My 5G home internet dropped up to 97 % of IPv4 packets while IPv6 over the same cell ran clean, and one TCP flow pushed 153 Mbit/s while its neighbors starved. A measurement writeup: what the data shows, what it rules out, the workaround that proves the point, what changed when o2 touched the network on July 16, and what five weeks of continuous monitoring afterwards did and did not settle.
Anatomy of a broken IPv4 path: flow-dependent packet loss on o2's network