# Full text: Blakean Entomology: Small Lives, Uncertain Bodies, and the Work of Attention

> Extracted from `Blakean_Entomology.pdf`
> Source: https://zenodo.org/api/records/23900583/files/Blakean_Entomology.pdf/content
> SHA-256: fc99a3dd629646a6e7cb1b41e0b6290388cbb4a09d7c0f91526e0c413dddf73a
> Version: 1.0.0
> Extraction: pypdf 6.16.2; 42 pages.

## Page 1

Blakean Entomology
Small Lives, Uncertain Bodies, and the Work of Attention
Daniel Ari Friedman
ORCID: 0000-0001-6232-9096
DOI: 10.5281/zenodo.23900583
2 November 2026


## Page 2

Graphical abstract 1: The Fly: likeness and injury
Figure 1: The Fly: likeness after injury. “Am not I / A fly like thee? / Or art not thou / A man like me?” The Fly; Songs Y, Bentley
plate / object 40, E23. Physical transcription lines 7-10. William Blake, Songs Y, printed ca. 1825. The Met, Rogers Fund, 1917;
17.10.40. Public Domain / CC0. Complete scan. ( Blake, 1825). The panels distinguish a reading, a rival and a conditional classification.
The lower panel separates 36,039 candidates, 47 authored review decisions and 35,992 unreviewed candidates. Its proportional bar keeps
the complete candidate denominator; neither the image nor the analogy adds a contextual decision.
2

## Page 3

Graphical abstract 2: A Dream: hearing and a route home
Figure 2: A Dream: capacities in relation. “I am set to light the ground, / While the beetle goes his round: / Follow now the beetles
hum, / Little wanderer hie thee home. ” A Dream; Songs Y, Bentley plate / object 26, E16. Physical transcription lines 19-22. William
Blake, Songs Y, printed ca. 1825. The Met, Rogers Fund, 1917; 17.10.26. Public Domain / CC0. Complete scan. ( Blake, 1825). The
panels distinguish a reading, a rival and a conditional classification. The lower cards separate the emmet’s need, the glow-worm’s light
and the beetle’s audible route. These are capacities offered in the text, not a reported rescue or observed insect behavior. A compound
name must be read before classification.
3

## Page 4

Graphical abstract 3: Thel: comparison, care and classification
Figure 3: Thel: a Worm is not a lineage. “Art thou a Worm? image of weakness. art thou but a Worm? / I see thee like an infant
wrapped in the Lillys leaf:” The Book of Thel F, object 6; Bentley 6 / Erdman plate 4. Physical transcription lines 3-4. William Blake,
Thel (1789), copy F. Library of Congress, Rosenwald 1798A. Printing dates differ. Complete scan; retained LOC rights qualification.
(Blake, 1795). The panels distinguish a reading, a rival and a conditional classification. The lower plot shows the three largest absolute
category-share changes under pooled and equal nonempty-group weighting. Circles and squares share a 0–100% axis; full-profile total
variation is 0.222387. Lexical profiles and the Worm–infant comparison establish no species identity or ancestral relation.
4

## Page 5

Contents
1 Small lives and the work of attention 6
2 Biological symbolism and its critical inheritances 6
3 F rom a captured expression to a literary claim 7
3.1 Words, designs and the material page . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.2 Historical names and conditional classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4 Provision and injury in the Songs and Gates 9
4.1 A Dream: assistance without a completed rescue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.2 The Fly: consolation after the creature’s injury . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.3 The Sick Rose: secrecy, desire and uncertain embodiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.4 The Human Abstract: pity and the nourishment of cruelty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.5 Gates of Paradise: bodies between copies and captions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
5 The wine-press: abundance beside unequal suffering 17
6 Recognition, generation and collective bodies 17
6.1 Thel: recognition and nurture through an infant analogy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
6.2 Polypus: joining and the powers of a body . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
6.3 Webs, prisons and the allocation of danger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
7 Seeing, injuring and judging 20
7.1 Microscopic display and the failure of demonstration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
7.2 Senses, cells and the limits of emancipatory speech . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
7.3 Animal suffering and the force of an augury . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
8 How the collection shapes what can be counted 21
8.1 A retrieval candidate is not an animal observation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
8.2 What weighting changes and what it cannot correct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
8.2.1 Compositions, omission and concentrated evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
8.3 What exact-text equality leaves unresolved . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
8.4 Near words and shared source units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
8.4.1 Twelve literal-pair controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
8.5 Zeros in a declared vocabulary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
8.6 Readings, decisions and their separate denominators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
8.7 Contextual judgment as a contestable record . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
8.8 The text occupied by a match and by a reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
9 Attention, symbolic inheritance and unfinished readings 34
Appendix A: Complete name-family coverage 37
Appendix B: T welve literal-pair controls 38
Appendix C: Material-witness concordance 39
Appendix D: Sources and reading boundaries 39
References 40
5

