You are given ONE film-still description, its location text, and the
world facts. Decide whether the scene ITSELF names writing that must be
readable in frame, and when it does, author the exact words — because
the image model receives only what you write here and must not invent
lettering on its own.

Ground every entry in what you were given, and say where it came from:

- scene_text — the shot or location text names this writing, or names a
  thing whose whole purpose is to be read.
- world_facts — the world facts establish it for this place and era.
- inferred — neither of the above. You reasoned it from what kind of
  place this is, or from what would plausibly be there.

Be honest about `inferred`. A plausible guess is still a guess, and
"this would make the place read as itself" is exactly a guess. Most film
stills name no writing at all — an ordinary room, street, or vehicle
carries none the scene calls for. When the scene names nothing, return
an empty list. That is the expected answer, not a failure.

Do not choose where the writing goes. Name only the words; the image
already establishes the surfaces.

Never write dialogue, captions, subtitles, or narration. Never use real
brand names, real institutions, or personal names unless the scene text
itself names them.

Return for each entry:

- text_native: the exact words, in the NATIVE language and script of the
  place and era the world facts establish — never another language,
  never romanization. Short, and worded the way it would actually be
  written there and then.
- source: scene_text, world_facts, or inferred, as defined above.
- reason_ko: 한 문장 — 무엇이 이 글자를 부르는가. `scene_text` 이면 그
  대목을 원문 그대로 인용한다.