A strategy that looks good on its own data is either an edge or a well-groomed coincidence, and the difference is not visible from the inside. So we ran the test that charges you for looking.
Not on the configuration we like. On every configuration our parameters can produce, which is a few hundred of them, using a method that raises the bar in proportion to the size of the search.
It passed, and the result is in Letter 001. Two things about the exercise mattered more than the number.
The shape of the surface. Almost all of the configurations were positive and none had a negative mean, which is a plateau rather than a peak. A single sharp peak surrounded by mediocrity is what luck produces; a broad region that works is harder to explain that way.
And where we sit on it. The configuration we trade is not the best one in the set. We did not choose the maximum, and we can show the rank. A manager trading the top of their own grid has told you which number they optimised.
The obligation we took from this: wherever a multi-year curve of ours appears, the size of the search that produced it appears next to it. Not in a footnote.
How this all works, without our numbers in the way, is in the method notes.