O.K. Here are my findings.
I spent about two hours testing all sorts of scenarios with regard to the subtle elevation changes we're experiencing.
1. This Mavic Pro is an AMAZING piece of technology. I'm just shocked at how far consumer drones have come in a relatively short amount of time. Unless you are seeking "perfection" with good lighting the OPTI mode provides an amazing amount of control. I turned off the sonars to experience plain ATTI and I'm so happy I don't need to use that in close quarters ever again. Kudos to DJI on the amazing advances.
2. Even when "rock solid" - indoors, no wind, apparently no slow rise or fall, full Opti, etc. there IS going to be a subtle moving of the video frame. At least with the current technology, the "battle" between the gimble and the drone trying to be stable creates just a subtle and perceptible movement. However, it is very minor. If you need absolutely NO frame movement for video, I don't believe you will accomplish this with the Mavic Pro in the air. You need a real tripod.
3. Speaking of "Tripod Mode" - I've decided it's really a bad name. The overall stability does not change. The mode simply reduces the effects of the stick input to allow for more subtle pans and slides. So don't think that tripod mode is actually making for a more stable video platform.
4. I noted during my tests that unlike previous times when I noted my Mavic slowly descending, I was consistently experiencing slight rising of the drone. About 3 inches in a minute was what I measured. For 99% of the applications (I'm guessing) this slow rise would be considered irrelevant. But it is there - at times. Sometimes it went away almost completely and all I can think is that the temperature / barometric pressure and sonars found a happy medium.
5. I'm guessing that the sonar software tries to lock the "current" position rather than estimating the current height and trying to maintain that current height. This is a subtle but significant difference. If this is in fact what it is doing, then there is hope for a firmware upgrade to help solve this "elevation drift" - but only when relatively close to a flat surface. So, in short, the Mavic should take a ground distance measurement and "lock" that (unless there is stick input) and try to maintain that distance rather than the "current" height.
6. Even if #4 was a "distance lock" - the sonar clearly has a variance in measurement that I'm guessing is about + or - 2.5 inches at maybe 10 feet. Obviously that variability increases and decreases with overall distance from the ground. I don't think the sensors being used could measure much more precisely than that.
7. I'm unclear about the barometric pressure being used for elevation control. If it is being used in conjunction with sonar, then that might explain some of the elevation drift issues if those readings don't precisely agree (which they likely won't most of the time). But the reality is that barometric pressure changes and the air and Mavic temperature could affect that measurement. It "seemed" to me that as the Mavic warmed up, the elevation drift moderated, but I was getting sporadic results in that regard.
So, in short, I think we need to live with a bit of elevation drift. Elevation is the poorest GPS dimension (and intentionally designed that way), the sonar is not super precise - especially the higher you get, and barometric pressure naturally deviates - especially with temperature variations. If you need a rock solid video platform where elevation drift is not acceptable, I don't think the Mavic Pro is your solution - but I also doubt any other flying drone will provide that without significantly more video stabilization built in.
Hope this helps.