Intelligent risk management and compliance monitoring system for 73+ exchanges. Provides real-time risk assessment, ML-driven anomaly detection, and regulatory compliance support.

• RiskManager — Central risk management orchestrator for pre-trade, in-trade, and post-trade risk controls
• RiskAssessor — Real-time risk scoring and limit enforcement

• RiskEnsembleModel — Ensemble model combining Random Forest, Neural Network, and XGBoost for risk prediction
• AnomalyDetector — Behavioral pattern recognition and outlier detection for suspicious trading activity

• RiskEvent — Structured event type for risk-triggered incidents
• RiskEventType — Enum of event types (POSITION_LIMIT, MARGIN_CALL, etc.)
• RiskLevel — Enum of risk levels (LOW, MEDIUM, HIGH, CRITICAL)
• RiskMetrics — Container for aggregated risk metrics

DurableRiskGate and the instrument admission types provide deterministic, local contracts for offline fake-client smoke tests. They do not source trusted account, position, exchange metadata, or policy facts, and do not constitute a real-account risk approval or enable a provider write route. A production composition must independently supply and verify those authorities.

A successful dispatch claim leaves a dispatch-inflight:<intent_id> account freeze. settle and ensure_settled require the matching durable claim and do not clear that latch. The generic resolve_freeze method refuses dispatch latches. resolve_dispatch_freeze requires an injected journal authority that holds the account-wide writer fence while it verifies an exact current journal row, the single dispatch attempt, and the account/intent/permit/claim binding, then yields a typed VerifiedDispatchResolution bound to the proof digest. A bare boolean cannot clear the latch; the receipt itself is not a signature or proof of provider truth outside that injected authority. The allowed resolutions are a rejected or canceled zero-fill terminal, or an ACKED_TRACKED transfer to an exact durable order exposure reservation. The tracked order's settled notional remains counted as risk capacity; this API does not clear positions, margins, fills, or later order lifecycle state. No authority is installed by default. The Iteration 41 CTP reconciliation observation alone is not sufficient: its independent native queries do not prove a common atomic snapshot or an account-wide writer fence. A verifier that cannot prove those facts must keep the latch active. Fake simulation evidence is restricted to fake/fixture scopes and cannot bind a native CTP scope.

• Market Manipulation Detection (spoofing, layering, front-running)
• Anti-Money Laundering (AML) and Know Your Customer (KYC)
• MiFID II and SEC Rule 606 trade reporting
• Position limits and concentration risk limits
• Credit risk and margin management

pip install bt_api_risk

• Python >= 3.9
• numpy >= 1.26.0
• pydantic >= 2.0.0

from bt_api_risk import RiskManager, RiskAssessor, RiskLevel, RiskEvent, RiskEventType

# Initialize the risk manager
risk_mgr = RiskManager()

# Create a risk assessment
assessor = RiskAssessor()
risk_score = assessor.assess_position(
    exchange="BINANCE___SPOT",
    symbol="BTCUSDT",
    position_value=100000.0,
    account_equity=500000.0,
)
print(f"Risk level: {risk_score.level}")  # RiskLevel.MEDIUM

# Emit a risk event
event = RiskEvent(
    event_type=RiskEventType.POSITION_LIMIT,
    level=RiskLevel.HIGH,
    exchange="BINANCE___SPOT",
    symbol="BTCUSDT",
    message="Position exceeds single-symbol limit",
)
risk_mgr.emit(event)

from bt_api_risk import AnomalyDetector

detector = AnomalyDetector(threshold=0.8)
is_anomalous = detector.is_anomalous(
    features={
        "trade_frequency": 150,
        "order_size_ratio": 0.95,
        "price_impact": 0.12,
    }
)

from bt_api_risk import RiskEnsembleModel

model = RiskEnsembleModel()
risk_prob = model.predict_risk(
    features={
        "position_concentration": 0.75,
        "leverage_ratio": 2.5,
        "daily_volatility": 0.04,
        "liquidity_score": 0.6,
    }
)
print(f"Risk probability: {risk_prob:.3f}")

Full documentation available at bt_api_py documentation.

MIT License