Cost, selection, and what an accuracy number leaves out

At what scale can a trading judgment matter? How many different opportunities exist, and what would useful selection need to improve?

The study followed seven questions rather than searching for one profitable setting. Its durable contribution is a set of comparisons: cost against distance, observed behaviour against a baseline, and ticket count against shared exposure.

We compared spread with trade distances, tested simple predefined directional reads, and measured several path properties at different horizons. Structural-level tests used shifted levels as placebos. Path comparisons used shuffled changes as another reference.

Portfolio comparisons measured related outcomes when instruments shared a currency or broader group. The accuracy ladder mixed hindsight-correct and hindsight-incorrect sides of the same entries under stated trade geometries.

The study also explored two scheduled-flow hypotheses. Minute and hourly data were drawn from the same broker history; gold and NZDUSD hourly series were rebuilt from minute data where necessary.

Spread relative to the trade's risk distance explained a substantial part of the differences in break-even requirements. Lower-volatility regimes were more expensive in those terms. They were not consistently more random on the measures tested.

The simple trend, strength, level, and impulse definitions did not establish a useful directional advantage after the relevant comparisons. That result applies to these definitions and windows. It does not show that every technical, contextual, or discretionary judgment is equivalent to a coin flip.

Grouping by instrument names only partly captured overlapping bets. Shared currency exposure and related outcomes revealed links across groups and differences within them. The number of distinct directions depended on the sample and method.

The month-end London fixing hypothesis produced a directional lead worth investigating. Its evidence comprised 26 monthly opportunities. Seven related dollar pairs did not turn one monthly dollar view into seven independent replications. The Tokyo fixing hypothesis did not show the proposed effect.

Changing the assumed share of correct calls changed average simulated R linearly within the construction. Correct calls could still stop out, and incorrect calls could still win. The ladder therefore improves on treating execution as a fixed percentage haircut.

A real trader can be correct on a different set of paths. These coefficients cannot convert an arbitrary accuracy claim into a promised win rate or income.

The tests used defined historical samples, limited execution assumptions, and many overlapping windows. Split-half agreement adds useful context but does not replace forward evidence. The suggested trade frequencies and ATR distances in the research notes remain hypotheses, not instructions in this guide.

The study's broad information and market-mechanism discussion also contains interpretations beyond what candle data alone can establish.