If you've only run into two images in two months, that's probably as good as you can hope for as when you would choose RC over P is going to be subjective... as is the algorithm. It is based on how much of the image is out of gamut and where the out of gamut areas occur. More weight is given to the center for example so it would allow more out of gamut near the edges than it will allow in the center. It's also a tradeoff between losing a little in out of gamut areas with RC versus desaturating the entire image (even in gamut colors) with P so that's a subjective threshold as well.
I will say that if nothing shows out of gamut, then nothing is really out of gamut BUT the new LCMS 2.17 color management engine does fix/improve the out of gamut warnings and that will be in QU 2025.102 which is in the very final stages of testing and I hope to get out tomorrow. So be sure to retest when that is out.
P.S. To be honest, I just use RC all the time unless I see a specific problem with colors and that usually only happens with images that are filled with bright saturated colors. With RC at least you know if the color can be rendered by your printer (it is not out of gamut) it will be rendered accurately. With perceptual, literally nothing is rendered accurately: even the in gamut colors that could be printed correctly are shifted. It fools our eyes because it's a relative shift of the whole image but I just like knowing that my printer rendered everything it can render accurately and I don't mind out of gamut colors being a little clipped. That's usually less objectionable than desaturating the whole image a little like perceptual does. That's pretty much how the auto intent works: it won't "give up" on RC unless there's quite a bit out of gamut that might produce objectionable banding or clipping when using RC with out of gamut colors.
Regards,
Mike