The safest way to switch a Japanese text editor interface without changing the manuscript is to grade two independent lanes. The selection lane records the exact characters chosen at a Japanese–Latin boundary. The preference lane records the stored language value and its restoration. Neither result is allowed to certify the other.
The retained visual anchor shows 東京Station前 with only 東京 selected and a two-character selection count. Its SHA-256 is 4c06a40a283bcf147c5e07d54a6f74e0a00870166f331ffc1fe10f4c7ebb73f2. A separate command log records ja → en → ja for the language preference. It does not show the selection repeated under the English interface, and the screenshot does not prove that the preference was switched. That separation is the direct answer for a Japanese text editor switch UI language workflow.
Define the one-action character boundary first
Before touching the interface, write the expected range as literal text. The controlled string is 東京Station前; the expected range is 東京; the expected length is two characters. “Select the Japanese part” is too vague because an operator may include the following Latin word or the final Japanese character.
The one-action condition also needs a fixed meaning. Grade the first selection result before any drag correction. If an operator adjusts the highlight afterward, record the extra action separately. Otherwise a convenient manual repair can hide a boundary that is consistently too wide or too narrow.
There is no universal linguistic boundary for every editing task. A name replacement may need only kanji, while a phrasing correction may include okurigana or punctuation. The test asks whether the first selected range matches the unit that the editor defined before the attempt.
Read the two-character result from the retained screen
Opening the source image at full size shows a purple highlight over 東京 in 東京Station前. The Latin Station and the following 前 are outside the highlight. The lower status area visibly identifies the selected range and reports two selected characters.
The receipt records the source path, digest, complete mixed-script input, visible selected text, and visible count. Recomputing the digest matters because a cropped derivative might omit the status area or neighboring characters that make the boundary interpretable.
This observation is intentionally singular. Other rows in the screen contain numbers, kana, punctuation, and Latin text, but they are not selected. The image does not prove that every Japanese script class follows the same algorithm, that the result repeats from multiple pointer positions, or that the selection was produced with one particular click sequence.
Audit the language preference on a separate track
The preference evidence begins with a settings export, reads ja, sets en, reads back en, imports the backup, and finally reads ja. The retained values are ja → en → ja; the backup is 2,580 bytes with SHA-256 7dc99b88717d7d431116f0e28960ca4f900bb66068cd8148a5aba6bc374fc177; the import reports 77 keys.
Those are stored-value facts. They do not establish which menus changed after a full restart, whether every notification was localized, or whether the two-character selection behaved identically while en was active. The screenshot, in turn, shows one selection result but not the preference transition.
Keep one receipt for the exact selected range and another for the preference round trip. An en readback cannot fill the selection field. A visible two-character highlight cannot fill the restored-preference field.
Leave five script-boundary rows explicitly unmeasured
A production comparison can contain six rows: kanji–hiragana, hiragana–katakana, Japanese–Latin, Japanese–digit, word–punctuation, and markup–prose. Each row should specify its expected range, first result, over-selection, under-selection, and correction count.
Only the Japanese–Latin row has a retained visual outcome here. Enter 東京, two characters, and the screenshot digest in that row. Leave the other five rows blank until an operator performs the same controlled action on the target build. Visible but unselected sample lines are not test results.
If the team needs a Japanese-versus-English interface comparison, duplicate the six rows into two named columns and capture both sessions with the same application and OS versions. The existing records do not provide that paired screen experiment, so this article presents it as the next acceptance task rather than a completed claim.
Recover selection and settings failures independently
If the first highlight expands to 東京Station, classify it as over-selection. If it stops at 東, classify it as under-selection. Do not toggle the interface language to repair either result. Return to the unchanged mixed-script specimen, preserve the expected range, and repeat under a documented starting position.
If the preference does not read back as en, or fails to return to ja, do not modify manuscript text or encoding. Save the error, import the fixed backup, and read the value again. A settings failure has a settings recovery path.
If a visual exercise changes the manuscript, discard the disposable copy and recover from its known digest. Separating selection, preference, and content recovery keeps one failure from being masked by an unrelated fix.
Approve with two receipts, not one success label
The selection receipt names 東京Station前, expects 東京 and two characters, and points to the image with digest 4c06a40a…73f2. Its failure categories are over-selection, under-selection, and an unexplained variable range. Its acceptance condition is the exact first range defined before the action.
The preference receipt starts at ja, records an en readback, imports the backup, and closes at ja. Restarted English menus, notifications, IME behavior, and selection under both UI languages stay in the unmeasured field.
For a Japanese text editor switch UI language decision, do not ask one result to carry both promises. Verify the character boundary from the selected text, verify the preference from its round trip, and schedule the missing paired-screen test separately. That protects the manuscript while making the next decision visible.