ORB Engine

ORB Engine needed to support research, validation, live session behavior, broker and data integrations, and observability without duplicating strategy logic across every process. The architecture keeps the core behavior consistent while allowing each path to handle its own inputs, timing, and operational concerns.

The result is a modular algorithmic trading research, execution, and observability platform built around Opening Range Breakout systems, with reliability and deterministic behavior treated as first-class engineering requirements.

Twelve configured crypto ORB sessions cover four assets across three market windows. YAML configuration controls which sessions are enabled and how they operate, allowing concurrent runners to share infrastructure without turning session behavior into copied application code.

The live path ingests five-minute candles, computes opening ranges, and scans for break, retest, and rejection behavior. Weekend handling, configurable session behavior, and signal-only mode are treated as explicit operating concerns. Live crypto scanning and historical research now use Kraken Futures candle data by default, keeping the backtest and streaming paths aligned to the same contracts rather than validating behavior on one instrument and operating on another. OKX and Kraken spot remain supported data venues, but they are not the default research/live pairing. The crypto session runner is signal-only, while futures and margin runners have optional order-placement paths.

The backtesting subsystem is driven by YAML parameters and follows the same conceptual session flow: opening-range calculation, break, retest, and rejection detection, long and short support, stop and target evaluation, session limits, and outcome statistics.

The research path exists to validate behavior, not to advertise returns.

Randomized parity testing verifies that the streaming signal engine produces the same decisions as the causal backtesting path. The parity harness now runs 200 randomized seeds across four configuration combinations, producing 800 automated live/backtest equivalence cases. The combinations represent base behavior, direction-aware rejection, close-related behavior, and combined close/range behavior.

Separate replay validation has also compared the streaming scanner and causal backtest across 500 randomized sessions and 2,466 emitted signals with zero mismatches.

Regression coverage also validates break selection and direction-aware rejection behavior, making parity a maintained contract rather than a one-time comparison.

The system includes circuit-breaker logic, configurable consecutive-loss limits, cooldown sessions, session trade limits, stop-on-first-loss options, and signal-only mode. These controls are expressed through configuration-driven behavior so operating constraints remain visible, testable, and separate from the core signal logic.

A FastAPI monitoring surface provides health information, logs, log-file access, dashboard views, signals, and summaries or aggregations where appropriate. The monitoring path is read-only, with structured signal parsing and safe log-path handling that protects against path traversal.

ORB Engine includes a structured post-session narration layer using OpenAI and Pydantic. The narrator summarizes completed sessions and produces structured output for interpretation after the system has finished its work.

AI is intentionally isolated from deterministic trading and risk logic. It does not influence trade decisions, position sizing, or circuit-breaker behavior, and it fails gracefully when AI services are unavailable.

• Python
• FastAPI
• Pandas
• NumPy
• PyYAML
• OpenAI
• Pydantic
• WebSockets
• Kraken
• Coinbase
• Interactive Brokers