Automate, Carve, Duck, or something else?

Automation of all (most) elements works for me.

Diddo... and also some judicious EQing of some of the surrounding parts can help a lot too. For instance, you may have the orchestral stuff sounding just the want you want it on its own, but you may have to make an EQ move there that you wouldn’t normally do, to help things settle in the track. Like pulling the lows or low mids back a touch or boosting somewhere in the mids or highs to help that element stand out without messing with the vocal. Or automating EQ on a few tracks just in the congested spots can help as well. But as drbill said, automation is your friend in these situations

Yeah, I did heavily automate the orchestra (and not just volume of course -  expression, etc.) - might as well give the vocal similar attention to detail.  Also realizing that I've probably push the orchestra too hard and it would probably help to pull back the entire ensemble in places. That might make it so I don't have to compress the 2-buss as aggressively.

NB: I noticed that the timpani/bass drum rolls are driven into distortion right as they approach the downbeat. I'm was going to fix, but I'm kind of liking how they fall apart as they arrive. I'm using pads and elements with distortion elsewhere so it's not like I feel obligated to have a super clean sound. It's always interesting to me when the unintended (some people call them mistakes) become a valued part of the performance/mix.

I guess in some ways a multi band comp and dynamic eq are sorta the same thing, but in operation they work differently. I use the Melda Audio Auto Dynamic eq.

As an example: you can bus all the orchestral elements and put the auto eq on that bus. Then activate the sidechain. Feed the sidechain from a send on the vocal track. Then select one (or more) bands of eq, that are now triggered by the vocals, in places that will help make room for the vocal track. In the melda plug it gives you the option of seeing a real time freq analyzer and/or spectrograph so you can see which frequencies are coming from the sidechain. You’re basically ducking the orchestra but only in specific eq bands. It’s a super smooth way of making elements gel and is sometimes easier than automating.

There's absolutely no reason to make it too complicated. If you can find what exactly is offending, just automate it down a few dB during the offending moments.

Yes,svart makes sense. just automate the vocal in the part you need more of, and automate the places where the mix is too high. There are a number of ways I guess. I would send the orchestral group to a bus, and automate the buss just enough before trying to raise the vocal. I'm not that skilled in Logic, so there may be better solutions.

I replied above, but I should have been more clear about leaving the vocal where it sits and automating the other offending tracks around it.  Most of the time if you have further compression on the whole mix, automating something up will only add to more compression congestion.

There's going to be a few instruments or tracks that are stepping on the vocals more than others, I'd just find those and automate them down a little during the sections that need it.