Every Unicode letter-or-number token is lowercased, then its first cased letter is uppercased. Small words receive the same treatment as every other token.
Punctuation separates tokens
Spaces, hyphens and underscores are preserved and end one token, so the next letter-or-number run starts a new token.
Names still need a human
Internal capitals in names and acronyms are lowered by Title Case. Capitalized Words is less destructive but still cannot know which spelling is correct.
Check locale and destination behavior
The active interface locale is part of the case-mapping contract, while the tool never detects the language of pasted text. Repeat a representative sample containing accents, an acronym, a proper name, a digit, punctuation, and an emoji. Compare the visible output and both code-point totals. Then paste into the real editor and check whether that editor applies normalization, autocorrection, smart punctuation, identifier rules, or another rewrite after CaseShift has finished.