Redline Exchange

A deterministic price-time-priority exchange matching engine written in Python.

This is a software-engineering project, not a trading-strategy dashboard. It focuses on data structures, deterministic behavior, invariants, and performance.

Features

• limit and market orders;
• buy and sell sides;
• price-time priority;
• partial fills across multiple price levels;
• cancellation by order ID;
• cancel/replace with explicit queue-priority rules;
• deterministic CSV event replay;
• resting-order (maker) execution prices;
• top-of-book and depth snapshots;
• integer price ticks to avoid floating-point money errors;
• invariant and property-based tests;
• reproducible throughput, p50, and p95 latency benchmarks.

Quickstart

git clone https://github.com/asoracca/redline-exchange.git
cd redline-exchange
python -m venv .venv
source .venv/bin/activate
python -m pip install -e ".[dev]"

python -m pytest -q
python -m orderbook.cli
redline-demo
redline-replay examples/events.csv --symbol DEMO
python benchmark.py --orders 10000 100000 1000000 --csv benchmark-results.csv

On Windows PowerShell, activate the environment with:

.venv\Scripts\Activate.ps1

Example

from orderbook import LimitOrderBook, Side

book = LimitOrderBook(symbol="DEMO")
book.submit_limit("sell-1", Side.SELL, quantity=100, price_ticks=10100)
trades = book.submit_limit("buy-1", Side.BUY, quantity=40, price_ticks=10100)

print(trades[0])
print(book.top_of_book())

10100 means 101.00 when the tick size is one cent. The engine stores integer ticks internally.

Matching rules

1. Better prices execute first.
2. At the same price, earlier orders execute first.
3. A trade executes at the resting order's price.
4. Unfilled limit quantity rests on the book.
5. Unfilled market quantity is canceled rather than resting.

Replace rules

replace(order_id, new_quantity, new_price_ticks=None) treats quantity as the new remaining quantity.

• Reducing quantity at the same price keeps the original sequence and priority.
• Increasing quantity loses priority.
• Changing price loses priority and may immediately trade.
• Zero quantity is rejected; use cancel instead.

Heap entries include price, sequence, and order ID. A stale entry from a cancel or replacement cannot be mistaken for the active version of an order.

Correctness checks

assert_invariants() verifies that active orders have valid heap entries, the book is not crossed after matching, trade sequences increase, quantities remain positive, and no order trades more than its maximum accepted quantity.

The Hypothesis test suite generates randomized submissions, markets, cancels, and replacements and checks these invariants after every event.

See event replay and benchmark methodology.

See docs/DESIGN.md for the data structures, complexity, and extension plan.

Market microstructure research

Redline Market Lab adds deterministic noise traders, informed traders, and an inventory-aware market maker on top of the unchanged matching engine. It tests how informed flow, quote latency, volatility, and inventory limits affect marked P&L and risk.

python run_market_lab.py
python run_market_study.py

The second runner repeats five scenarios over 50 common random seeds, reports approximate 95% confidence intervals, decomposes P&L into execution edge and inventory revaluation, and creates a scenario chart. See the methodology and results.

Suggested development sequence

• Version 1: deterministic price-time engine and unit tests.
• Version 2: cancel/replace, CSV replay, invariants, property tests, and latency benchmarks.
• Version 3: agent-based market microstructure experiments and multi-seed inference.
• Version 4: C++ implementation and Python/C++ benchmark comparison.

Do not add a trading signal until the engine itself is correct, tested, and measured.