We used an optimization criteria based on S/√(B+(ΔB)^2+(ΔS)^2)). We used an optimization criteria based on S/√(B+(ΔB)^2+(ΔS)^2)). For the sensitive channels we assigned a ~20$percnt signal uncertainty, ~25$percnt for the irreducible and lost lepton backgrounds, ~50$percnt for the fake lepton background, and ~100$percnt for photon→lepton and charge flips. The “issue” with preselection plots is that they do not show background rejections other analyses can achieve. For example, for e±mu± the MTmax rejects a lot of photon fakes that make the S/B much lower at low Mll. Therefore, in the end the sensitivity using the upper metric isn’t very different when using the tighter Mll selection, but lowering the cut allowed us to keep a little more signal acceptance, and since we have low signal yields, we opted for that choice.

-- HannsjoergWeber - 2019-01-22