@jeffrey: "'ly' and 'for' are still wrong · yeah and pass is missing · there is an off by one error that continues through the line". Exactly right, and it explains every symptom at once. Counting her sung events after "just" there are ten; counting syllables in "waiting very patiently for time to pass" there are eleven. I had assigned patiently two events, pa·tient and ly. It is pa·tient·ly — THREE — so every word after it inherited the wrong event: what I pinned as "for" was really patiently's last syllable, "for" was actually where I had "time", and so on down the line, until "pass" fell past the 22.45 s cap and vanished entirely. That is why ly and for stayed confused no matter how the boundaries were nudged: the boundaries were fine, the ASSIGNMENT was off. Re-pinned one word later — for 18.54 · time 19.48 · to 21.20 · pass 23.00 — and the cap moves to 24.25 so pass is back. The proof is the melody. The per-word pitches now come out as for D#4 · time F4 · to C#4 · pass C#4 · very A#3 · patiently D#4 · just C4 · waiting C#4 — which is the lane's original harvest table, measured independently back in v1 from her raw take. A wrong assignment could not reproduce that. Two supporting fixes. Notes are RE-MEASURED over the span the chart will actually play: f0_hz came from the aligner, over the span the aligner believed in, so a repinned word kept the old note — "for" was still reading C#4 after being moved a whole word. And silence inside a word now barely stretches (weight 0.04 against a vowel's 1.0): her 0.20 s pause between "patient" and "ly" was being stretched to 0.40 s, which detached the syllable so it read as a word of its own. A pause is not a phoneme.