Not bad, uh?
While from the "Fitted y-intercepts" plot we get an insight of the system linearity. All the y-intercepts should be zero because at a signal with zero length should correspond a pulse with a zero length. As we can see on the graph they all are compatible with zero, but not the 18-prescaler dataset. This dataset, though, is the worst one because is has just two data and its calibration can not be trusted.
Following there is a table with the acquisition rates for each prescaler setting.
|Prescaler||Acquisition rate [kS/s]|
|128||9.74 ± 0.04|
|64||19.39 ± 0.06|
|32||37.3 ± 0.6|
|16||75.5 ± 0.3|
|8||153 ± 2|
I also tried an unweighted fit of the rates because you can see that they roughly double when the prescaling halves, this looks like an inverse proportionality law. So I fitted the rates vs the prescaler settings with a simple law of
I got a value for a of
with a χ²=3.14 and 4 degrees of freedom, this means that the law is accepted with a 95% confidence level!