Yeah, that's the exact tradeoff. 108MP (or even whatever the real photosite count is that they're shift-capturing or otherwise trick-shooting to get that number) on a sensor that small is genuinely revolutionary. But only giving that sensor as much light to work with as a matchhead-sized lens can capture for it, there's no way to avoid relying very heavily on the ISP to yield an intelligible image. Again, that does an incredible job for what little it's given to work with - but doing so requires it be what we could fairly call "inventive," with the result that anywhere near 100% zoom, "suggestions" are exactly what you're seeing. The detail is as much computational as "real."
People make much of whatever Samsung it was a couple years back, that got caught copy-pasting a sharper image of Luna into that one shot everyone takes and then gets disappointed with the result because, unlike the real thing, our brain doesn't make the moon seem bigger in pictures. But they all do this and they have for years. I tried taking pictures of some Polistes exclamans wasps with my phone a couple years back, in good bright lighting with a decent CRI (my kitchen, they were houseguests). Now if you image search that species name, you'll see these wasps are quite colorful, with complex markings in shades ranging from bright yellow through orange, "ferruginous" rust-red, and black.
In the light I had in the kitchen, I could see all these colors clearly with my eyes, through the glass of the heated terrarium that was serving as the wasps' temporary enclosure. (They'd shown a distinct propensity for the HVAC registers, and while I find their company congenial, having a dozen fertile females exploring the ductwork might have been a bit much even for me...) But as far as I could get the cameras on this iPhone 13 mini to report, from as close as their shitty minimum working distance allows, these wasps were all solid yellow from the flat of their heart-shaped faces to the tip of their pointy butts. No matter what I did, even pulling a shot into Photoshop to sample pixels and experimentally oversaturate, I couldn't squeeze more than a hint of red out of anything without resorting to hue adjustments, i.e. there is no red there to find.
So all I can conclude is the frigging thing made up a wasp - oh, not in the computer vision, generative AI sense we would mean that now, or even in the Samsung sense that only works for the one subject anyway, but in the sense that even in the most favorable of real-world conditions, it's working from such a total approximation of the actual scene that, unless that scene corresponds closely enough to what the ISP's pipeline was "trained on" by the engineers who design phones' imaging subsystems, the poor hapless thing really can't help but screw it up.
This is why people who complain about discrete cameras' lack of brains are wrongheaded to do so. I see how they get there, but there are some aspects of physics that really can't be replaced by computation, including basically all the ones that matter, and the physical, optical singlemindedness of the discrete camera's sole design focus is what liberates it to excel in that realm. Just as with humans, all cramming a phone in there will do is give the poor thing anxiety.