Unicode offsets

All spans are zero-based and half-open: start is included, end is excluded. They count Unicode code points, not UTF-8 bytes, UTF-16 units or grapheme clusters. For က😀, the emoji spans [1, 2) although JavaScript and Dart use two UTF-16 units to represent it.

const span = Array.from(result.input).slice(start, end).join('');

Use the input attached to the response. Normalization/conversion can change length or ordering and do not provide an offset map. Transliteration spans reference input, not Latin output.