Pro Trader Features
The Pro Trader is built on 7 years of live trading experience across crypto market microstructure. Every feature exists because of a real loss or a real edge discovered in production. Nothing here is theoretical.
Signal Intelligence
Multi-Exchange Order Flow
The bot ingests real-time order flow from multiple centralized and decentralized venues simultaneously. It tracks buying and selling pressure across spot and perpetual markets, detecting divergences that single-exchange strategies miss entirely. When spot buyers are aggressive but perp sellers are dumping — or vice versa — the bot recognizes the imbalance before price reflects it.
Orderbook Imbalance Detection
Real-time monitoring of bid/ask depth ratios across multiple exchanges. When one side of the book thins out relative to the other, it's an early signal of directional pressure. The bot uses smoothed aggregate imbalance from multiple venues for directional reads, while raw single-exchange imbalance drives fast protective reactions.
Squeeze Detection
The most reliable directional signal in the system. Detects when spot and perpetual order flow diverge in a pattern that historically precedes forced liquidation cascades. When the conditions align, the bot shifts its bias to trade with the squeeze — positioning ahead of the forced buying or selling that follows.
Volatility Forecasting
A statistically validated forecasting model predicts short-term volatility regime changes before they happen. The model was validated through recursive feature elimination over 360+ days of tick data, testing nearly 200 features and narrowing to those with proven predictive power. When high volatility is predicted, spreads widen preemptively. This alone has prevented some of the largest potential drawdowns.
ML Volatility Predictor
A separately trained machine learning model (cross-validated, 72-75% accuracy) predicts whether the next period will be high or low volatility. This provides a second independent volatility signal that works alongside the statistical forecaster — two different approaches confirming the same conclusion increases reliability.
Open Interest Divergence
Monitors the relationship between price movement and open interest changes. Price rising with growing open interest means new money is entering (strong move). Price rising with falling open interest means positions are closing (weak move). The bot amplifies or dampens its directional signals based on whether the underlying move is structurally strong or weak.
Market Structure Patterns
Detects higher-highs/higher-lows (bullish structure) and lower-highs/lower-lows (bearish structure) in real time. These patterns, validated through backtesting, provide a structural framework that filters out noise. The bot won't fight established structure — it trades with it.
Confluence Scoring
Multiple signals are fused into a single confidence-weighted directional score. The system aggregates squeeze state, open interest flow, market structure, orderbook imbalance, value area position, and momentum indicators — weighting each by its statistically validated reliability. High confluence = tight spreads, aggressive positioning. Low confluence = wide spreads, conservative stance.
Momentum State Detection
Leading indicators that fire before traditional signals confirm. Detects the very start of directional moves using flow acceleration metrics — catching trend beginnings while lagging indicators are still neutral. When these early signals align with subsequent confluence confirmation, the win rate is substantially higher.
Value Area Analysis
Integrates multi-timeframe volume profile data (daily, weekly, monthly, and their previous periods). The bot knows where the highest-volume price zones are and treats them as dynamic support/resistance. Positions near value area boundaries get special handling — the bot recognizes when price is at a statistically significant level and adjusts accordingly.
Value Area Rejection Detection
Goes beyond static value area levels by detecting rejection patterns — when price tests a value area boundary multiple times and gets pushed back, confirmed by order flow. This pattern historically precedes reversals, and the bot shifts bias to trade with the rejection.
Market Profile Patterns
Advanced volume profile patterns including poor highs/lows (unfinished auction areas that price tends to revisit), single prints (fast-moving gaps in volume that act as support/resistance), and naked points of control (untested high-volume nodes that attract price).
