Two steps. The check runs in your browser. Like my scorer, it reads each letter by its code point after NFC. Unlike it, it looks only at ş ţ ș ț, and it skips ş in sentences that look Turkish.
Or any Romanian text: a web page, a database export, a translation.
The fix swaps ş ţ for ș ț sentence by sentence. In a sentence with ı, ğ, ç, ö or ü it leaves ş alone, so most Turkish keeps its ş, but a short Turkish sentence without those letters, like "Hoş geldiniz!", still gets changed. ţ is always swapped: Turkish has no ţ. It's the fix_romanian() from the post.