Description
Summary I'm building a Binance BTC/USDT spot bot that reads live order flow — executed tape + top-of-book depth — and trades classic microstructure patterns (absorption, sweeps, book/tape confluence, spoof detection). I have a working Python prototype and need someone experienced to harden it. Non-negotiable: I was burned before by a TradingView backtest that looked great and lost money live because it ignored spread, slippage, and partial fills. Realistic execution modelling — walking the recorded book to compute real fills, charging real spread and fees — is the whole point. Scope: WebSocket ingestion (aggTrade + depth20) with reconnect handling Deterministic order-flow signal engine (CVD, imbalance, walls, voids, spoofs) — no LLM in the hot path Recorder + replay simulator with book-walking fills, post-cost P&L, hit rate by signal type Spot execution with hard-coded risk caps + drawdown kill switch Two-speed architecture: fast reflex loop for stops/targets, slower cadence for signals Structured journal for every signal and fill Optional: Claude API off the hot path for post-hoc trade review You: Strong Python, prior order-flow work, know why aggressor tagging and iceberg orders matter, insist on realistic backtesting. Bonus: Binance API experience. In your application, include: An order-flow strategy you've built and its main live failure mode How you'd model slippage in a crypto backtest Code samples (redacted fine) Estimate and any scope suggestions