Quant Export

What it is

Quant Export is a curated extraction of the reusable core from Jordan's earlier crypto trading bot: roughly 116 Python files across core, engines, execution, models, training, and utils, with no framework glue. The engines layer decomposes the full order lifecycle into small single-purpose modules (order builder, normalizer, validator, sizer, ack tracker) plus explicit market microstructure models for fees, slippage, and latency, and separate realized/unrealized PnL trackers. A Master Constitution dataclass encodes a fixed priority hierarchy that the RiskManagerCore and every other module defer to, with dynamic volatility- and confidence-scaled position sizing, a kill switch, and hard stop-loss checks. It ships trained model artifacts: PPO reinforcement-learning checkpoints (stable-baselines .zip) and per-symbol XGBoost models across seven crypto pairs, produced by an iterative "infinite" retraining loop.

Highlights

• ◆Order lifecycle fully decomposed into ~30 single-responsibility engine modules (builder, normalizer, validator, sizer, ack/confirmation trackers, execution router) rather than one monolithic executor
• ◆Explicit microstructure modeling: separate fee_model, slippage_model, latency_model, and tick-rate controller so backtest and live share the same cost assumptions
• ◆Governance-first design: a Master Constitution priority hierarchy the RiskManagerCore explicitly obeys ('overrides everything except Mac safety'), with 2% per-trade risk and 5% hard stop baked in
• ◆Dual ML approach shipped as artifacts: PPO RL checkpoints (17 training cycles on BTC) plus XGBoost models across 7 pairs (BTC, ETH, SOL, ADA, DOGE, LTC, XRP) from an iterative retrain loop
• ◆~7,400 lines of Python; safety-conscious utils layer with heartbeat, watchdog overseer, anti-exploit, recovery, and health-check modules