## Page 6

Selected textual and material witnesses are read alongside an exhaustive search of declared forms in a pinned corpus. Corpus-
wide contextual and visual review remains incomplete.
Abstract
How do Blake’s small lives make another being’s needs matter? This article argues that provision, feeding and aggregation give
different forms to that question: help can enable another life, appetite can use it, and a collective body can redistribute its
powers. Copy-specific readings of four short poems, nine selected historical scans and bounded prophetic and prose passages
test this account against symbolic, spiritual and human-centered alternatives. The emmet’s offered route, the fly’s reciprocal
likeness and the Worm/infant comparison become visual arguments in which the complete impression remains beside the exact
words, a rival reading and a limit on biological identification. A reproducible concordance supplies breadth without turning
retrieval into a bestiary: 102 categories yield 36,039 category-source-span candidates in 2,067 captured units; 2,498 belong to
199 identified Blake-primary reading units. Within that scope, 82 name-family categories supply 1,239 candidates. Pooled
and equal nonempty-group compositions differ by total variation 0.222387 across 26 positive groups. Source-role contrasts,
finite proximity experiments and recorded reading windows expose dependence on particular inputs. The contextual ledger
contains 47 authored decisions and leaves 35,992 candidates unreviewed. These measures describe a search and its selective
examination, not animal prevalence or complete-oeuvre coverage. Baine’s biological-symbolic scholarship, accessed here through
a contemporary review, makes contextual variability a continuing challenge. The argument locates ethical force in particular
acts of hearing, sustaining, injuring and joining; neither a shared name nor a modern lineage determines that force.
1 Small lives and the work of attention
The emmet’s difficulty in “A Dream” is practical: she is lost, tired and separated from her family. The poem makes that difficulty
audible before a glow-worm offers light and a beetle’s movement offers a route home. The speaker contributes sympathy, but
does not perform the entire rescue. Distributed capacities become useful because they respond to a particular creature’s need.
The animal scene therefore gives this article a question more precise than whether Blake likes insects: what forms of attention
and association make another being’s life possible? The reading concerns “A Dream,” E16, checked against retained Erdman
record 26 and Archive Songs copy A, object 27, especially physical text lines 5, 16 and 20–21. Archive source.
“Entomology” names the inquiry rather than restricting it to the modern class Insecta. Worms, spiders, centipedes, scorpions,
snails and Polypus belong within the requested field. Their differences matter. A glow-worm’s compound name must not turn
it into an earthworm, and a mythic Polypus must not acquire a modern species identity through a database field. The project
records Blake’s expression first and biological classification, when warranted, separately. Ambiguous boundaries between animal,
vegetable, human and collective life can themselves be literary evidence.
This breadth also prevents a familiar selection bias. Reading only the short lyrics privileges the fly, worm and emmet while
overlooking prophetic inventories, prose, manuscripts, annotations and visual descriptions. Reading all source planes together
creates the opposite error: a curator’s image description or editorial note can become a false Blake quotation. The present
method retrieves broadly while preserving the distinction between a captured lead and an admitted interpretive claim.
The argument follows three relations rather than a stable bestiary: provision that enables another life, feeding that uses a
living body, and aggregation that reorganizes its members’ powers. A glow-worm’s assistance and the Clay’s nurture make
dependence answerable; a worm’s destructive intimacy and the wine-press’s suffering prevent abundance from guaranteeing
reciprocity. The Polypus asks what happens to distinct capacities when lives join. Each relation can intersect the others, but
neither a shared animal name nor a biological analogy supplies a single moral verdict.
2 Biological symbolism and its critical inheritances
Rodney M. Baine’s The Scattered Portions: William Blake’s Biological Symbolism , prepared with the assistance of Mary R.
Baine, is a necessary book-length predecessor. Its insect chapter is described in an inspected contemporary review; the chapter
itself was unavailable for direct reading in this research pass. Its formal heading and boundaries therefore remain unverified.
(Baine, 1986; Hoagwood, 1989, p. 40)
Hoagwood values Baine’s historical sources, examples and illustrations while questioning the relation between symbolic classifi-
cation and context. His insect discussion reports an interpretation of metamorphosis through individuals and states, alongside
spiders associated with institutional entrapment. The review also recognizes Baine’s acknowledgment of variable meanings. Its
objection is more demanding than a rejection of symbolism: a particular passage or design may hold contrary implications
together instead of assigning them to separate moral categories. This account belongs to Hoagwood’s review, not to a direct
reading of Baine’s chapter. ( Hoagwood, 1989, pp. 39–42)
The strongest rival to the present argument follows from that problem. Provision, feeding and aggregation might all stage
human faculties, spiritual states or regeneration. Calling them relations does not make them cease to be symbols; recording
6

