The sign of gamma identifies regions where variance cancellation can (in theory) lead to improved S/N. We can also visualize gamma itself. The ideal gain in S/N might often not be all that desirable in practice if the bias from signal suppression is large compared to the statistical uncertainty after cleaning (this can be assessed from the ratio of bias to noise after cleaning). We can mitigate this bias by, for example, restoring the signal based on measurements of the g-I correlation but only to the point where the mean-squared-error (MSE) is minimized. The fractional gain in S/N at the MSE optimum is shown here for the case where cleaning and signal-correction are both performed on the same patch (fsky=0.03, l=100). Prospects are significantly improved when the g-I spectrum can be measured on a larger patch: for example, here is the gain in S/N when the patch where g-I is measured has fsky=0.8 and the cleaning patch still has fsky=0.03 (still for l=100).
