Editing Openai/69186d20-2d14-8013-b9c6-c09000f173f6 (section)

===== Given a real Italo-Celtic stage, the hierarchy is roughly: =====
# Shared innovations in morphology and sound law - These must exist and are the primary evidence: superlative morphology, certain case/tense formations, particular sound changes. - This is what people cite when they argue “Italo-Celtic” at all.
# A smallish set of lexical isoglosses - A few roots that look like: PIE > Italo-Celtic innovation > survive only in Italic+Celtic. - But you don’t expect hundreds in the basic lexicon; you expect a handful and they’re often semantically shifted or obscure.
# No strong prediction about “feels like more cognates” - Because: - Celtic’s later phonology scrambles transparency, - Germanic–Italic contact adds spurious “extra closeness”, - and lexical replacement is fast and messy compared to morphology.

So: your conceptual point is right—a genuine subgroup should have some layer of extra shared retentions—but in practice that layer is small and heavily obscured, and doesn’t manifest as “I see more obvious cognates in my Duolingo vocab.”

If you want to go more rigorous, the next step would be: look at a published list of proposed Italo-Celtic lexical isoglosses, then separately at Italo-Germanic ones, and look at which branch of the tree those actually support once you control for PIE retentions and later contact. That’s the level where your retentions intuition can be made precise.