Put the source material here — the actual writing or transcribed speech of the character you're
imitating. Plain text (.txt) or markdown (.md), one file per coherent sample. Name them so they
sort sensibly: 01_*.txt, 02_*.txt, ...

Only .txt and .md files in this folder are read as voice samples (this file has no extension, so
it is ignored). Clean the source first if it contains anything the deployed model must not
reproduce — see ../prompts/content_rules.md.

Optional extras:
  - A biography file: reference it from a mode's `context_files` in ../project.toml so the
    synthesizer can ground "personal question" answers in real facts.
  - A glossary file (recurring terms/phrases): set `glossary_file` in ../project.toml. It is
    appended to the synthesis prompt but NOT treated as a voice sample.