Risk & Protection
Adaptive Spread Engine
Spreads are never static. Multiple independent risk signals each contribute a multiplier to the base spread. High volatility, toxic flow, rapid price movement, competitive pressure, and anomalous behavior each widen spreads independently — and they compound. The result is spreads that are tight in safe conditions and wide when danger is detected, reacting faster than any human could.
| Condition | Response |
|---|---|
| Predicted volatility spike | Spreads widen before the move hits |
| Toxic/informed flow detected | Spreads widen, reducing adverse selection |
| Trade cascade (rapid fills market-wide) | Spreads widen substantially |
| Fast directional price movement | Spreads widen proportionally to velocity |
| Anomalous price behavior | Bayesian detection triggers spread widening |
| Competitive spread compression | Detects when other MMs tighten aggressively |
| Calm, balanced conditions | Spreads tighten for maximum fill rate |
Toxic Flow Detection (VPIN)
Volume-Synchronized Probability of Informed Trading — a well-known market microstructure metric that detects when informed traders are active. When VPIN rises above threshold, the bot widens spreads automatically. This is the same class of signal that institutional market makers use to avoid getting picked off.
Adverse Selection Tracking
After every fill, the bot tracks what happens to price. If fills are consistently followed by adverse price movement (you buy, price drops), the bot detects this toxic period and widens spreads until conditions normalize. Learns from its own execution quality in real time.
Wick Suppression
Detects rapid price spikes and suppresses requoting during wick events. Without this, the bot would chase wicks — placing orders into the spike that get filled at terrible prices and immediately reverse. The suppression window prevents this entirely.
Trade Cascade Detection
Monitors the rate of trades across the market. When trade rate spikes dramatically above baseline (indicating a cascade of liquidations or panic), spreads widen substantially. This prevents the bot from getting caught in a waterfall of fills during flash crashes or liquidation events.
Velocity-Based Protection
When price is moving fast in one direction, the bot scales its spread proportionally to the speed of the move. Slow moves = normal spreads. Fast moves = wide spreads. Prevents getting steamrolled by momentum.
Liquidity Fade Detection
Monitors orderbook depth changes in real time. When market makers pull their orders from one side of the book (a "fade"), it's often a precursor to a directional move. The bot detects this and can pull its own orders preemptively — avoiding fills right before a move against it.
Post-Exit Guard
After closing a position, the bot enters a mandatory observation period. During this window, it watches the market without placing orders, collecting data on the new regime before re-entering. This prevents the common pattern of closing a winner then immediately re-entering at a worse level.
Underwater Protection
When holding a losing position, the bot never exits at a loss through its normal quoting logic. Instead, it stays on the build side to average down, waiting for price to return. The stop-loss uses a limit chase exit to guarantee maker-only execution (100% fill rate, zero taker fees). Because exits are always maker-only, it is strongly recommended to also set a manual stop-loss directly on the exchange as a safety net — if the bot is unable to close in time during extreme moves, the exchange-level stop will catch it.
Position Limits
Hard caps on maximum position size prevent the bot from building an outsized position in trending markets. When approaching the limit, new order placement is blocked entirely.
Rapid Fill Pause
If multiple fills occur in rapid succession (indicating the bot is getting run over), it pauses all activity briefly. This circuit breaker prevents runaway position accumulation during flash moves.
Execution Intelligence
Dynamic Take-Profit Tiers
Take-profit targets scale with position size. Small positions target wider profits (more room to run). As position grows, TP tightens to lock in gains. The tier structure is configurable per-asset and adapts to the user's risk tolerance.
Trailing Limit Exits
Instead of market-order exits that pay taker fees, the bot uses progressive limit orders that trail favorable price movement. The exit order walks toward mid-price in small steps, capturing additional profit while maintaining maker-only execution. Floors at minimum profit to guarantee the exit is always profitable.
Inventory Skew
When holding a position, the bot asymmetrically adjusts its quoting to encourage closing trades. Long position = tighter ask (encouraging sells), wider bid. Short position = tighter bid, wider ask. The skew is calibrated to be subtle enough to maintain fill rate while gradually reducing exposure.
