Alpha Intelligence Engine

The Alpha Intelligence Engine scans trades across connected exchanges, identifies top-performing traders, reverse-engineers their strategies, and generates signals from their behavior.

Architecture

graph TD
    subgraph "Data Collection"
        EX[Exchange APIs] -->|trades| SC[Trade Scanner]
        EX -->|leaderboards| LB[Leaderboard Tracker]
    end

    subgraph "Analysis"
        SC --> AS[(Alpha Store)]
        LB --> AS
        AS --> PR[Trader Profiler]
        PR -->|profiles| AS
        AS --> CL[Strategy Classifier]
        CL -->|fingerprints| AS
    end

    subgraph "Signal Generation"
        AS --> AF[Alpha Follower Agent]
        AF -->|Signal| ARB[AIS Arbitration]
        ARB --> RE[Risk Engine]
    end

Components

Trade Scanner

Polls connected exchanges for notable trades:

• Whale detection — flags trades above a configurable notional threshold (default: $50,000)
• Volume anomalies — detects trades that are 3x+ the recent average for a symbol
• Deduplication — tracks seen trade IDs to avoid double-counting

Leaderboard Tracker

Monitors copy-trading leaderboards from Binance, Bybit, and other exchanges:

• Periodic snapshots — captures leaderboard state every hour
• Rank history — tracks how traders move up/down over time
• Consistency detection — identifies traders who maintain top positions across multiple snapshots

Trader Profiler

Builds statistical profiles from observed trade data:

Metric	Description
Win rate	Fraction of trades with positive P&L
Sharpe ratio	Risk-adjusted return (annualized)
Sortino ratio	Downside-risk-adjusted return
Max drawdown	Largest peak-to-trough decline
Profit factor	Gross profit / gross loss
Consistency score	Stability of win rate over time (0-1)
Avg holding period	Mean trade duration in minutes
Trade frequency	Trades per day
Preferred symbols	Most traded instruments

Traders are classified into tiers:

Tier	Criteria
Elite	Top 1% — sustained edge, high Sharpe, 60+ composite score
Strong	Top 5% — consistently profitable, 45+ score
Notable	Top 15% — above average, 30+ score
Average	Median performers
Weak	Below average or insufficient data

Strategy Classifier

Reverse-engineers trading styles from trade patterns:

Style	Detection Criteria
Scalper	Average holding < 15 minutes
Momentum	Intraday holding, moderate win rate
Mean Reversion	High win rate (>65%), low return variance
Breakout	Lower win rate (<50%), large winner/loser ratio (>2x)
Swing	Holding 8h-7d
Trend Following	Holding > 7 days
Contrarian	Low win rate but very large winners (>3x losers)

Produces a StrategyFingerprint that captures entry timing, exit patterns, sizing behavior, and market condition preferences.

Alpha Follower Agent

Generates AIS Signals when top-tier traders open positions:

1. Queries AlphaStore for recent activity on the target symbol
2. Filters for trades from traders meeting the minimum tier requirement
3. Scores each activity using: base_confidence(tier) + consistency_bonus + win_rate_adj + sharpe_adj
4. Applies recency decay (fresher = higher confidence)
5. Emits the strongest signal through the standard AIS pipeline

The agent: - Extends the standard Agent ABC - Produces standard Signal objects - Is subject to mandate governance and risk validation - Participates in weighted arbitration alongside other strategy agents

Configuration

# config/intelligence.yaml
intelligence:
  enabled: true
  scanner:
    whale_threshold_usd: 50000
    volume_spike_multiplier: 3.0
  leaderboard:
    refresh_interval_seconds: 3600
    max_rank_to_track: 100
  agent:
    min_tier: notable
    max_activity_age_seconds: 3600
    agent_weight: 0.6

Data Flow

sequenceDiagram
    participant EX as Exchange
    participant SC as Scanner
    participant AS as Alpha Store
    participant PR as Profiler
    participant CL as Classifier
    participant AF as Alpha Follower
    participant AIS as AIS Pipeline

    loop Every 60s
        SC->>EX: get_my_trades(symbol)
        EX-->>SC: TradeRecord[]
        SC->>AS: append_activity(whale/anomaly trades)
    end

    loop Every 1h
        PR->>AS: get_activities(trader_id)
        PR->>PR: compute metrics
        PR->>AS: upsert_profile(profile)
        CL->>AS: get_activities(trader_id)
        CL->>CL: classify style
        CL->>AS: save_fingerprint(fp)
    end

    loop Every AIS cycle (60s)
        AF->>AS: get_activities(symbol, recent)
        AF->>AS: get_profile(trader_id)
        AF->>AF: score + select best
        AF-->>AIS: Signal(alpha_follower)
    end