I would say that the incremental roll out of features was a fantastic choice and will prove to be the only way FSD is developable in the end.
However, to figure out what Mr. Dome should be refunded or credited is much more difficult.
The most straightforward part of the analysis would be that some number of buyers purchased FSD for an extra $3k from late 2016 until late 2018. That breaks down into two questions: (1) should a portion of that be refunded now? and (2) should all of it be credited if the owner trades in for another Tesla? I would argue that as to question (2) the answer would be an easy "yes" now, and probably through 2020.
If, the computer upgrade is done, and the other FSD features are released, then I guess the question shifts first, to my question (1): To Dome, who will have three years of no FSD, its not the same as some one who purchases a car in late 2019 and only has a couple of months in the same situation.
When all the hardware is upgraded and the features released, at least at that point there will be a measurement.
The problem is then going to be assigning some money to expectations. For that, one has to somehow calculate what Mr. Dome reasonably expected in the fall of 2016. Then, of course, what of those who bought in late 2017 or late 2018? They would not have ended up waiting as long.
By early 2019, the pricing had changes so much that even though FSD purchasers have, as of now, paid for something that is as yet undelivered, the roll out of the remaining FSD features should solve that issue.
The question, discussed to death elsewhere, of whether a particular person is satisfied with a feature, really does not lend itself to any rational analysis. Depending on their point of view, a particular buyer might be entirely satisfied (me) or completely un-satisfied until the car can drive itself with no occupant (not unreasonable either).