Okay, First shot of Balloons taken 20+ years ago, with the exception of this year’s sensors probably outside the dynamic range of digital, so then one is looking at bracketed exposures for HDR which would have motion blur. ETTR therefore not really applicable, though it could be argued that for the lightest tones bracketed it is. I strongly suspect that lab tested dynamic range for the darkest tones visible would not translate well in real world dynamic range in low light without substantial noise in the darker tones.
Second shot of the woman, ETTR would be applicable if the percentage of highlight tones was visible on the in camera histogram, but I would sway towards ETTL to make sure the dark tones were kept and no lifting was required or run the risk of chromatic noise.
The third photo is an example of ignoring ETTR for effect.
Camera meters are great, 90% of the time they work and give a OOC Jpg that people are content with, ETTR works on RAW 99% of the time. I fully admit that these are edge cases, but as a former regional news photographer that had to do 10 jobs a day in whatever the weather conditions and time of day (no luxury of the golden hour), the edge cases came up a lot more of the time.
Having to use Velvia with only 4~5 stops of range made choosing the amount of highlight detail and discarding other tones, second nature. My general advice is to take lots of pictures, until one is actively immersed in the subjective and more into visualisation of the finished print, it’s too easy to get bogged down in getting a full range of tones in the RAW only to end up dodging out and burning in details that would otherwise distract from the main subject matter.