JSON and CSV, without guessing which string is text
A JSON file is not a translation file. The analyzer reads structure you can point at, and when your schema uses its own field names you map them — instead of hoping the tool guessed right.
Read directly
- Structured JSON and CSV records: keys with their text, and key/value pairs that are clearly copy rather than configuration.
- The UI-text fields that dominate real databases:
name,title,description,category. - Dialogue context already present in the file: speaker, listener, both genders, scene, conversation, node, choice group, voice and portrait — from explicit field names only.
- Nested speaker objects and top-level character databases: an id becomes the character’s display name and gender by structure, not by resemblance.
- Locale variants collapse:
fb.jsonwithfb_ja.jsonbecomes one entry with its variants attached instead of unrelated duplicates.
Known limits
- Non-canonical field names (
char_from,sex_a,body) are mapped by you in the Context Mapping review. Nothing is inferred from position or from the text itself. - A schema whose text nothing identifies is reported as unclassified rather than extracted.
- Binary files are classified, never scanned for printable strings: a byte scan is a guess, not an extraction.
Never touched
- Your original file is not rewritten. Export produces a new file in the same format you opened.
Statuses come from the project’s tested compatibility record. Every “read directly” line was validated against real game files or fixtures; partial support is written as partial, and an external tool is named as a tool you configure yourself.
Open a file in the browser editorGame archives are read by the Windows build, where the archive tools run on your machine.Windows app