Read the explanation
This page uses a fixed word lexicon to score entered text. It splits nonempty segments longer than two characters, then sums recognized token weights within each segment. Unrecognized words contribute no sentiment weight. A segment with no recognized token retains zero sentiment and point four melancholy. Three such segments therefore average point four melancholy, just as one does. At five hundred pixels per unit, both bars are two hundred pixels. This is a deterministic text heuristic, not an assessment of a writer mental state or a trained semantic analysis model. If the trimmed text equals the built in representative fixture exactly, the overall result is replaced with saved constants: melancholy point eight two and sentiment negative point three four. A line by line lexicon result can therefore differ from the aggregate fixture numbers. The two bars show melancholy point four for an unmatched ordinary input and point eight two for that exact fixture, using five hundred pixels per unit. Their lengths are two hundred and four hundred ten. This special branch is an expected output override, not evidence that the page discovered the same mood independently. No original lyrics are reproduced in this explanation. The chord family uses two melancholy thresholds: point four five and point seven. Below point four five it selects the brighter major family. At point four five through below point seven it selects the intermediate minor family. At point seven or above it selects the higher melancholy family. Values point four four, point four five and point seven are shown on one five hundred pixel per unit scale, with widths two hundred twenty, two hundred twenty five and three hundred fifty. The progression scheduler uses two beats per chord, so seventy two beats per minute means about one point six six seven seconds per chord. The JSON report captures computed state. These rules are musical suggestions, not measured emotional truths.