Bug or feature?

Example:

^(?:[^a-zA-Zs(]+)(?: · [^a-zA-Zs(]+)*K[a-zA-Z]+ (?=(?:(?: · [^a-zA-Zs(]+[a-zA-Z]+ )* 【 [^ ]+(?: ·[^ ]+)* 】)?rn)

will match the transliterated parts “shibashiba”, “dare” and “hakkiri” after the initial sequence of hiragana characters in these texts:

しばしばshibashiba 【 屡々 ·屡屡 ·屡 ·数数 ·数々 ·数 】

だれdare · たれtare · たta 【 誰 】

はっきりhakkiri

Those parts are properly highlighted by the search.
When I attempt to replace the matched parts by anything, no replacement happens, and the search skips to the next match. Even if I attempt to replace by empty string.

All characters in the example are within the Unicode Basic Multilingual Plane.

Do I misunderstand something?

I understand that I can’t use a lookbehind

(?<=^(?:[^a-zA-Zs(]+)(?: · [^a-zA-Zs(]+)*)[a-zA-Z]+ (?=(?:(?: · [^a-zA-Zs(]+[a-zA-Z]+ )* 【 [^ ]+(?: ·[^ ]+)* 】)?rn)

because the part

^(?:[^a-zA-Zs(]+)(?: · [^a-zA-Zs(]+)*

is variable length.

Both the part discarded by K and the lookahead after the match are necessary to avoid matching other lines of text. This pattern does match all variations I show above, and it also works well to avoid matching other parts of text, e.g. “(adv) often; again and again; frequently →Related words: 度々”.

Notepad++ v8.6.5 (32-bit)