## Page 7

a source span does not establish an ethic of nonhuman life. Nor does a speaker’s care necessarily answer the creature’s need.
The article must consequently show where an offered action, a displaced voice or an unresolved ending constrains a particular
interpretation. The question becomes: when do the needs of another being qualify the speaker’s understanding, and when does
human analogy, moralization or collective incorporation absorb them? The cases below test that distinction without making a
relational reading the automatic successor to symbolic scholarship.
The study does not claim to discover that Blake writes about invertebrates. Janelle A. Schwartz’s chapter on Blake and
taxonomy explicitly addresses the varied worm and its relation to decay and generation. Only its publisher abstract and
bibliographic information were accessible in this pass; the chapter’s full argument has not been read here. Its existence rules
out presenting a worm-based approach as an unprecedented research subject. ( Schwartz, 2012, chapter 4, publisher abstract
only).
Caroline Dauphin’s discussion of the Polypus connects its changing poetic forms with problems of generation, bodily boundaries
and collectivity. The earlier selected paragraphs 3–6 and 21–25, now extended through paragraphs 7–28 and 49–50, inform
the present question: how can an aggregated body become both an image of connection and a threat to distinct lives? Her
historical genealogy is a scholarly interpretation, not independently established by the search totals reported here. ( Dauphin,
2026, paragraphs 3–28 and 49–50).
Perkins’s complete article on animal rights and “Auguries of Innocence” and Srigley’s complete article on “The Sick Rose”
now supply substantive, divergent readings of injury. Material scholarship by Essick and Viscomi and selected Fara sections
place reproduction, generation and the terms of observation beside the poems; their limited reading scopes are specified below.
Gould, Baker and Erasmus Darwin offer historically available comparisons, with no demonstrated reading or influence chain
from these books to Blake. ( Perkins, 1999; Srigley, 1992; Essick and Viscomi , 2002; Fara, 2020)
The contribution attempted here is a reproducible relation between broad retrieval and situated reading. Every retrieved
lead should be traceable to its source span and edition. Submitted contextual decisions additionally retain their reviewer and
rationale; a close reading in this article does not create a decision in that separate ledger. Every declared search form should
remain inspectable, including zero-result forms and noisy expansions. This is useful precisely where a polished argument might
otherwise hide its selection of examples. The method does not remove interpretive judgment; it makes the decisions through
which a concordance becomes an argument available for examination.
3 From a captured expression to a literary claim
The corpus is an independently copied, private application bundle exported by the public blake repository. Its pin identifies
the captured manifest, recovered projection, SQLite index and portable bundle manifest separately. Original passage bodies,
source hashes, author identities, edition/copy fields, locators, rights and roles survive export. Export verifies the consumed
input bytes before and after the copy. A Git repository’s code, a source capture and an application’s result epoch remain
separate objects.
The baseline searches 2,067 captured units: 540 reading documents, 796 nonempty Erdman pages and 731 raw XML units.
Of the reading documents, 199 carry the identified Blake author and primary role; 341 remain unverified. The Erdman pages
and four XML parts are mixed/editorial source material. The 727 Archive XML documents remain unverified as whole units
because they combine text, apparatus, metadata and descriptions. These counts describe source units, not 2,067 distinct works
or independent historical witnesses. The counts and identities are bound in research/corpus-pin.json and the text-free
baseline summary.
The initial vocabulary has 102 categories and 376 explicit forms. Historical forms include Emmet and Catterpiller. The
user-specified emmett is an explicit candidate spelling, not silently asserted as an attested Blake form. Compound families
include earth-worm, tape-worm, glow-worm, silk-worm, honey bee and cock chafer. The implementation uses the existing
Blake Unicode normalization and retains original character offsets. Whole-token matching prevents ant from matching giant
and bee from matching been. It does not silently stem fly into every motion verb.
Separate contextual categories include web, silk, honey, hive, shell, wing, swarm, sting and spinning. These help recover
discussion that does not name an animal, but they introduce many unrelated cases. A wing can belong to an angel or bird; a
spinner can be a human worker; a sting can belong to a plant. Such forms cannot supply an insect total without contextual
review.
The engine emits every category/source/span candidate without a display cap, plus all nine role/plane cells for every category,
including zeros. Equivalent forms in one category share an identical source span; different categories may overlap. Thus
honey bee can yield a compound candidate and a bee candidate. Neither category sums nor repeated printed copies become
distinct animal observations. Exact-text grouping is reported separately and does not establish that nonidentical editions are
independent witnesses.
Definition 1 (A source-bound retrieval candidate). Fix a captured corpus, a declared vocabulary and one matching
profile, together denoted by 𝜌. A retrieval candidate is an event represented analytically as ℎ = (𝜌, 𝑐, 𝑠, [𝑎, 𝑏)) : category 𝑐
matches the original character interval [𝑎, 𝑏) in retained source unit 𝑠. The interval satisfies 0 ≤ 𝑎 < 𝑏 ≤ |𝑥 𝑠|, where 𝑥𝑠 is
7

