I agree with the emphasis on "preference".
Comparison to best is a reasonable option, but becomes less and less useful the more you messed up on your best lap (or there was traffic).
If in turn 123 you had traffic or a mistake and you lost 0.5 seconds, each time you try to beat your best you'll need to remember that as much as +0.5 until turn 123 can still give you a new best, as long as you don't mess up 123, and you do the remainder of the lap exactly the same. Then you end up either with a "how well do I remember what I did on my best lap" problem while driving to extract anything reliable from the time/speed delta.
Same thing applies to "predicted lap time". If best lap is used as comparison, and this time there's no traffic / no chance of mistake, the predictive lap time will be slower. It will basically assume that you're going to repeat every single mistake from your best lap on the unfinished part of the track.
Comparison against optimal will of course be too optimistic in most cases, and your time delta will almost always be red. This is how it works in iRacing, and I generally like it.
When it comes to predicted lap time, using opt as a reference will almost always give you a prediction faster than you end up doing. I think that would still be useful though, at least as an option.
Which makes me wonder if we could use some kind of a percentile (such as median) instead of opt. "Lap time predicted based on your 90th percentile best sectors"?
