$ styleprofile score --by-paragraph examples/draft.md examples/writer/old-maps.md '<tmp>/demo-writer.json'
[exit 0]
--- stdout
STYLE COMPARISON   2 documents, 2 chunks, 1,182 words   vs demo-writer.json (55 chunks)

By document   furthest from the reference first
  2 documents: 2 close
  document                      words  verdict             Delta  LLM-likeness
  examples/draft.md               542  close                0.64  a few traits
  examples/writer/old-maps.md     640  close                0.55  like reference

Across 2 documents: close   Delta 0.60
  Lower is closer. The reference's own held-out writing at these lengths (542-640 words per chunk) scores 0.59 typically
  (a mean over 2 chunks, up to 0.85).
  Pooled over every document from here on; the table above has each one's own.

LLM-likeness: like the reference   0.50
  The reference's own writing at these lengths (542-640 words per chunk) scores 0.24 typically (a mean over 2 chunks, up
  to 0.53); the LLM drafts 8.41.

By area   Delta ÷ the top of the reference's usual range in each area; bars on a log scale
  close up to 1x, somewhat different to 1.5x, clearly different to 2x, very different above
  Voice and stance             0.93x  ████░░░░░░░░░░░░░░░░  close
  Punctuation                  0.78x  ███░░░░░░░░░░░░░░░░░  close
  Function words               0.72x  ███░░░░░░░░░░░░░░░░░  close
  Syntax                       0.67x  ███░░░░░░░░░░░░░░░░░  close
  Rhythm                       0.60x  ███░░░░░░░░░░░░░░░░░  close
  Sentence shape               0.55x  ███░░░░░░░░░░░░░░░░░  close
  How sentences open           0.52x  ██░░░░░░░░░░░░░░░░░░  close
  Vocabulary                   0.49x  ██░░░░░░░░░░░░░░░░░░  close
  Formatting                   0.21x  █░░░░░░░░░░░░░░░░░░░  close

Where it drifts (experimental)   2 of 15 paragraphs drift, read in spans of at least 100 words
  examples/draft.md, line 9   a few LLM traits (0.92), close (Delta 1.14)
                              "But the store is more than a place to buy things — it's a…"
                              Em dashes ▲▲▲, LLM marker words (delve, crucial) ▲▲▲, Long words (7+ letters) ▲▲▲
  examples/draft.md, line 15  a few LLM traits (0.92), close (Delta 0.90)
                              "Ultimately, the future of the village hardware store depends…"
                              "these" ▲▲▲, Nominalizations (-tion, -ment) ▲▲▲, Long words (7+ letters) ▲▲▲

By paragraph   examples/draft.md   spans of at least 100 words; * drifts, above 1.00x its range
     lines      words   x range  LLM-likeness
     3             70     0.16x  like the reference      The hardware store in the village has…
     5             85     0.23x  like the reference      It's a narrow building, long and dim…
     7             84     0.23x  like the reference      I go in for one thing and come out an…
  *  9             60     1.18x  a few LLM traits        But the store is more than a place to…
     11            86     0.40x  like the reference      The big store in the city is cheaper. I…
     13            63     0.40x  like the reference      Walt is seventy-one. His daughter works…
  *  15            33     1.02x  a few LLM traits        Ultimately, the future of the village…
     17            61     0.63x  like the reference      I bought a pound of eightpenny nails…

By paragraph   examples/writer/old-maps.md   spans of at least 100 words; * drifts, above 1.00x its range
     lines      words   x range  LLM-likeness
     3             80     0.19x  like the reference      The best map of this town was drawn in…
     5            114     0.13x  like the reference      I used to stand in front of it while I…
     7             97     0.17x  like the reference      Old maps are wrong in interesting ways.…
     9            104     0.15x  like the reference      My father had a map he'd drawn himself…
     11            93     0.20x  like the reference      There's a kind of person who loves maps…
     13            88     0.38x  like the reference      What I like about an old map, finally…
     15            64     0.28x  like the reference      I've been asked, now and then, to give a…

Biggest differences   (each ▲ or ▼ is one standard deviation, up to 3)
                                                   this text   reference
  "by"                                              10.1 /1k     4.4 /1k   ▲▲
  "while"                                            3.3 /1k     0.7 /1k   ▲
  "but"                                              4.3 /1k     1.8 /1k   ▲
  Parse tree depth (clause nesting)                     4.59        4.11   ▲
  "on"                                              15.2 /1k    10.0 /1k   ▲
  "or"                                               0.8 /1k     4.2 /1k   ▼
  "has"                                              5.3 /1k     2.5 /1k   ▲
  "there"                                            9.3 /1k     5.4 /1k   ▲

Pass --all to see every metric.
