BACK TO WORK
LIVE

Backtesting Engine

A backtesting engine with eight different execution modes — vectorized for speed, event-driven for realism, tick replay for full fidelity. Walk-forward validation prevents overfitting, and the C++ hot-path handles order book updates in microseconds.

8engines
165+metrics
<10µslatency
Parquetdata format

TECH STACK

PythonC++17pybind11Parquet

The Problem

Most backtests lie. Overfitting, survivorship bias, and look-ahead bias make 90% of strategies look great in research and fall apart in live markets.

The Approach

Built an engine that enforces walk-forward validation by default, models transaction costs realistically, and uses a C++ hot-path so tick-level replays don't take days to run.

Key Features

Walk-Forward Validation

Train on the past, test on the unseen future. Always.

Multiple Execution Models

Vectorized, event-driven, tick replay. Pick the fidelity you need for the test you're running.

Regime-Adaptive Sizing

Position sizing adapts to detected market regime — trending, ranging, high vol, low vol.

C++ Hot-Path

Order book updates handled in C++ via pybind11 for sub-10µs latency.