A useful novel-proofing comparison asks whether an editor can preserve the reasons for all 17 known items, not how many warnings it can produce. Prepare three copied chapters with 12 change candidates, three deliberate dialogue exceptions, and two pending decisions. Give every item an ID before opening a candidate. Score whether the candidate helps you find, classify, resume, and change one rule without disturbing the other 16 rows.
This 17-item set is a comparison sample, not a measured product result. Automatic literary-quality judgment, collaborative proofreading, and price ranking are outside the article. Operational search and replacement and same-word highlighting are separate workflows. This article focuses on editor selection under one controlled test.
Start with a 17-row decision ledger
An editor that reports 20 problems is not automatically better than one that reports 12. A novel can contain intentional pronunciation, historically appropriate spelling, or dialogue that breaks a narration rule on purpose. Give every row one outcome: change, keep as an exception, or pending decision. Stop when a candidate loses an item or changes an exception without its reason.
Use columns for item ID, chapter and line anchor, observed form, expected classification, editor result, action, reason, and restart result. IDs V-01 through V-12, E-01 through E-03, and P-01 through P-02 expose a missing row. Write the expected classification before opening the candidate.
Put 12, three, and two items in different contexts
Do not repeat one word 12 times. Include two forms of a character name, Arabic and full-width numerals, two punctuation styles, a spacing difference, and a term standardized in narration but retained in dialogue. This tests whether a result includes enough context to support a decision.
Give the three dialogue exceptions a character and scene reason. Give the two pending rows no forced answer; success means returning to the same passage later. Duplicate the same three chapters for every candidate and retain one untouched source.
Score finding and classification separately
Run the same search from chapter one through chapter three and mark whether all 17 IDs were surfaced. Then record whether each result distinguishes narration from dialogue and the working copy from an old draft. A hit without a reliable return point cannot support classification.
Apply only the 12 pre-approved changes. Preserve E-01 through E-03 and leave P-01 and P-02 unresolved. Re-run the condition. The 12 changed forms should leave that result set, while the three exceptions and two pending rows should still lead to their intended passages.
Resume with exceptions and pending rows first
Before closing the session, record the last item ID, chapter, and next pending action. After reopening, check the reasons and locations for E-01 through E-03, then the unresolved state of P-01 and P-02. The condition is restoration of decision authority, not accidental return to the same window position.
If an exception existed only in a temporary filter or the operator’s memory, mark it lost. Opening an old draft, treating a pending row as solved, or returning an exception to the ordinary change list also stops the candidate. If the editor does not store reasons, record an external ledger as a condition of use.
Move only P-01 on the second pass
Change P-01 from pending to change, write its expected form, and repeat the search. The candidate passes when P-01 follows the new rule while the 12 completed changes, three exceptions, and P-02 remain stable.
Classify failures as wrong source, lost exception evidence, overly broad replacement, or incomplete restart notes. Do not turn every observed failure into a vague usability score.
Return Rune Studio’s confirmed scope to the comparison sheet
Rune Studio documentation describes Japanese long-form editing, workspace search, manuscript organization, preview, and EPUB output. Observed behavior covers only part of that scope: bounded search, three replacements across two files followed by reopening, tree search, horizontal EPUB generation, and preservation of the link between a vertical-writing tab and its preview.
The 17 classifications, a visible manuscript-preview check, persistence of exception reasons, restart of all 17 rows, and the second pass that changes only one pending item remain unconfirmed. Rune Studio therefore remains a candidate for this test, not a product that has passed this sample. Run the full copied manuscript through both passes, reopen it, and record the reason beside every unchanged exception before deciding. Repeat the same search after reopening so that an item is credited only when its location, classification, and explanation can all be recovered.
Choose the editor that preserves the reasons
For the query “recommended text editor for novels proofreading,” choose the candidate that keeps the reasons for V-01 through V-12, E-01 through E-03, and P-01 through P-02 after reopening. A controlled rule change should move only P-01 and preserve the other 16 rows.
Begin by placing IDs and expected classifications beside the 17 known cases in a copied manuscript. If Rune Studio is included, check the Rune Studio product page and begin the unmeasured exception, visible manuscript-preview, and restart rows as pending.