This page runs the same engine the bot trades — a JavaScript port of the Python engine, pinned to it trade-for-trade by an automated parity check. It fetches public 1-minute klines straight from Binance (no account, no key; nothing leaves your browser), aggregates them, and simulates the v2 strategy with the parameters you choose: conviction-score entries with hysteresis, volatility-target sizing, fees and slippage on every fill, walk-forward out-of-sample validation on demand.
90 days ≈ 130 requests to Binance's public API (~1 min). If Binance is unreachable from your region, upload a CSV made by scripts/download_data.py instead.
Walk-forward re-optimises a small grid (theta_in × trail × ER min) on rolling 30-day windows, trades the top-3 combos on the following 10 days, and reports only out-of-sample results. That number — not the fixed backtest — is the honest one.
| Test window | IS Sharpe (top-3) | OOS return | OOS trades |
|---|
| Entry (UTC) | Exit | Held | Reason | Weight | Entry px | Exit px | Net % | Net USDT |
|---|
Simulated results ignore partial fills, order-book depth and latency. A fixed-parameter backtest over one window is an optimistic upper bound — the walk-forward numbers are the honest ones, and even they are statistically inconclusive at these sample sizes (~13 effective independent trades). Educational use only; not financial advice.