REJECTED folk-etymology: morpheme_decompose('tupa') returns source='segmented' (greedy-longest FALLBACK, NOT a lemma), yielding tu (num 'two') + pa-head (n 'head') — there is NO 'tupa' morpheme entry in mdb, which is exactly why the DB falls back to greedy splitting; this confirms rather than supports a tu+pa analysis. No semantic path from 'two'+'head' (or 'two winters') to 'desire/heed' exists, and the slot-IV pluractional -pa would force a never-attested *tu-re alternation. The string tu+pa survives only in the unrelated salmon noun. Therefore the verbal root tupa is treated as synchronically atomic / bound (near-cranberry), with the apparent internal structure a false-segmentation artefact.