Do not choose an EPUB editor for Japanese fiction from a “vertical writing supported” label. Give candidates A and B the same five samples: punctuation, line-start and line-end restrictions, a two-digit upright-number case, ruby, and right-to-left progression. Score them in two independent rounds. A sample with conflicting round-one and round-two results stays pending. Only a candidate with agreement on all five requirements advances to a real-manuscript trial. Print DTP and identical rendering in every reader are outside this selection.
Turn five visual samples into stated requirements
Write one expected state for each sample before viewing a candidate. Punctuation must occupy the intended vertical position. The selected prohibition case must not break at the declared line boundary. “12” must stand as one upright group. Ruby must attach to the named base text. The transition to chapter two must progress from right to left.
This article does not send one chapter through editor, package, and readers or perform a one-character edit and regeneration. That end-to-end test belongs to STUDIO-231. The task here is narrower: compare candidates with the same five Japanese-typesetting decisions.
Round one records five explicit observations
Use five rows for each candidate. The columns are sample, expected state, observed state, round-one decision, and evidence note. With the candidate name visible, classify each row as pass, fail, or unable to judge.
For the two-digit row, “12 forms one upright group” is an expected state. Describe what was observed rather than writing only “looks good.” Unable to judge is not a low score; it means the required view or artifact was unavailable.
Round two hides the first decision
For round two, hide the candidate name and the first decision, then score the same five samples in the same order. Use another reviewer when possible. A single reviewer can still close the first record and work only from the expected states and observation targets.
| Round one | Round two | Final state |
|---|---|---|
| Pass | Pass | Requirement met |
| Fail | Fail | Remove candidate |
| Unable to judge | Any result | Pending |
| Pass | Fail | Pending; review observation conditions |
| Fail | Pass | Pending; review observation conditions |
Do not use majority voting. A split decision identifies an unstable observation and remains pending.
Between rounds, change only whether the evaluator can see the candidate name and the previous result. Keep sample wording, display scale, line width, and chapter position fixed. If an observation condition must change, start a new sheet instead of combining it with the current two rounds.
Write evidence against the expected state. “Candidate A preview” is not enough. A note such as “A-NUM shows 1 and 2 in separate orientations instead of one upright group” lets the second round recognize the same condition without seeing the candidate label.
One split row prevents selection
Candidate A may agree on punctuation, restrictions, ruby, and progression but split on upright digits. Its recommendation remains pending. Candidate B may agree on four rows while ruby cannot be observed; it also remains pending. A total score would let a required condition disappear inside an average.
If a manuscript does not use ruby, declare that row not applicable before evaluation begins. Do not remove a requirement after seeing a poor result. The five-sample sheet is a way to state this project’s non-negotiable Japanese-typesetting requirements in advance.
Place Rune Studio as a candidate with structural evidence
In the August 11, 2026 dedicated-workspace test, a Rune Studio vertical EPUB had verified vertical-rl, right-to-left progression, valid=true, nav, and NCX. That evidence supports package structure and reading direction.
The five-sample visual assessment of punctuation, restrictions, upright digits, and ruby across viewing environments has not been performed. Rune Studio can therefore enter the comparison sheet, but the current evidence does not show that it passes all five rows. An integrated workflow from manuscript editing through cover production has not been tested here and is not part of this decision.
Advance only five rows with two-round agreement
The final sheet keeps five rows per candidate with round one, round two, final state, and pending reason. Advance only a candidate with pass/pass on every applicable row. A split row states exactly which observation condition must be clarified before another evaluation.
This method does not prove the entire EPUB production workflow. It gives the software selection one specific job: compare non-negotiable Japanese-typesetting requirements with the same two-round scale. Choose from five agreed observations, not from a vertical-writing label.
Start today by writing only the five expected states. Define the decision language before opening a product or revealing a candidate name; that prevents the standard from being relaxed after a result is seen.
