fix(tokenizer): stop partial furigana readings from marking words known (#137)

This commit is contained in:
sudacode authored and GitHub committed 2026-07-05 17:02:44 -07:00
1 parent eef4500599
commit b14f977e33
5 files changed
+150 -2

No files matched your search

@@ -0,0 +1,4 @@
type: fixed
area: overlay
- Fixed never-mined compound words (e.g. 待ち合わせてる) being highlighted green as known: subtitle tokens now carry complete readings instead of kanji-only furigana joins, and the known-word reading fallback rejects readings that don't cover the token surface. Stored word readings in the stats database are no longer truncated for new lines.