A famous exception to the ‘phonetic spelling system’ of Welsh is the use of <y> for both /ǝ/ and the retracted high vowel /ɨ(:)/. This double use of <y> was almost universally adopted by c. 1330, when a grammarian labelled /ǝ/ and /ɨ/ as ‘dark y’ and ‘clear y’ and illustrated them with polysyllables such as ystyr /ˈǝstɨr/ ‘meaning’, in which the value of <y> was predictable from the position of <y> in the word. At that time the three-way system of urn:x-wiley:00791636:media:trps12205:trps12205-math-0001 for /i(:)/, ‘dark’ <y>, and ‘clear’ <y> was two centuries old, being first attested in Braint Teilo (‘The Privilege of St Teilo’), c. 1130. Yet the ‘Teilo’ system is rarely attested before c. 1300; instead all three phonemes might be represented by urn:x-wiley:00791636:media:trps12205:trps12205-math-0002, as commonly before 1100, or by <y>; or <e> might be used for /ǝ/ and/or for /ɨ(:)/, as had sometimes occurred in Old Welsh as well. This article argues that one reason, apart from scribal conservatism, for the delay in adopting the ‘Teilo’ system was its failure to distinguish the value of <y> in proclitics such y /ǝ/ ‘the’ and y /ɨ/ ‘his/her’ and ‘to’. For this the ultimately abortive ‘Caligula’ system (c. 1250) had offered a solution.