qlab
A quantitative research framework built around one principle — the validation layer matters more than the strategy layer.
0.910.05
Why
Anyone can produce a backtest with an impressive Sharpe ratio. The hard part is knowing whether it means anything. qlab is built so that self-deception is difficult — and so that when a result does survive, the survival means something. Nothing in it is a trading system.
What it contains
- Validation: embargoed walk-forward testing, a locked holdout, the probabilistic and deflated Sharpe ratio.
- Honest backtests: transaction costs are mandatory, every tested configuration is logged, and explicit checks guard against lookahead and survivorship bias.
- Market microstructure: a limit-order-book simulator for studying where market-making PnL comes from.
- Test suites that run on every push via GitHub Actions — including the check that a perfect-foresight signal produces an absurd Sharpe and a same-bar signal produces roughly nothing.
The result that shaped the method
Across a 66-configuration momentum sweep, the best out-of-sample Sharpe of 0.91 deflates to 0.05 once corrected for the number of trials. The findings are documented in seven research notes.