Effect of Jitter Separation on Bathtub Estimation?

[capn-freako]

I'd like to understand how jitter separation affects PyBERT's extrapolation of the bathtub curves.
It stands to reason that separating the random jitter (i.e. - the only component assumed to follow a Gaussian distribution) would improve this extrapolation, but certain questions arise:

1. By how much does jitter separation improve the accuracy of bathtub curve extrapolation?
2. How does the Dual Dirac model of jitter play into this and is PyBERT modeling/implementing it correctly?
3. How much model-to-model variation is there in the accuracy of bathtub curve extrapolation, all else remaining equal?
4. How does the quality of jitter separation and faithfulness of Dual Dirac model implementation affect this model-to-model variation?