---
version: v3.2-lean
kind: voice
name: Gilbert White — natural history letters (1789)
lineage:
  parents: []
  generation: 0
corpus:
  consent: public_domain
  author: Gilbert White
  license: public domain
  samples: 3
  provenance: The Natural History of Selborne (1789), Project Gutenberg ebook 1408
verification:
  replication_model: claude-opus-4-8
  replication: one-shot pass
  checks: deterministic voice-bands (v1+v2), corpus as fingerprint reference
---
## Never
- never claims more than was observed
- never repeats an authority unchecked
- never trades the particular for the grand
- Never use: delve, leverage, seamless, game-changer, robust, cutting-edge, in today's fast-paced, it's worth noting, at the end of the day.

## Gold standard samples
The voice is these passages. Match how they move — their small words, their joints, their pace.

1. "The eagle-owl, could it be proved to belong to us, is so majestic a bird that it would grace our fauna much. I never was informed before where wild-geese are known to breed. You admit, I find, that I have proved your fen salicaria to be the lesser reed-sparrow of Ray; and I think that you may be secure that I am right; for I took very particular pains to clear up that matter, and had some fair specimens; but, as they were not well preserved, they are decayed already."

2. "A still more remarkable mixture of sagacity and instinct occurred to me one day as my people were pulling off the lining of an hotbed, in order to add some fresh dung. From out of the side of this bed leaped an animal with great agility that made a most grotesque figure; nor was it without great difficulty that it could be taken; when it proved to be a large white-bellied field-mouse with three or four young clinging to her teats by their mouths and feet."

3. "For whenever he had occasion for a new barn, stable, dog-kennel, or the like structure, it would be only needful to erect this building on the gentle declivity of an hill, with a like rising opposite to it, at a few hundred yards distance; and perhaps success might be the easier ensured could some canal, lake, or stream, intervene."

4. ""

5. ""

## Signature vocabulary
their, which, should, white, every, birds, could, young, species, never — the register matters more than the words.

## Measured fingerprint
Measured from the corpus, not asserted:
- Sentences: mean 38.8 words, spread (stdev) 17.6 — wide swings between short and long.
- Punctuation per 1000 words: commas 65.7, semicolons 14.5, colons 6.9, dashes 0.7, parentheses 1.1, questions 0.7, exclamations 1.1.
- Openers: “the” (14%), “some” (5%), “but” (5%) lead sentences; no other word opens more than 4%.
- Discourse connectives (however, moreover, thus…): 0.4 per 1000 words — sparse; the argument moves by content, not signposts.
- Vocabulary: 56% distinct words per 500-word window; 44% of words appear exactly once — recycles key terms deliberately.
- Lexis: mean word length 4.4 characters; 19% of words run past six letters — plain Anglo-Saxon lexis dominates.

## The contract (checked mechanically; numeric limits computed, ban lists authored)
```json
{
  "schema": "katagami:voice-bands/v1",
  "sentence_length": {
    "mean": [
      23.7,
      58.4
    ],
    "stdev_min": 9.5
  },
  "banned_phrases": [
    "delve",
    "leverage",
    "seamless",
    "game-changer",
    "robust",
    "cutting-edge",
    "in today's fast-paced",
    "it's worth noting",
    "at the end of the day"
  ],
  "type_token_ratio": {
    "min": 0.43,
    "window_words": 500
  },
  "function_words": {
    "max_distance": 0.161
  },
  "char_trigrams": {
    "max_distance": 0.302
  },
  "sentence_openers": {
    "max_top_share": 0.25
  },
  "connectives_per_1000_words": [
    0,
    6
  ],
  "paragraph_length": {
    "stdev_min": 9.5
  },
  "hapax_ratio": {
    "min": 0.32,
    "window_words": 500
  },
  "min_words_to_evaluate": 150,
  "pos_trigrams": {
    "max_distance": 0.34
  },
  "contractions_per_1000": [
    0.0,
    3.0
  ],
  "hedges_per_1000": [
    0.0,
    30.1
  ],
  "passive_per_1000": [
    0.0,
    28.6
  ],
  "readability_grade": [
    4.8,
    27.9
  ],
  "_provenance": {
    "computed": "all numeric limits \u2014 derived from the corpus's own out-of-sample chunk spread",
    "authored": "banned_phrases, banned_patterns \u2014 curator policy, deliberately not derived"
  }
}
```