## Page 8

that unit’s captured text. The source identity, text hash, role and locator accompany the event. This representation describes
the counting unit, not the implementation’s opaque occurrence ID. Different categories may overlap in one passage; repeated
witnesses retain different source identities. Neither event identity nor a category name establishes an organism, a distinct poetic
scene or a biological taxon.
The distinction in Definition 1 is the first safeguard against turning a search result into a literary conclusion. It preserves the
document and the matching operation that made a lead visible, while leaving its use in that particular scene to be read.
The first run returns 36,039 category-span candidates, including 2,498 in the Blake-primary reading scope. These totals
deliberately include contextual leads and non-animal senses. They are measurements of the declared retrieval program, not a
count of invertebrate mentions. Candidate totals for selected categories illustrate why source planes must remain separate:
Table 1: Selected retrieval categories across source planes. Reading units include all recorded roles; neither column counts
independently identified animals.
Category Reading units, all roles Mixed Erdman pages Raw XML
Worm and declared derivatives 468 85 2,354
Ant / emmet family 37 8 230
Fly / flies 211 61 4,881
Spider 78 20 617
Bee 45 10 104
Polypus family 73 22 110
Snail 0 2 93
tbl. 1’s reading column includes unverified passages. Its XML column includes catalog prose, code-like identifiers, apparatus
and repeated source descriptions. The snail’s absence from the reading plane does not erase the Four Zoas passage at E323.
Nor can a large XML fly count establish that Blake describes thousands of flies. The different planes serve different discovery
functions.
3.1 Words, designs and the material page
An animal word and a depicted animal are different kinds of evidence. A transcription can establish that a poem names a fly
without establishing that a particular mark in its design depicts one. Conversely, a visually prominent body can matter without
supplying an additional lexical occurrence. The illuminated page makes this distinction productive: lettering, vegetation, figures
and coloured grounds guide the movement through the poem, sometimes offering an analogy and sometimes withholding the
object that a reader expects to find. The relevant question is therefore not simply whether the picture confirms the text, but
what each medium enables the reader to notice, and what remains uncertain after they have been considered together.
Material scholarship also discourages treating colour as a transparent record of a single original conception. Essick and Viscomi
distinguish impressions printed in one ink and subsequently coloured from colour printing, in which colour is transferred from
the plate. They also show how cropping, straightening and mounting can conceal the relation between an impression and its
sheet. Their argument concerns printing processes and bibliographical evidence; it does not license assigning an intention to
each difference between two digital reproductions. ( Essick and Viscomi , 2002, pp. 74–76, 91–92) Here the complete captured
scans remain available beside the resized figures. Details are enlarged only with their crop boundaries recorded. None has been
rotated or recoloured.
The distinction matters particularly for small lives. Enlarging a minute shape may make its outlines easier to inspect while
also detaching it from the proportions and verbal setting through which it becomes meaningful. The four Songs impressions
reproduced here belong to copy Y at the Metropolitan Museum of Art. Their text is checked in that copy; they do not inherit the
copy-A reading credit of the earlier casebook. The Thel comparison likewise names each physical witness. This is an addition
of seven selected visual witnesses, not a representative sample of all copies or a visual census of Blake’s animals. ( Blake, 1825,
1795, 1818)
3.2 Historical names and conditional classification
The modern reference contains 37 admitted NCBI taxon records and an explicit mapping policy for all 102 historical search
categories. Each admitted name, rank and ancestry is checked against two retained EFetch responses, identified by retrieval
URL, access date and SHA-256. The schematic compresses omitted ancestors while preserving captured ancestry. Its layout
comes from the shared validated evidence-tree API. This operation uses no sequence alignment, evolutionary model or dating
analysis. NCBI explains that its classification uses literature-based curation and that its Common Tree is an abbreviated
classification display. These captured records are a versioned reference, subject to revision, rather than a final nomenclatural
authority. ( National Center for Biotechnology Information , n.d.; National Library of Medicine , n.d.)
8

