The way the "koma" is placed above 困っ makes me suspect they first tokenize the text, get furigana for each segment (most tokenizers have this built-in) and then derive the romaji without correctly handling the sokuon っ when it appears just before a segmentation boundary. Usually it's represented in the romanization by doubling the following consonant, but of course that doesn't work if there's no following consonant because each token is treated as a separate string. They do get it right for "erikku" where there's no segmentation boundary.