Skip to system research
Build A Trading SystemRules / Evidence / Replay

A testable trading edge

An explanation needs a counterexample and a comparison baseline.

Original edition: 8 October 2026. Source limits appear below. Expanded edition: 13 October 2026.

Educational research design and invented arithmetic examples. No strategy returns, broker execution or production certification are supplied.

The specific research question

State what should happen, why it might happen and what would reject the premise. Define the benchmark before seeing results and retain the cases in which no trade was possible.

Name an alternative explanation

A result could reflect a broad exposure, a favourable regime, a survivor-only sample or an execution assumption. Choose a baseline capable of distinguishing those explanations from the proposed mechanism. The relevant comparison is determined by the question, not whichever benchmark makes the strategy look strongest.

Decide what rejection means

A hypothesis can fail its reserved evaluation, remain untestable because a required historical feature is missing, or be invalidated by a software defect. These are different outcomes. Report the exact one instead of presenting a missing experiment as evidence of a failed market.

Make the check reproducible

Describe the cause, expected effect and observable alternative explanation. Decide which experiments can reject the idea before searching for parameters. Record every material research choice, including unsuccessful runs; the best final curve is not the whole research population.

What this page does not establish

Educational research design and invented arithmetic examples. No strategy returns, broker execution or production certification are supplied.

Sources and scope

Causal mechanics do not supply an economic cause

The fixture checks whether available completed closes produce reproducible 3/5 crossover intentions. It does not posit or validate a profitable market mechanism. The bars were deliberately constructed, so their outcomes cannot count as independent observations of such a mechanism. A real hypothesis needs an appropriate market, baseline, eligible population and counterexample design. Passing warmup, prefix or cash tests helps establish software integrity but does not change that evidential absence. An engine result and a testable economic explanation remain different objects even when both are called a system.

A negative result needs the right label

The six synthetic positions finish net negative 1.20 under fixed fees. That is neither a verified no-model master nor evidence that a real crossover family fails the authorised research process. It is a complete software-reference result. A genuine financial rejection would require the full predeclared method, appropriate source data and independently checked evaluation. A positive synthetic variation would have the same limitation. Preserve the original result and settings instead of searching the invented bars for a prettier parameter combination and describing the selected curve as independent proof.