## Page 9

Conditional modern animal senses connect fly to Diptera, spider to Araneae, ant/emmet to Formicidae and bee to Anthophila.
These are reference lineages, not decisions that every retrieved token refers to an organism in that lineage. Anthophila is
more selective than treating all Apoidea as bees. The explicit emmett and emmitt query expansions remain unattested in
these captured matches; they have not been erased from the old registry. The old insect retrieval group for larva is retained
historically, while the mapping policy marks the life-stage term as unresolved across animal groups. A query tag is not a
biological classification.
The five mapping states distinguish a resolved reference lineage, a broad group, an ambiguous mapping, a non-taxon lead and
an outside-scope label. Worm is a polyphyletic literary term whose possible animal senses cross annelids, flatworms, nematodes
and insect forms; these examples are not exhaustive. Glow-worm cannot be assigned to Annelida by its final word. Blindworm
may fall outside the invertebrate scope. Polypus, zoophyte and animalcule retain historical and figurative uncertainty rather
than acquiring species names. Parts, products and actions such as wing, silk, honey and spin receive no taxon target. No
category total is allocated to a taxon by this policy, and “invertebrates” is not presented as a clade.
The lineage diagram is one side of a three-part inquiry: identify the printed expression, read its local use, and only then consider
a modern classification. These relations are neither a linguistic family tree nor a history of influence. Similar spelling can join
unlike bodies; unlike words can name an animal family. The opening visual arguments make these distinctions concrete in the
same material witnesses used for their quotations.
Table 2: Words, local acts and conditional reference. Physical transcription locators retain copy Y or F; the table makes
interpretive distinctions, not new corpus decisions. ( Blake, 1825, 1795).
Printed expression Local operation Conditional biological lead What remains unresolved
Emmet, “A Dream,” Songs Y
26.05
A named wanderer needs a
route home
Ant family, Formicidae No species or completed
rescue
glow-worm, Songs Y 26.16 A compound name; a speaker
offers light
Context must precede a
narrower assignment
The suffix does not supply
an earthworm lineage
“A fly like thee?” Songs Y
40.08
A reciprocal human/animal
comparison
Diptera only for the animal
sense
Neither the speaker nor every
likeness becomes a fly
“That flies in the night”
Songs Y 39.05
A verb qualifies the invisible
worm
No additional fly
identification
Motion does not determine
the worm’s anatomy
Catterpiller and Fly , Songs Y
47.17
Feeding on the tree of
Mystery
Conditional insect
associations
Poetic feeding does not
diagnose an ecological
interaction
“like an infant” Thel F 06.04 Thel compares the Worm
with infancy
No allocation to a worm
lineage
Comparison and depicted
form remain distinct evidence
4 Provision and injury in the Songs and Gates
The earlier casebook read all verse lines of these four poems in copy A. The present comparison adds separate complete
transcriptions in copy Y and the four Met impressions reproduced below. The copy-A textual credits remain intact; no
full-copy reading or all-copy visual comparison follows.
4.1 A Dream: assistance without a completed rescue
“A Dream” makes homecoming depend on several kinds of agency. The emmet speaks her own distress; the dreamer hears
it; the glow-worm can illuminate the ground; the beetle provides an audible orientation. On the reading proposed here, the
poem’s coordination is organized around an unresolved need. The closing imperative offers a route home; it does not narrate a
completed rescue. The glow-worm’s office as watchman also leaves an alternative reading through assigned social roles. Both
readings must account for the sequence of distress, the speaker’s tear, practical advice and the ending. This is an interpretive
proposal about the poem’s sequence, not a biological claim about ant navigation or a measured social network. The glow-worm’s
familiar association with illumination is insufficient by itself: what matters is when and for whom its capacity becomes useful.
In copy Y’s “A Dream,” orange-brown lettering occupies almost the entire designed field, while curling stems, tufts and leaves
pass around and between its stanzas. Blue washes gather along parts of the margins. The reader’s route through speech is also
a route through this vegetation. A small upright, pale figure at the lower right becomes clearer in a bounded enlargement, but
its identity is less secure than the text’s naming of the glow-worm and beetle. It would be misleading to turn every adjacent
mark into an insect. What can be seen with greater confidence is a page that makes the written journey dense with branching
and interlacing forms. [ Blake (1825), plate 26, object 26, physical lines 3–22; Met 17.10.26]
William Gould’s An Account of English Ants supplies a historically available contrast, without establishing that Blake read this
book. At the beginning of chapter VIII, Gould presents ant labour as an example of public care and an admonition to human
indolence. But that moral use coexists with a demand for repeated observation. Later in the chapter he disputes inherited
9

