What I'll do is:
- continue the fix-up, setting them all to "adj-ix";
- once that's done, check the counts for the 〜のよかった/〜の良かった forms, and based on that run two batch POS updates: one for "adj-f" and one for "adj-f,adj-ix".

仲の良かった	48625
居心地の良かった	1311
運の良かった	1286

True, some of them do inflect. But I still think we need to indicate that they're prenominal.
Could your script be modified to check the ngram counts for 〜のよかった/〜の良かった forms? We could set the PoS to [adj-f,adj-ix] if the counts are over a certain threshold. If the threshold isn't met, it would just be [adj-f].

調子のよかった	2672
調子がよかった	27551
性の良かった	57
性が良かった	2482

There's a bit of inflection going on with the Xのよい forms. Would it be better to leave them as "adj-ix"? I can easily flip them to "adj-f", but in cases like 調子のよい that doesn't seem completely accurate.
I'll continue creating Xのよい forms as "adj-ix" as it's relatively easy to do a bulk amendment later.

gg5
体のいい	        8498
体のよかった	No matches

As it's prenominal (and doesn't conjugate), adj-f is correct. I was going to raise this as an issue on GitHub. All the newly added 〜のいい entries should actually be adj-f. The same applies to all expressions where "の" functions as a subject marker.

@@ -26,4 +26,6 @@
-<gloss>nice way of saying</gloss>
-<gloss>diplomatic way of saying</gloss>
-<gloss>fine</gloss>
-<gloss>plausible</gloss>
+<gloss>nice-looking</gloss>
+<gloss>seemingly fine</gloss>
+<gloss>nice-sounding</gloss>
+<gloss>polite-sounding</gloss>
+<gloss>plausible (e.g. excuse)</gloss>
+<gloss>glib</gloss>

"ix" these days.

@@ -25 +25 @@
-<pos>&adj-f;</pos>
+<pos>&adj-ix;</pos>