We took nine years of results and split them into stages: the bare shape of the trade on its own, then with the forced-exit rule added, then with the regime filter on top. Each stage measured on the same data, in the same denomination.
The point was to find out which part of the machine earns.
The answer was uncomfortable and useful: almost none of it comes from the signal. The signal on its own is worth very little. Nearly all of the result comes from two rules layered on top of it, and one of those two carries most of the weight. The table is in Letter 001.
We publish it because it dictates how we work. If nearly all the value sits in one component, then nearly all the risk sits there too. That component has failed on us once, quietly, and we did not notice for longer than we would like. It is now the first thing checked on every run, with tests written around it that fail if it stops working.
The habit generalises. A strategy is a stack of ideas, and the honest question is not whether the stack makes money but which layer does. Taking ours apart cost an afternoon and we had not done it in nine years of results. What it showed was that the clever part contributed almost nothing and a plain mechanical rule contributed nearly everything, which is not a flattering finding and is exactly why it was worth measuring rather than assuming.