, and you are responsible for any trading decisions you make.