that requires LLM as judge

no it doesn't, you just diff against the real source code. probably something more fuzzy/continuous than actual diff, but still

Besides functional equivalence, a significant part of the value in neural decompilation is the symbol (function names, variable names, struct definition including member names) it recovered. So, if the LLM predicted "FindFirstFitContainer" for a function originally called "find_pool", is this correct? Wrong? 26.333% correct?

Proving that two pieces of code are equivalent sounds very hard (incomputable)