RAW vs compressed RAW

If the firmware wanted to, it could actually make a lot of raw images even smaller than loss-y cRAW without any loss, but the highest compression levels take a significant amount of time to do, and that would slow the fps considerably. I remember once converting cr2 files to uncompressed DNG, where the raw data sits inside DNG as an uncompressed TIF, and used 7Zip to compress the images the best it could, and some were down to around 20% the size of the compressed cr2.
"Exposure to the left", if one is comfortable with the possible extra read noise, makes images easier to compress, since it leaves the most significant bits highly contiguous in the image, in the same color channel. When you "expose to the right" the least significant bits are all very busy with noise.

One of the workarounds for lossless is to make 14 bits an option, especially at high ISO settings. The vendor knows where the noise is.

Yes, using a large number of raw values is incidental, not causal, when higher analog gain is used for higher ISO settings. It is the analog gain (or higher conversion gain, when applicable), not the bit depth, that causes lower exposure-referred read noise at higher ISO settings that use more analog gain. So, high ISO settings could have an option to drop the bit depth some, without increasing exposure-referred read noise. In experiments I conducted many years ago, I found that if you quantized the raw values such that the read noise standard deviation was still at least 1.2 or 1.3 in the new, quantized levels, there was no increase in visible noise in the demosaiced image. That was with older Canon CMOS sensors with significant banding noise, where I needed 1.2 or 1.3, but other people have tested the practical limits of quantization in sensors with more spatially random noise, and even 0.8 was not problematic (I think that it was Jim Kasson and Jack Hogan that did those).

The potential of such quantization as an option is for smaller or similar raw file sizes at higher ISOs, rather than the much larger ones that we usually get with compressed raws in historical cameras.

Good reference...thanks! I have been using CRAW and am pleased with the results. I like the smaller file sizes, moves everything along a bit faster and of course takes less space.

Storage space is cheap, so I always go with RAW. This also has the benefit that you can get more out of the data once better RAW processors are available in a decade or so.

After lots and lots of studies, it was not visible to me. If you do the studies yourself, I recommend tripod captures. Do 3 captures with raw and 3 with craw. Do several different scenes/subjects. Then do blind pixel peeping comparisons.

I didn't do 'lots' of studies, I did a comparison at some point. In craw, the deep shadows lose some detail, also the noise stops being purely random which isn't good for noise reduction. The loss is minor, yes, but to me the reduced file sizes is not worth it. I used craw when shooting concerts and being afraid of the shoot not fitting into the memory cards.

Here's a good third party comparison (already posted above):
www.the-digital-picture.com/Canon-Cameras/Canon-C-RAW-Image-File-Format.aspx

Deep shadows lose on quality, and I don't want even those minor losses when shooting landscapes.

Did you actually look at the examples and read the analysis in your citation? As explained, noise is random so you can expect to see some minor differences. Again I suggest you try the experiment for yourself. If you do, shoot 3 images for each scene with raw and the same scene with 3 images for craw. Try a few different scenes and then no blind comparisons. When I tried this, I saw some slight differences at the pixel level but raw was not consistently better than craw.

Yes the IQ degradation is minor. The noise in craw doesn't look totally random, again it may be bad for noise reduction algorithms.
There is definite lots of information in the lossy craw compression, purely by definition. Why lose it?

Normally I get around 20-50 shots from a landscape photoshoot. The file sizes don't matter at all.

When I shoot concerts/action, I may get up to 2000 shots in a few hours, this is where I use craw sometimes. However, because I use ISO invariance, it implies pretty heavy lifting of the exposure slider in Lightroom, so I'm still avoiding craw if possible.