(A) Repetition of the analysis from the manuscript after altering activity within the non-driving cortex by removing the signal captured by the top PC. We did so by reconstructing each neuron’s response from all PCs (which would normally provide a perfect reconstruction) except the first. We took the original data matrix (with one column per neuron) and replaced it with , where is the matrix of PCs with the first column missing. Following this manipulation, generalization was lower when predicting the activity of driving-cortex neurons from non-driving-cortex neurons, rather than driving-cortex neurons (lines slope downwards). This pronounced decline is in contrast to the very modest decline observed in the data (see Figure 7F,G). (B) As in A, but using a different manipulation of non-driving cortex activity. Instead of removing signals, we created a mismatch in the size of signals in the driving versus non-driving cortex by manipulating the latter. We projected the original data matrix into its PC space, yielding . We divided the first two columns of by two, and multiplied columns three and four by two, yielding . We then reconstructed the activity of each neuron: and analyzed . Thus, all signals are present in both hemispheres, but to different degrees. Again, this creates a mismatch in generalization when predictions are based on driving- versus non-driving-cortex activity. These results illustrate that the original analysis can reveal if a signal is present in the driving-cortex, but absent, or of a different magnitude, in the non-driving cortex. Thus, the results of the original analysis indicate that any such differences are modest and/or restricted to higher PCs that capture relatively little variance.