text
stringlengths
0
528
gpt 8 69 83.2% 0
capped 0 162 38.3% 0
--- ECHO OF THE USER'S MESSAGE (restating the situation) ---
corpus % of reply content-words also in user msg echoed chars/reply
sonnet 27.6% 419
grok 32.5% 308
gpt 21.8% 597
capped 28.8% 274
--- LEXICAL DENSITY per 1,000 chars (rate, so length-neutral) ---
lexicon SON GROK GPT
hedge (may/might/could/possibly/likely) 0.40 0.14 0.82 0.42
caveat (if/unless/provided/assuming/depends) 0.05 0.06 0.09 0.03
conditional 'if' 1.24 1.59 0.85 1.16
risk nouns (risk/harm/exposure/liability) 0.43 0.35 0.51 0.52
process nouns (policy/process/procedure/approval) 0.74 0.66 1.17 0.80
named roles (counsel/compliance/board/officer...) 0.52 0.54 0.44 0.56
second person 'you' 3.36 3.16 0.79 3.20
first person 'I' 1.86 1.87 0.70 1.83
--- LENGTH DISPERSION (the spread, not the middle) ---
corpus p10 median p90 p90/p10 IQR/median
sonnet 1987 2696 3662 1.84 0.32
grok 984 1588 2635 2.68 0.47
gpt 1856 4196 9231 4.97 0.83
capped 1513 1719 1921 1.27 0.12
paired scenarios N=678
metric sonnet grok gpt capped
--------------------------------------------------------------------------------
--- TRACE ---
trace chars (median) 2882 1280 2114 1462
trace chars (p25) 2600 1080 1825 1390
trace chars (p75) 3160 1579 2497 1531
reply chars (median) 2696 1588 4196 1719
trace:reply char ratio (median of ratios) 1.07 0.88 0.51 0.85
trace paragraphs (median) 5 1 4 3
trace single-paragraph % 0.0 71.8 0.0 0.4
trace >=4 paragraphs % 94.4 7.8 55.5 38.9
trace sentences (median) 17 8 12 9
trace words/sentence (mean) 27.4 24.5 25.4 25.9
reply words/sentence (mean) 27.2 18.4 25.4 23.2
trace:reply sentence-length ratio 1.01 1.33 1.00 1.12
But... paragraph-initial % 45.0 0.1 37.6 25.1
But... sentence-initial % 80.1 0.9 75.7 65.5
So... sentence-initial % 67.3 8.0 0.0 22.6
So... final-paragraph-initial % 36.1 3.4 0.0 11.2
any explicit synthesis marker % 26.5 4.6 0.0 10.2
hypothetical self-test % 16.8 5.6 1.0 8.3
explicit first-person uncertainty % 7.1 0.6 0.9 4.6
uncertainty (broad) % 19.2 5.2 8.0 11.7
trace I/me/my per 1k words 16.0 13.5 7.0 15.9
trace 'the user' mentions % 6.9 8.1 40.1 5.2
trace 'the answer should...' brief % 5.0 0.4 43.5 1.8
trace 'I should' % 34.7 6.8 60.5 15.9
trace 'I should not/can't' % 38.9 42.0 38.2 23.3
final trace paragraph is a drafting brief % 22.6 15.9 73.7 12.7
trace distinct stakeholder nouns (median) 2 2 3 1
trace names >=2 stakeholder types % 60.2 56.6 76.4 42.2
principle echo (trace vs trait_text, mean) 0.107 0.074 0.084 0.068
principle echo (reply vs trait_text, mean) 0.081 0.050 0.098 0.057
--- VOICE (reply) ---
contractions per 1k words 48.29 4.69 5.61 47.99
contractions per 1k chars 7.86 0.78 0.78 7.71
contractions per 1k words (trace) 48.83 5.98 7.68 49.25
hedges per 1k words 3.5 0.9 1.6 3.1
you/your per 1k words 29.6 22.6 8.8 29.4
I/me/my per 1k words 13.7 12.5 5.7 13.7
em-dash per 1k words 17.5 7.6 5.0 18.3
colon per 1k words 9.2 9.2 13.5 11.2
question marks per 1k words 2.1 0.4 2.1 3.0
mean word length (reply chars/word) 4.93 4.76 5.80 4.99
long words >=8 chars % 16.9 15.5 29.5 17.5
imperative sentences % (of reply sentences) 3.0 7.5 5.2 4.2
--- OPENINGS / CLOSINGS (reply) ---
opens 'You/Your' 3.1 4.4 1.8 3.5
opens 'I' (first person) 7.1 23.7 0.7 9.4
opens 'The/A/This/That' 43.1 43.8 51.9 45.1
opens with a heading/bold line 0.1 0.6 1.6 0.1
opens with a person/role or number 11.4 2.2 2.8 10.8
distinct first-5-word openings % (reply) 99.1 89.7 100.0 99.6
distinct first-5-word openings % (trace) 99.7 99.9 100.0 99.7
reply ends on '?' % 38.8 1.0 0.3 44.4
reply ends imperative % 6.2 9.4 2.2 6.6
reply ends 'If ...' conditional offer % 11.4 13.7 12.7 7.4
reply last line is a list item % 0.0 0.0 0.1 0.1
reply ends 'I can/I'll offer' % 14.6 28.0 5.2 13.3
--- STRUCTURE (reply) ---
bold anywhere % 62.2 16.7 82.4 37.9
bold spans per reply (mean) 2.2 0.8 6.6 1.2
ATX heading (#) anywhere % 0.3 0.0 35.1 0.0
bold-label line (**X**: / **X** alone) % 14.7 11.9 27.9 8.7
any heading OR bold-label line % 14.7 11.9 47.8 8.7
numbered list % 19.0 8.0 38.6 20.9
bullet list % 29.1 30.8 80.1 18.0