Text Case Converter

Seventeen case styles, and the hard part of every one of them is deciding where a word ends.

The rules are the five `src/lib/csv/naming.ts` states, reused here rather than restated: anything that is not a letter, a digit or a combining mark is a separator and is removed; a lower case letter followed by an upper case one starts a new word; in a run of capitals the last one before a lower case letter starts a new word, so XMLParser splits as XML then Parser; a digit followed by a capital starts a new word, so col2Name is two; and a letter next to a digit is not a boundary, so version2 is one.

The version this replaced got the two capital rules right and everything else wrong. Only a dash, an underscore, a dot and a slash were separators, so every other character stayed exactly where it was: `First Name (US)` came out of snake_case as `first_name_(us)` and `Price ($)` as `price_($)`, neither of which is snake_case. It had no digit rule, so `col2Name` became `col2name`. Measured against a 22 input corpus on 2026-09-03: 278 of 352 conversions were correct; after this pass, 352 of 352, through the browser.

Four other things changed. The URL slug used `[^\w\s-]` to strip punctuation, and `\w` is `[A-Za-z0-9_]`, so it deleted every accented letter along with its accent: `Café Münster` became `caf-mnster` and a Japanese place name vanished entirely. Accented letters are kept now, which RFC 3987 allows in a URL, and there is an ASCII option that removes the accent FROM the letter, giving `cafe-munster`. Alternating and inverse case ran over UTF-16 code units, so an emoji counted as two positions and the alternation after it was one out; they run over grapheme clusters now. Sentence case capitalised after every full stop, so `dr. smith arrived` became `Dr. Smith arrived`; it uses the sentence rule from the Word Counter now, which knows an abbreviation, an initial and a decimal point from the end of a sentence. And random case called `Math.random` during render, so the value changed every time the component redrew; it is seeded on the text now, so the same text gives the same answer.

Title case is the one that has no single answer, and the version this replaced hid that behind an unattributed list of eighteen small words. There are three published rules and they disagree: the Chicago Manual of Style (17th edition, 8.159) lowercases every preposition however long, so it writes "A Portrait of the Artist between the Wars"; the AP Stylebook capitalises any word of four letters or more, so it writes "Between"; APA follows AP and adds the word after a colon. All three are offered, the one in force is named on the page with its source, and the words it keeps lower case can be expanded and read. A word that is already all capitals is left alone, so US does not become Us, unless the whole text is capitals and there is no way to tell an acronym from a shout.

Every example printed under a style name is computed from the same input, `First Name (US) v2`, under the settings in force, so it always shows what that style will actually do rather than a string typed next to it that might have drifted.