Select an academic text editor with an acceptance table, not an “autosave included” label. The sample contains 25,000 characters, 18 footnotes, 44 citations, and two Shift_JIS conversion candidates. Published specifications, static source/localization alignment, and existing CLI manual-write evidence are kept separate; an unobserved autosave event is not marked as a pass.
Treat references as data, not decoration
In a Markdown-based paper, a note might use [^1] and a citation might use [@sato2024]. If brackets, punctuation, or a key changes during conversion, the prose may remain readable while the note or bibliography link stops working.
Before changing encoding, record three counts: 25,000 manuscript characters, 18 footnotes, and 44 citation keys. Also look for duplicate note identifiers and citation keys that do not appear in the bibliography source. These checks make the article specific to academic writing rather than just another large-file save routine.
Decide two character substitutions explicitly
If two characters may not survive the destination encoding, place them in a decision table. Record the source character, approved substitute, location, and whether it occurs inside a note identifier or citation key. Make the substitution only in a duplicate, then save a file such as paper-utf8-safe.txt.
Reopen that file and compare the manuscript count, unconvertible list, footnotes, and citations. Do not let a matching overall length conceal a changed key.
Match 25,000/18/44/2 at three checkpoints
The dedicated academic sample produced this table:
| Check | Before manual write | After manual write | After reopen |
|---|---|---|---|
| Characters | 25,000 | 25,000 | 25,000 |
| Footnotes | 18 | 18 | 18 |
| Citations | 44 | 44 | 44 |
| Shift_JIS candidates | 2 | 2 | 2 |
| Unique added marker | 0 before edit | 1 | 1 |
This is a count audit, not a semantic proof. It establishes that the note and citation patterns did not disappear during the manual save route. It does not prove that every note points to the correct passage, every citation points to the correct bibliography item, or the citation style is valid.
Accept or hold the autosave candidate with the same table
The evidence has three layers. The published specification (Stage 3) describes Japanese-encoding detection, an unconvertible-character indication, and an autosave pause while a problem remains. Static source/localization inspection connects corresponding logic and message resources. Existing CLI/output verification (Stage 4) establishes 25,000/18/44/2 before manual write, after write, and after readback, plus one unique marker.
| Acceptance item | Pass condition | Current evidence | Decision |
|---|---|---|---|
| Manuscript | Read back 25,000 characters | Manual route matched | Manual route passes |
| Footnotes | Keep 18 | Three checkpoints matched | Manual route passes |
| Citations | Keep 44 | Three checkpoints matched | Manual route passes |
| Conversion candidates | Retain and decide both candidates explicitly | Three checkpoints matched | Manageable |
| Autosave | Confirm destination, state, and readback without manual save | Specification and static alignment only | Unobserved; hold |
Static alignment does not prove rendering or runtime firing. Autosave was not fired in this verification, and no saved-state indication was observed. Keep the candidate on hold for the paper-protection route and use manual saves plus versioned copies until the open row is completed.
Separate documented safeguards from the measured manual write
Rune Studio’s current Mac documentation describes major Japanese-encoding detection, indication of unconvertible characters, and pausing autosave while a conversion problem remains. Static inspection also found corresponding logic and localized message resources. Neither evidence layer is a runtime observation of red marking or autosave firing.
The measured operation was a manual write and reopen with 25,000 characters, 18 footnotes, 44 citations, and two Shift_JIS candidates preserved as counts. Autosave firing and visual red marking were not observed. Semantic reference correctness, citation-style validity, damaged-data recovery, OCR, and binary repair were also outside scope. Keep manual saves and versioned copies until your own autosave trial is complete.
Count the manuscript’s relationships
An academic file is acceptable when its text and reference patterns survive together. The manual route matched 25,000/18/44/2 before write, after write, and after reopen. Autosave remains unobserved, so hold that candidate and retain manual saves and versioned copies until destination, saved state, and readback are established.
The current Mac app’s Japanese-text safeguards are described on the Rune Studio website.