Research framework

Research before conviction.

NQ Route is built around a simple rule: a useful market idea has to survive outside the sample that created it. Training results, out-of-sample results and live forward results are kept separate.

451Resolved sessions in the 2023–2024 training sample used for the frozen open-location benchmark.
214Resolved sessions in the 2025 out-of-sample set used to test that benchmark without refitting it.
77.6%OOS route-classification accuracy for open location alone. This is not the same thing as trade win rate or profitability.
What survived

Open location remained the strongest first filter.

Adding previous-day candle information, AVWAP and simple higher-high / higher-low structure only improved raw classification slightly and reduced or failed to improve ranking quality enough to justify heavy additional scoring.

That led to a deliberate design choice: keep the initial prior simple, then let live auction behavior do the heavier lifting.

Important: the public percentages describe specific research tasks with specific definitions. They should not be read as a promise that 77.6% of trades will win.
Validation discipline

Contexts have to earn their place.

Current context priors are required to show adequate train and OOS sample size, preserve the same directional preference and improve meaningfully over the relevant side baseline before they are promoted.

TrainUsed to discover or fit the hypothesis.
Out of sampleUsed to check whether the effect survives on unseen historical data.
ForwardUsed to track what happens after the model is frozen and running live.
Public snapshot

Examples of validated behavior.

We publish enough to show the research philosophy without publishing the entire live scoring recipe.

PDL

Acceptance context

Several PDL contexts showed materially stronger acceptance rates than the side baseline in both train and OOS samples.

VALIDATED • OOS CONSISTENT
PDH

More selective edge

PDH-side contexts were generally weaker and more selective, reinforcing the need for live confirmation instead of symmetric assumptions.

ASYMMETRY MATTERS

Failed-loss behavior

When an accepted break later reclaims the level, the route often changes character and rotation targets become relevant again.

STATE CHANGE > STATIC BIAS

Full research notes are for members.

Detailed context tables, sample sizes, side baselines, route definitions and forward-validation notes will live inside the community area rather than on the public marketing page.

Join the beta