A backtest tells you how a trading strategy would have performed on past data. It is the core of quantitative research. Manifold-BT runs 500,000 of them in a few seconds, on a Rust engine with a Python API. It also ships an MCP server, so Claude or Cursor can run the tests directly and answer with real numbers instead of a plausible guess. Ask in plain English, get measured results.
Systematic trader here. Testing an idea used to mean five or six versions in an afternoon, then picking the best-looking one. Which is a great way to fool yourself.
The engine is now written in Rust. 500,000 backtests in a few seconds, and fifteen years of 1min bar data runs in about a second.
Speed alone would just make overfitting faster, so the lookahead detection and validation tools are built in, not bolted on.
With the MCP server, Claude runs the whole loop itself. Hypothesis, test, result, next hypothesis. AI-driven quant research where the numbers are measured rather than guessed.
There is a free tier, so go run your own numbers rather than trusting mine.