Microprice Fair Value
Instead of using raw mid-price, the bot calculates a depth-weighted fair value using the Stoikov microprice model. This accounts for asymmetric orderbook depth — if there's 10x more size on the bid than the ask, fair value isn't at mid, it's closer to the ask. Orders are placed relative to this fair value for more accurate pricing.
Conviction Scaling
When confluence is high, spreads tighten and the bot quotes more aggressively. When confluence is low, spreads widen and sizing decreases. The bot doesn't treat every cycle the same — it scales its aggression to its confidence level.
Self-Learning Signal Tracking
Every trade the bot takes is journaled with the full market context at entry — which signals were active, orderbook state, flow metrics. On exit, the outcome is recorded and each signal's effectiveness score is updated: signals present during winning trades gain confidence, signals present during losing trades lose it. Over time, a salience leaderboard emerges — the bot knows which signals are reliably predictive and which are noise. Signals that consistently underperform are automatically demoted. This creates a compounding data advantage: the longer the bot runs, the more refined its understanding of what actually works in current market conditions.
Adaptive Spread Optimization
The bot records its own fill data and learns which spread levels produce the best execution over time. After a calibration period, it can autonomously suggest or apply optimal spreads based on its own performance history — effectively tuning itself.
Dormancy Mode
When price isn't moving, the bot stops requoting. This saves API rate limits, avoids unnecessary order churn, and prevents getting filled on a wick during a dead market. The bot wakes up instantly when price moves beyond threshold.
Cross-Asset Hedging
Optional hedging of directional exposure using a correlated asset. A BTC long can be hedged with an ETH short (or vice versa). The hedge ratio, trigger threshold, and PnL activation are all configurable. When the primary position hits stop-loss, the hedge is closed simultaneously.
Microstructure Analysis
The bot computes institutional-grade market microstructure metrics in real time:
- Price Impact — Measures how much a given trade size moves the market. High impact = illiquid, dangerous conditions.
- Illiquidity Index — Tracks the ratio of absolute returns to volume. Rising illiquidity = wider spreads needed.
- Institutional Flow — Detects patterns consistent with large players entering or exiting positions.
- Orderbook Level Duration — How long price levels persist before being consumed. Short-lived levels indicate aggressive taking.
- Footprint Analysis — Volume at each price level with absorption detection. Identifies where large orders are sitting and absorbing flow — invisible resistance/support.
These metrics feed into the confluence scorer and spread engine, giving the bot a structural read on market quality that goes beyond simple price action.
Trading Modes
HFT Scalping (Default)
The primary mode. Quotes one side at a time, alternating rapidly when flat to capture spread on both sides. When a position is established, commits to the profitable exit side. Sub-second cycle times in production.
Swing Mode
Transforms the bot from a scalper into a swing trader. Wider profit targets, trailing activation at higher thresholds, and longer post-exit observation windows. For users who want fewer, larger trades on higher-conviction setups.
Exit-Only Mode
Emergency or wind-down mode. Activates automatically when max inventory is reached — the bot stops opening new positions and only places limit orders to close existing ones at a profit. All exits are maker-only limit orders that trail favorable price movement, ensuring zero taker fees. The bot will wait for price to return to profit before closing — it never market-orders out at a loss. Useful for gracefully winding down exposure without paying spread or taker fees.
Data Infrastructure
C++ Orderbook Engine
The orderbook is maintained by a C++ bridge fed via WebSocket, eliminating REST API latency entirely. The bot reads from an in-memory orderbook that updates in real time — no polling delays, no stale data.
Multi-Exchange Data Pipeline
Real-time WebSocket feeds from Bybit, Binance (spot + futures), and Coinbase. Cross-venue analysis surfaces divergences that single-exchange strategies cannot detect.
Orderflow Recording
Every orderbook snapshot and trade event can be recorded to a local database for ML training. This creates a proprietary dataset that improves the bot's models over time — a compounding data advantage.