Another point that I think has been mentioned elsewhere: xpubs are another vector of attack. Unfortunately this one is very difficult (impossible?) to measure, so it’s hardly a critique of your quantitative analysis other than, you should probably mention this additional risk. Many services have existed that actually require users to share xpubs; in some cases it’s even been public data, but of course usually not: usually it’s a relationship between a client and a service. Such as, tax reporting software, certain blockchain monitoring services, security type services etc. Any databases can be leaked, so it’s hard to know what xpubs are in the public or quasi-public domain. But to state the obvious, a publically exposed xpub means that coins held at an address A that was never used before is not unexposed, but exposed (to CRQC threat). It’s probably a huge number of addresses in this category (that is, BIP32 on an unhardened branch, holding funds and never before used), but nearly impossible to know what fraction (probably low) of that total is exposed by the aforementioned xpub threat.
(Sorry that my comments do not address any of the questions you wanted feedback for; this was just the part that captured my interest).