Skip to content

INPUT_FORMAT.md

Latest commit

 

History

History
37 lines (28 loc) · 1.51 KB

INPUT_FORMAT.md

File metadata and controls

37 lines (28 loc) · 1.51 KB
Raw

Input format contract

This tool consumes three things. Any producer (manual authoring, a future synthesis project, another writing tool) just has to emit this shape.

1. The text — my_text.md (.md or .txt)

Plain prose. Mark each claim with one or more citation markers after the sentence(s) that claim supports:

Nasal irrigation reduces viral shedding. [[smith2020]]
AGI may reshape labor markets and productivity. [[jones]] [[lee]]
  • Marker syntax: [[key]]; key charset is [A-Za-z0-9_-].
  • Consecutive markers group onto the preceding sentence(s).
  • A sentence with no marker gets the own verdict (the author's uncited claim — thesis, argument, transition; shown indigo, nothing is checked). A small default-on pass tags own claims structural / opinion / fact, and "fact" gets a dismissible "citation needed?" nudge.

2. References — how keys map to source files

Resolved in this precedence:

  1. --references <file>
  2. a sibling my_text.md.refs.txt next to the text file ← default
  3. a trailing [References] block inside the text

Format is one key = filename per line; lines starting with # are comments:

smith2020 = smith_2020_nasal_sprays.pdf
jones     = jones_agi_labor.txt

3. Sources — the --sources <dir> folder

Contains every file named in the references. .pdf (parsed with pypdf) and .txt are supported.

Output

<output-dir>/analysis.json (verdicts + evidence + coverage) and <output-dir>/viewer.html (self-contained, opens with no server — see README).