Raised in the jdk-dev ml:
https://mail.openjdk.java.net/pipermail/jdk-dev/2020-September/004727.html

---
For scripts Deseret, Osage, Old Hungarian, Warang Citi,
Medefaidrin, and Adlam, for strings with upper- and
lowercase variants of the same letter, the following
code fails:

Collator collator = Collator.getInstance();
collator.setStrength(Collator.PRIMARY);
assertThat(collator.compare(lower, upper)).isEqualTo(0);