Hi Racecar,
Typically any image (or footage) should be undistorted, before used in the Camera Calibrator. Of course the is an amount of distortion that will be OK. However this is a process that calculates the data of a camera based on the idea of Vanishing points, which means that any distortion might change that and lead to less precise results. In that way, your question is spot on.
In short:
The idea is normally, that any lens correction will alter the quality of the image. As each change of the position of a pixel, which is not based on full pixel movements will mix pixel values. But exactly this is typically the case while correcting any lens distortion. The idea is, if not requested (director) otherwise, that the original footage as well the final footage stays untouched, for quality reasons. This means that anything in the pipeline has to be changed back and fore.
As a consequence the CG footage would suffer by adapting the distortion to it, later on. For that the Physical camera has some options. The suffering by itself sounds not nice, but most of the time I found myself adding a little blur to the CG footage anyway, as it is mostly more crisp than practical footage. If that isn’t good enough, you need to render larger, if the Physical camera is not your first choise anyway.
How to do that back and fore distorting? I like to work with “Lens Grids”. These are poster like Checkerboards, that one photographs and applies then the lens correction on. From there you know what the lens does, and the values from that process can be applied positive or negative to gain the results your need.
There is one thing that needs to be clear. An work on CG footage in post that was not lens corrected, will alter the size of the material. In some cases you need to render larger with a wider field of view—if—the CG content inside of the frame touches that area (area means here if the process shrinks to a certain degree the image, and the footage isn’t covering the full frame anymore)
To test your round-trip, your initial test should be, if you can work with the checker board inside Ae, and apply first the distortion to it and in a second step, reverse that effect. Then compare the original checkerboard and the double adjusted one, are they the same (perhaps Blend mode Difference).
If you like to work with footage, keep in mind that lens distortion can be tricky, and is not as simple as some believe. E.g., a Fisheye for example is in most cases recognized as lens distortion in the “best way” but it draws the lines as it should be… however there is an option with cheap constructions that these lines are not “drawn” as they should be, then I would talk about lens distortion. (This might be seen differently by some people…and there are different FishEye definitions available, so—long story) What I like to say is, there are several “layers” what a lens can do to your footage, a reason why I suggest to know your lens and the needed process.
I have shortly discussed the quality for images in my Siggraph Presentation
http://www.cineversity.com/vidplaytut/siggraph_rewind_2012_dr._sassi
As you might have the best quality, you might follow the list that I share there.
In short you need to know what you have as source footage or “background-footage” This will lead you pick one of the ways I have mentioned above.
All the best
Sassi