54.5% win rate at a profit factor of 2.32 means winning trades average almost twice the dollar value of losers — the system is designed for asymmetric payoffs, not high accuracy. And the 7.7× return-to-drawdown ratio is the number that matters most: it means for every dollar the account drew down at its worst, the system generated nearly eight in total return.

There's a meaningful long/short asymmetry: short trades win 58.4% of the time vs 49.3% on longs, and produce nearly 2× the P&L. That's because markets fall faster than they climb — a behavior the engine's risk model is built around.

Strategy specifics are kept private — but the results, architecture, and engineering decisions behind the system are openly shared below.

• Reconnection logic. WebSocket drops happen. The bot detects them, backs off with exponential retry, and resumes exactly where it left off — without losing track of open positions.
• Auto-flatten before close. Holding futures overnight isn't allowed on prop firm evaluations. A hard cutoff guarantees all positions close before the 3 PM CT bell — no exceptions, no surprises.
• Structured logging, every event. State transitions, orders, fills, errors — all logged as JSON. When something weird happens, I can replay the exact sequence of decisions that led to it.
• Risk manager as a separate module. Daily loss limits, per-trade limits, and max trade count are enforced by a dedicated class — not sprinkled throughout the code where they can be accidentally bypassed.
• Mobile dashboard. A Flask + WebSocket app pushes live state to a phone-friendly UI: start/stop buttons, equity curve, current state, last trade. I can check in during errands without needing a laptop.
• Auto-start on boot. A launchd plist plus a cron job ensures the bot is running by market open every weekday — even if the machine has rebooted overnight.