Keeping Japanese and English site copy in one canonical source

The input document for semantic design was rebuilt as the canonical source shared by both languages. It treats Japanese as the canonical meaning and English as an output candidate, while seven decision records, C-01 through C-07, resolve conflicting instructions. Audience, intended emotion, voice, medium, length, and read-aloud behavior were brought together with truth, legal, and publication boundaries.

• The Home information structure and navigation were fixed, with shared conditions for calls to action, AI-persona and AI-generated-image disclosure, and the boundary that the support loop does not guarantee outcomes.
• Six exact examples—Product, YouTube, 3D model, Design image, Music, and Business experiment—were retained without limiting all activities to those six categories.
• Meaning and structure, bilingual expression, legal decisions, and implementation decisions were separated, with a boundary against treating unverified URLs or display states as settled copy.

The copy was aligned to 42 shared keys. Every key received an exact source, canonical Japanese meaning, allowed expression range, status in both languages, prohibitions, and dependencies. English status was split into 32 output-candidate, six pending, and four blocked entries so unresolved copy could not appear complete.

Comparison of the first consolidated draft found that audience, emotion, voice, channel, length, read-aloud behavior, the support loop's non-guarantee boundary, the six exact examples, and each key's exact source and allowed range had been omitted. Aya's existing Design image responsibility had also disappeared. These details were restored and the profile-responsibility decision was recorded as C-07 before the source was finalized.

The output document became the sole active copy source, carrying the same 42 keys in the same order as the input. Each key pairs its Japanese semantic baseline with a natural English candidate, placing the Hero, calls to action, profiles, disclosure, and URL status in one sequence. The former English-only input and output became migration stubs that explicitly prohibit copying, revising, implementing, or publishing them as current copy.

• Ten profiles were aligned to five fields each—display name, Role, responsibility, role value, and image path—for 50 fields in total.
• Display name, Role, and image path became locale-invariant values, while responsibility and role value became Japanese-English pairs with the same meaning. Reader-facing card copy was reduced to the role value, with responsibility retained as supporting source context.
• Ten Role names were aligned to shared English labels and all ten Japanese role values were rewritten as plain single sentences. All ten responsibilities retained their prior meaning; role values fit 24–34 Japanese characters and 7–12 English words.

Checks counted and ordered the 42 keys, compared meaning across languages, inspected five fields for all ten profiles, searched for remaining placeholders, and measured Role names and role values. Profile states before and after the rewrite were also compared to confirm that the other 41 keys and all ten responsibilities had no unintended changes.

The primary diff was checked line by line to confirm that the behavior described above matched the change at the cutoff.