While posting an answer to this question, I looked back at an answer I'd written on another site, which mentions the strange case of Latin diēs < PIt *diēm.
In particular:
The accusative form of [the Proto-Indo-European root behind Zeús and Juppiter], *dyēm (cf AGrk Zēn), also survived in Latin, in the form *diēm > diem. Uniquely, the dy here didn't become *dj > j; some scholars suggest that it evolved in a different Italic language/dialect, which didn't have the palatalization rules, then got borrowed back into Latin.
But, to my shame, I can't seem to figure out who these "some scholars" I cited actually are: I recently lost my web history, and with it all records of which particular articles I'd been reading.
And so I'm asking here: how did *diēm avoid the palatalization that created Jovis? I was under the impression that this palatalization happened back in Proto-Italic times, before Latin had split off; was there actually some Italic language that avoided the process entirely, from which Latin could have borrowed an unpalatalized *diēm? And if not, where could this notorious exception have come from?