Finding two spellings is not permission to replace one everywhere. A novel may have a style rule for one pair, a character-voice exception for another, and an unresolved proper name that requires an external source. The safe process records candidates, counts, context, approved form, and exceptions in a search ledger. Replacement happens only after classification, a backup, and a dry run. Preserved exceptions also need reasons, so that the next proofreading pass does not reopen the same decision. Matching counts and correct meaning are separate acceptance tests. This guide turns search and replace into a reviewable decision process rather than a one-click cleanup.
Create a six-column search ledger
Use columns for candidate A, candidate B, approved form, search counts, exceptions, and review date. Add an evidence column for proper names, pointing to a character sheet or approved source. Mark undecided entries explicitly instead of leaving them blank.
The ledger is a record of decisions, not merely a queue of replacement commands. Initial counts become acceptance values for the verification search. Exceptions preserve deliberate dialogue, quotation, and period-specific forms that would otherwise be investigated again.
Collect a small, project-specific candidate set
Begin with variations discovered during reading, frequent number formats, names, and words central to the story. Do not import an enormous general dictionary on the first pass. A set of about twenty high-impact pairs is easier to complete and produces immediate evidence about the manuscript.
Save one sentence of context when the rule depends on meaning. A form may differ between an auxiliary use and a literal use. If the rule cannot be expressed clearly, keep the item in needs-decision status and exclude it from global replacement.
Count both candidates and classify every context
Search candidate A and candidate B separately and record their totals. Classify results as standardize, intentional exception, or needs decision. A character's speech pattern and a cited title may remain exceptions even when narrative prose follows a house style.
Process large result sets by chapter. Record standardized and exception counts after each chapter. Reviewing twenty contexts at a time makes it easier to discover that an early rule is too broad before it damages hundreds of occurrences.
Dry-run the exact file scope
Save a backup or version snapshot, then run a preview replacement and compare its file list and count with the ledger. Stop when counts differ. Confirm that reference notes, generated HTML, and archived drafts are not accidentally inside the replacement scope.
In an August 11, 2026 Rune Studio CLI test, two UTF-8 LF files contained two matches for a Japanese name. A dry run reported two files and two replacements; the actual replacement succeeded, and a new search found the replacement form twice. The tool executed a defined edit; it did not decide which name was semantically correct.
Search both old and new forms after replacement
Search the old form and confirm that it reaches either zero or the documented exception count. Then search the new form and confirm the expected increase. Record how many new-form occurrences existed before replacement so that the final total can be explained.
Open representative matches from the beginning, middle, and end of the scope. Read surrounding sentences and inspect punctuation. Correct counts do not prove correct meaning, especially when patterns, width variants, or regular expressions are involved.
Preserve exceptions as reusable knowledge
Write why each exception remains: speaker voice, quotation, historical spelling, title, or technical notation. On the next review pass, the same occurrence can be recognized instead of researched again. If exceptions dominate, split the rule by meaning, speaker, or document type.
For a series, copy approved rules into the next volume's initial ledger while recording work-specific overrides. The goal is not a frozen universal dictionary; it is inherited reasoning with visible differences.
Keep discovery and replacement separate
A search session can discover candidates without changing any text. Schedule the decision pass separately, especially when an editor offers convenient replacement controls. This pause gives you time to consult the style sheet and isolate the correct file scope.
Rune Studio supports Japanese long-form search and replacement, but semantic approval remains an editorial responsibility. Automation should execute a reviewed rule and report its scope, not invent the rule from frequency alone.
Conclusion: make every replacement auditable
A reliable consistency pass combines search, context classification, dry run, replacement, and two-sided verification. The ledger connects each text change to a rule, count, exception, and review date. That structure prevents both indiscriminate standardization and repeated investigation.
Start with only two pairs that recently caused hesitation. Count them, classify each context, and write a one-sentence rule. Move to a dry run only when the expected file and replacement counts are clear.