Methodology

The rules we backtest under, so you can decide if the numbers mean what we say they mean.

Universe

MNQ continuous front-month, 1-minute bars, ET session clock.

Training / out-of-sample split

Models are developed on a training window and then validated on an untouched later window. We report both, side-by-side.

Costs

• Fills assumed at bar close — no optimistic fills on wicks.
• Slippage modeled where material; stated on each model page.
• Ladder chassis: 3 contracts, 30pt stop, 20 / 40 / 60pt TPs unless specified.

Honesty checks

• WRONG-DIR > baseline = real signal (direction inversion).
• NULL > baseline = filter is doing the lifting, not the thesis.
• Oracle filters (peek-ahead) are flagged and never ported live.

Full methodology writeup in progress.