I never change the K System reference level during any production. One could go from a lower level to a higher level but I never do it in production. It is not good. If I start a project at K -20 then I finish it there with a K-20 pre mastered mix.
Mastering is the best way to covert a K-20 mix to say a K-12 mix. You start by adding 2dB in the editor and then the other 6 dB in the mastering stages that follow.
Firstly you open the K-20 track in an editor. You can easily hard limit all the peaks there to -3 dB and add 2 db of rms gain right there to the track without changing any dynamics.
Then the following three stages of mastering apply eg EQ, Compression and limiting. It is easy to gain 2 dB in each of those stages. None of them doing much but all adding up to a further 6 dB of gain.
You will end up with a punchy loud K-12 master in no time. You will never get it that good by suddenly altering ref levels during a project and have to recalibrate everything at that level.
In a way it is best to do everything at K-20 because you can so easily turn that into a K-16, (iTunes level) K-14, K-12 or even higher as I do with K -10 masters. And for my hip hop friends I push stuff up all the way to K-8 but not so much these days. For broadcast you can leave a K-20 master alone because it often works out to about -22 LUFS anyway.
In answer to the OP if I am working at K-14 I just generate a rough mix at K-14 and not sweat where the volume control ends up in the car. Does not matter. If I am working at K-20 then I generate a rough mix at K-20. I can quickly open the editor and hard limit everything to -7dB
(and that won't be limiting too many peaks anyway) then adding 6 dB of rms gain to simulate a K-14 master. Rough and ready but gets the job done. In the editor I sometimes add a touch of compression and limiting and add a little rms gain and it is up to K-14 very quick smart. I don't alter my recording system calibration, just how I prepare the rough mix before it is burned or copied and goes out for testing elsewhere.