## Page 10

Figure 4: Animal words lead to classification only through a contextual reading. The 199 primary reading units supply 2,498
category-source-span candidates across all 102 categories. Displayed counts name individual categories and are never allocated or
summed to taxa. Solid branches compress unshown classification ancestors; spacing is schematic. Teal dashed links are conditional
animal senses; amber dotted links are ambiguous. Worm spans several lineages; Polypus and other historical forms retain unresolved
alternatives. This is not a sequence-inferred phylogeny, identification, complete animal tree, language genealogy or monophyly claim.
The emmet’s need in A Dream, the reciprocal questions in The Fly, and the Worm/infant comparison in Thel remain different literary
acts ( Blake, 1825, 1795; National Center for Biotechnology Information , n.d.).
10

## Page 11

Figure 5: Hearing precedes help in A Dream ( Blake, 1825). Bentley plate 26; Archive object 26; Erdman E16; accession 17.10.26.
Vegetation passes around and between the stanzas; the small lower-right figure remains uncertain. William Blake, Songs of Innocence
and of Experience, copy Y; work [1794], printed ca. 1825 (Archive: 1825). The Metropolitan Museum of Art, Rogers Fund, 1917. Public
Domain / Open Access (CC0). Whole captured scan proportionally resized; no recolouring, rotation or taxon identification from
ornament.
accounts of winter granaries among the English ants he has examined, while reserving the possibility of different behaviour
elsewhere. ( Gould, 1747, pp. 66–68, 78–81) Natural history here is capable both of making social examples and of correcting
exemplary stories. Blake’s lost mother need not be reduced to either a lesson in industriousness or a report about ant behaviour.
Her interrupted journey places need before usefulness, although the glow-worm’s assigned watch still permits a reading through
established social offices.
4.2 The Fly: consolation after the creature’s injury
“The Fly” makes another relation possible through the thoughtlessness of a human hand. Its human/animal comparison can
support sympathy, but the comparison also raises a difficulty: does the speaker attend to the fly’s life or use it to think about
his own mortality? The poem’s alternating questions give reason to retain both possibilities. A review record should therefore
distinguish an explicit fly reference from its rhetorical function and from the reader’s judgment about the speaker. The source
anchor is Archive Songs copy A, object 33; the retained Erdman text begins at E23, record 33. In this advancement the complete
copy-A physical transcription, lines 3–22, was read: the final happiness follows an explicit conditional relation between thought
and life. It is the speaker’s conclusion, not testimony from the fly about its injury. Full-poem transcription credit does not
include the design, other copies or all of Songs. ( Blake, n.d.e, object 33, physical lines 3–22)
11

## Page 12

Figure 6: Reciprocal likeness follows the speaker’s injury in The Fly ( Blake, 1825). Bentley plate 40; Archive object 40; Erdman E23;
accession 17.10.40. Human support and racket play coexist with the poem’s named fly; the shuttlecock-like mark is not identified as a
fly. William Blake, Songs of Innocence and of Experience, copy Y; work [1794], printed ca. 1825 (Archive: 1825). The Metropolitan
Museum of Art, Rogers Fund, 1917. Public Domain / Open Access (CC0). Whole captured scan proportionally resized; no recolouring,
rotation or taxon identification from ornament.
“The Fly” makes the relation between text and design more difficult. In copy Y, a large adult figure bends towards a small child
and holds its raised hands. At the left another figure raises a racket towards a shuttlecock-like object. Trees enclose the group,
and the poem occupies two upper columns. The most legible bodies are thus human bodies engaged in play and support, not an
anatomically resolved fly. [ Blake (1825), plate 40, object 40, physical lines 3–22; Met 17.10.40] These actions can intensify the
analogy between human and insect play. They can also expose a disparity: the child is visibly supported, whereas the creature
brushed away has no comparable helper within the depicted scene. Neither possibility proves that the adult illustrates the
speaking “I. ” The design instead keeps care, diversion and vulnerability beside a poem whose ending returns to the speaker’s
happiness. The fly’s fate remains less fully told than the human argument made from it.
4.3 The Sick Rose: secrecy, desire and uncertain embodiment
“The Sick Rose” offers a compact retrieval control. The worm is an explicit animal-form reference; flies is the verb describing
its movement, not evidence for a second animal named a fly. Both spans belong in the candidate concordance, but they require
different decisions. The worm’s invisibility and destructive intimacy also prevent a species identification from resolving the
poem. A literal insect, a figure for desire, and a more general destructive agent are interpretive alternatives to evaluate, not
labels the tokenizer can settle. Source: Songs copy A, object 48; E23, record 33. The complete two-stanza transcription was
12

