While trying to figure out some benchmark run to run fluctuation I improved the compare feature so it shows the compile name and level, which may have worked in the past but has been broken for some time, here are examples:

Orig:
java.lang.invoke.MethodHandleNatives::refKindIsField (32 bytes) () 1 1
org.jfree.util.ObjectList::indexOf (6 bytes) () 1 1d
org.jfree.chart.axis.NumberTickUnit::<init> (25 bytes) () 1 1d

New:
java.lang.invoke.MemberName::getInvocationType (56 bytes) (c1(3)) 1 1
org.jfree.data.xy.XYDataItem::<init> (29 bytes) (c1(3)) 1 1
sun.java2d.marlin.DHelpers$PolyStack::ensureSpace (115 bytes) (c2(4)) 1 1

This extra info helped me find a source of run to run variation.
Also I added a new test case to exercise this feature.