Deterministic replay
Event order and state transitions are designed to be reproducible so behavior can be inspected after a simulation run.
Event-Driven Market Engine
A deterministic event-driven simulation for exploring state, risk checks, replay, persistence, and streaming systems.
overview / problem
The project explores how stateful backend systems behave when strategy decisions, risk checks, simulated execution, persistence, streaming, and replay all have to agree on event order.
what I built
Engine independence, deterministic replay, state transitions, risk limits, execution simulation, persistence, streaming, Redis deduplication, and adapter separation.
architecture
Market events → Strategy engine → Risk engine → Execution simulation → Portfolio state → PostgreSQL / Redis → FastAPI → React dashboard
interesting decisions
Adapter separation
Historical data, execution simulation, persistence, and dashboard APIs are separated so simulated market data does not imply live brokerage connectivity.
Qualified benchmark
The repository documents roughly 97,000 events/second on an Apple M-series machine for the pure in-process Python event path without database writes.
limitations
- No real brokerage connectivity.
- No real trades.
- Simplified fill model.
- No order book.
- Benchmark excludes database I/O.
- Historical data adapter does not make this a live trading system.