Jacob Brazeal posted about "A puzzle of two unreliable sensors" a few days ago. I thought that it was interesting, and wanted to try to dig into it deeper; I wanted to find where the non-linearity came from such that the optimal linear combination of the two sensors were 58% A + 42% B.

It does feel like linear combinations end up being related to rotations, or "phase", when you plot the signed error of A vs the signed error of B.