Findings No. 86
Mid/side was the best arm, 75/76 = 98.7%
Mid/side topped five audio arms by one bar, but that edge sat inside the noise band, so a registered test decided it, not the raw leaderboard.
Mid/side scored 75 of 76, 98.7 percent, the highest of the five arms we tested. That is one bar above mono, and taken on its own it reads like a result worth keeping.
The context matters more than the number. All five arms in the test spanned only three bars of score, which is inside the noise we expect from a run like this. A one bar edge sitting inside a three bar spread is not evidence of anything by itself. It could just as easily be where mono, or any other arm, would land on a different pass.
We did not take the raw percentage at face value. The registered matrix, peaceful-dreams crossed with the five arms, was built to answer exactly this question before we saw the data: whether the mid/side edge was real or just noise. Running the comparison through that pre-committed matrix is the fix that shipped, in the sense that it replaced an eyeballed leaderboard with a test we could not quietly reinterpret after the fact. The matrix's answer was no, the edge was not real.
The general rule: a ranking is not a result. When your candidates already sit within each other's noise band, decide the comparison method before you look at the scores, not after, or you will talk yourself into a winner that a flat coin would have produced just as easily.
Receipt: Stem Shovel, the peaceful-dreams x five-arms registered matrix.