## Page 13

read for the present casebook. Its movement from sickness to the found bed and destructive secret love offers no diagnostic
anatomy with which to resolve the agent. ( Blake, n.d.e, object 48, physical lines 3–10)
The design of “The Sick Rose” offers bodies without making the poem a diagnostic report. In copy Y, thorned stems and
serrated leaves enclose the text; small bent figures occupy the upper vegetation. A larger exposed human form curves out of the
lower blossom beside a narrow, bent, apparently segmented shape. A detail crop permits closer inspection of that adjacency,
but not a secure species identification. The bodily and botanical forms overlap too closely for the design to supply a simple
labelled predator and host. [ Blake (1825), plate 39, object 39, physical lines 3–10; Met 17.10.39] In the verse, the worm’s
destructive movement is explicit while its invisibility is part of the problem. Giving the design an unqualified zoological label
would prematurely close the uncertainty through which the verbal address works.
Michael Srigley supplies a substantial counterreading to a straightforward encounter with an insect. He connects the nocturnal
worm to traditions of spectral beings and to Paracelsian explanations of illnesses produced by disordered imagination. On his
account, sexual repression and solitary fantasy make the destructive love intelligible. The article reconstructs several possible
channels through which Blake might have encountered such ideas, but does not demonstrate a particular act of reading the
proposed source. ( Srigley, 1992, pp. 4–8) Its historical hypothesis should remain a hypothesis. Nevertheless, it changes the
question: the creature may embody the damage of a social and imaginative relation rather than describe an external organism
entering an otherwise innocent world.
That alternative need not expel the animal word from consideration. The poem can be an invertebrate passage precisely
because the worm’s available senses are being made unstable. A literal animal reading must explain the secret love; a sexual
or psychological reading must explain why this injury takes worm-like, nocturnal and mobile form. The two stanzas are not
a code with one expendable noun. Their second-person address also matters: the rose is told of its sickness and destruction,
but does not answer. The reader hears an attribution of harm rather than an exchange in which both parties describe their
relation.
4.4 The Human Abstract: pity and the nourishment of cruelty
The caterpillar and fly of “The Human Abstract” complicate a uniformly tender reading. They feed within an account of fear,
cruelty and mystery that ends in the human brain. A contextual annotation must preserve the predication and setting rather
than importing the sympathy of “A Dream” because both passages contain insects. Reading the complete transcription in this
pass makes the final location of the sought tree in the human brain part of the feeding scene’s context. The caterpillar and fly
therefore cannot supply an empirical plant-host relation through their nouns alone. Source: Songs copy A, object 41, physical
lines 3–26, especially 17–18 and 23–26; E27, record 37. ( Blake, n.d.e, object 41)
“The Human Abstract” resists the opposite simplification, in which insects always stand for persecuted innocence. Its caterpillar
and fly feed on Mystery after Cruelty has prepared a snare, watered the ground and brought forth rooted Humility. The final
stanza locates the tree in the human brain. In copy Y, an aged, bearded figure raises his arms amid cord-like forms under the
text; a dark trunk and projecting growths occupy the margins. These visible relations support an interpretation of entanglement,
without requiring the figure to be named or the cords to be identified as an insect web. [ Blake (1825), plate 47, object 47,
physical lines 3–26; Met 17.10.47]
The opening attack on pity that depends upon making another person poor prevents compassion from remaining an unexamined
virtue. Read beside “A Dream,” it asks whether assistance answers a prior need or benefits from maintaining dependence. The
feeding insects are implicated in a growth whose nourishment is morally compromised; they do not independently explain its
origin. This distinction keeps appetite from becoming either automatic guilt or automatic innocence. What matters is the
relation between feeding, the conditions that sustain it, and the voice that makes those conditions intelligible.
4.5 Gates of Paradise: bodies between copies and captions
Two leaves from For Children: The Gates of Paradise , copy E, complicate a reading of Blake’s invertebrates as either zoological
specimens or freely interchangeable symbols. Yale’s frontispiece, What is Man! (B1978.43.1484), places an arched, segmented
creature above a leaf bearing a human-faced, tapering body. The juxtaposition makes larval and human forms available to the
same act of looking. It does not identify a species, demonstrate a biological transformation, or establish that the two bodies are
successive states of one individual. The human face matters: comparison does not require an observer to erase the particular
being whose vulnerability makes the comparison consequential ( Blake, 1793c).
The accompanying text must be kept specific to its witness. The inspected copy-E impression carries the question and a
publication inscription dated 17 May 1793. It does not carry the light-and-organ couplet that ASU’s Erdman transcription
presents under the frontispiece of For the Sexes (E260). The latter transcription also places “The Catterpiller on the Leaf /
Reminds thee of thy Mothers Grief” before the numbered Keys (E268). The spelling is already represented in the declared
retrieval vocabulary; this reading supplies no newly discovered lexical form. More importantly, a later textual frame is not
silently printed onto an earlier photographed leaf. The mother’s grief directs attention to embodied dependence, while the face
and leaf keep that dependence materially visible. Neither observation turns an emblematic body into a taxonomic identification
(Blake, n.d.c).
13

