Let's stick with the 500k/1M n-gram thresholds for now. The whole spec and gai families need redoing/revising, but that's something for next year. I'll probably raise an issue on it.

@@ -14 +14 @@
-<re_pri>spec1</re_pri>
+<re_pri>spec2</re_pri>

I think those are the precise thresholds Johan used to use (and I've mostly adhered to them too I think, picking them up from him)

Without definite thresholds, there's no obvious way to choose between spec1 and spec2.
I don't think it needs to be case-by-case unless A) the term is too recent for the ngrams, or B) a common term and an obscure term share the same kanji form and we want to give priority to the former (to aid lookups).
Any figure is arbitrary but does 500k+ for spec2 and 1m+ for spec1 sound sensible?

The spec1/spec2 pre-dated the m-grams by many years. They started out as tags that indicated they were entries in a certain compact dictionary and hence likely to be reasonably common. I was criticized by the compiler of the dictionary, so I flipped the tags to just "spec". Since then they've become a general "common term" tag.
We could pick on, say, 500k+ for the aggregation of n-gram counts as a guideline, but I think it should be a case-by-case decision.