split('') breaks the string into UTF-16 code units, reverse() flips the array in place, and join('') glues the code units back together. For pure ASCII, this is correct, fast, and exactly what every interview question expects. The hidden bug is that any character outside the Basic Multilingual Plane (most emoji, mathematical italics, several historical scripts) is stored as a surrogate pair of two code units, and naive reversal swaps the high and low halves, producing invalid UTF-16. Reach for this only when you can guarantee ASCII input, like reversing a number for a palindrome check.

The spread operator on a string iterates by Unicode code point (it uses the string's [Symbol.iterator]), so each surrogate pair stays as one entry in the resulting array. Reversing that array and joining preserves multi-code-unit characters like the globe 🌍 (U+1F30D) and mathematical bold A 𝐀 (U+1D400). This handles 95% of real-world reverse-a-string needs cleanly with no extra dependencies. The remaining 5% is grapheme clusters, where one user-perceived character is actually multiple code points (combining accents, ZWJ-joined emoji, regional indicator pairs for flags), which the next accordion handles.

A grapheme cluster is what a human sees as a single character: a base letter plus its combining marks, a country flag (two regional-indicator code points), or a multi-person emoji joined by zero-width joiners. [...str] still splits these into separate entries, so the spread version reverses inside the cluster. Intl.Segmenter with granularity: 'grapheme' walks the string in user-perceived characters, which is the only way to reverse without breaking emoji families or detaching combining accents. It is available in modern Node and all current browsers, and the cost is one allocation per call. Use it whenever the input is user-facing text in any language.