Conclusions

QUaD shows that bolometer pair differencing can work very well on the ground
Sidelobe performance was bad and we paid for it
Next gen expt will surely do much better
...but they'll need to due to higher sensitivity
QUaD achieved adequate temporal response match (post deconv)
Hints of spectral response mismatch
Use CMB as calibration source?
Interesting pair centroid offsets observed
...but irrelevant to QUaD's sensitivity
Only real way to understand expt requirements is end-to-end simulations
Something I worry about for the future: differential readout cross talk from channel X into A&B of a pair