You can only test. 
I’d say this: CDs have been pushed for over a decade. MP3s are polluted by the damage, but reconstruction is as much a software issue as hardware, so an intelligent transcoding algorithm can resolve this, and I bet it’s far less of an issue than it once was; once files are transcoded to 32 bit float space, the peaks can be very easily scaled into range without artifacts.
Of course actual clipping is still clipping. Talking about a single missing peak sample between a pair of FS values located 1/44,100 sec apart is fairly trivial in most band-limited programs, but as soon as you have 3 consecutive 0 dBFS samples in a row, you’re missing relevant data about both peak and slope which cannot be recovered. In these cases no hardware or software can restore your program to it’s original sound, and that’s what others here are referring to above.