Hebrew RTL text fixer

Unicode text is stored in logical order (the order you'd type it), and the bidi algorithm decides how to lay it out visually. For a right-to-left script mixed with numbers, punctuation and Latin text, that means separating the string into short runs of one direction each and deciding which side each run's neighbors sit on. Most of the time it gets this right; a tool that authored, exported or copied the text with its own idea of direction can leave the runs in an order the algorithm renders wrong.

Numbers are a special case: a run of digits is always read left to right even mid-sentence, so "יש לי 3 ספרים" keeps "3" reading normally while the Hebrew around it reads right to left. When the surrounding text's direction is misdetected, the whole line can look like the numbers moved, even though nothing about the digits changed.