wtf is xau-dxy-session-divergence?

This project is a research study that tests a trading idea about gold prices and the US Dollar Index moving apart from each other during specific trading session windows, like the London or New York market open. It is part of a public series of studies where the author writes down the exact rule and test method before looking at the confirming data, so the result cannot be quietly adjusted afterward to look better. The main, pre-registered version of the idea did not pass the bar the author set for themselves ahead of time. The average return per trade was slightly positive, but the statistical test used to check if that result could just be random chance came back just above the cutoff needed to call it a real effect. So the headline claim is reported honestly as not confirmed. However, when the author broke the results down by time window, a more interesting pattern showed up. Results tied to the London trading session were inconsistent depending on how the price data was measured, five minute bars versus fifteen minute bars gave very different answers, which is a sign of noise rather than a genuine pattern. Results tied to the New York session, on the other hand, stayed consistent across several different measurement windows, suggesting there might be something real happening there, even though this was not the exact claim that was locked in ahead of time. The author also ran an extra, clearly labeled after the fact check using a wider four hour window instead of the original two hour one. This is described honestly as exploratory only, since it was run after seeing the results, and it showed the New York pattern getting stronger while the London pattern got weaker, further supporting the idea that only the New York effect might be worth investigating further. The repository includes the locked methodology document, Python scripts that collect the price data and run the official analysis, an extra exploratory script, the raw data collected, and the result files as CSV documents. The project is shared under the MIT license, and the author points to a separate follow up repository where the New York specific idea gets its own fresh, properly locked in test.