Entity (non-alias) sort name style contains several provisions about how stylized Latin script characters should or should not be used.  However, it does not cover Japanese wordplay - specifically, the reading/speaking/sorting of Arabic numerals in Japanese text depends on artist intent.  As a result,

1. CDs in a shop (Japan has many physical CD stores still) would store the Japanese wordplay with the Japanese alphabet, not in an Arabic numeral / other characters section.
2. People speaking the names should say it the same way it is sorted in Japanese.

In MusicBrainz, there is a currently a mix of incompatible entity sort name strategies:

(A) Transliterate the non-Japanese characters that are supposed to be read, pronounced, and sorted as a Japanese character

(B) Do not change the Arabic numeral

I think a guideline should be added to use (A).

What should be non-controversial examples for this guideline of Arabic numerals that are not used as numbers:

1. 765PRO ALLSTARS is read, spoken, and sorted as ナムコプロ オールスターズ, so it should have the entity sort name "NAMCO PRO ALLSTARS".
2. 96猫 is read, spoken, and sorted as くろねこ, so it should have the entity sort name "Kuroneko".

The guideline could be further extended to additionally cover Arabic numerals that are used in numerical contexts.  For example:

• 22/7 is read, spoken, and sorted as ナナブンノニジュウニ, so it should have the entity sort name "Nanabun no Nijuuni".

Note that whether 0/2/3 of these examples are adopted for guideline or not, Alias sort name style says that Japanese is to be sorted the way Japanese would expect:

The sort name should be in the appropriate form for the alias locale, since different languages sort names differently (for example, use kana for Japanese sort names).

Therefore, the correct Japanese alias sort names in MusicBrainz already do not equate with the legacy Latin-script entity sort name.