## Page 14

Figure 7: An invisible agent resists visual diagnosis in The Sick Rose ( Blake, 1825). Bentley plate 39; Archive object 39; Erdman E23;
accession 17.10.39. Human, floral and narrow bent forms overlap; the design does not diagnose the invisible worm. William Blake,
Songs of Innocence and of Experience, copy Y; work [1794], printed ca. 1825 (Archive: 1825). The Metropolitan Museum of Art, Rogers
Fund, 1917. Public Domain / Open Access (CC0). Whole captured scan proportionally resized; no recolouring, rotation or taxon
identification from ornament.
14

## Page 15

Figure 8: Feeding belongs to a human-made tree in The Human Abstract ( Blake, 1825). Bentley plate 47; Archive object 47; Erdman
E27; accession 17.10.47. An aged figure raises his arms among cord-like forms; neither a character name nor an insect web is inferred.
William Blake, Songs of Innocence and of Experience, copy Y; work [1794], printed ca. 1825 (Archive: 1825). The Metropolitan
Museum of Art, Rogers Fund, 1917. Public Domain / Open Access (CC0). Whole captured scan proportionally resized; no recolouring,
rotation or taxon identification from ornament.
15

## Page 16

Figure 9: Human likeness and transformation in two objects from For Children: The Gates of Paradise , copy E, catalogued 1793. Left:
What is Man! , frontispiece, B1978.43.1484. Right: Alas!, museum plate 9, B1978.43.1492; the leaf displays 7 below the image and
handwritten 8 above. The later wording discussed in the text is not present on these photographed leaves. Yale Center for British Art,
Paul Mellon Collection; museum records label both Public Domain and link CC0. Museum-provided recto crops, with complete designs
and captions; no author recolouring or reconstruction. These are selected literary and visual witnesses, not species identifications.
Yale’s Alas! (B1978.43.1492) supplies a different difficulty. A larger human figure raises a hat; small airborne forms occupy
the space above, while another small humanoid figure lies below. The gesture and the fallen body invite comparison with the
casual force of the hand in The Fly , but that is a proposed reading across works, not proof that the smaller figures are flies.
Human proportions, aerial movement and insect association can coexist without resolving into one literal animal referent. A
record can preserve that association while declining a species or order-level assignment ( Blake, 1793b).
Here, too, numbering and text diverge across documentary frames. The museum calls the object plate 9; the photographed leaf
displays 7 below the image and a separate handwritten 8 above it. ASU’s For the Sexes transcription labels the corresponding
passage 7 (E263), surrounding Alas! with additional wording about the figures and the divine image. Those additions are
not visible on the inspected copy-E leaf. A usable locator therefore records the work, copy, accession and numbering system
together. “Plate 7” alone cannot do the work of all four.
Höltgen places Gates beside emblem-book practices while stressing how its compressed captions complicate a conventional
emblem’s explanatory closure. That argument offers a relevant counterweight to treating the designs as illustrated natural
history: a caption, an image and an interpretation are related evidence, not synonyms. The present comparison adopts neither
a general influence claim nor a complete decoding of Blake’s figures ( Höltgen, 2002).
Höltgen’s specific reading gives the alternative more force: his butterfly-girls and the youth’s capture place the scene within
a mortal, spiritless material existence. That is a serious challenge to treating aerial motion as uncomplicated liberation. His
quoted expanded wording is a different documentary frame from this early leaf’s sparse caption. The photograph cannot settle
his whole interpretation, nor can the interpretation establish a taxon. The conjunction of gesture, fallen body and later verbal
framing remains a literary argument rather than a recovered observation of insects. ( Höltgen, 2002)
These two selected objects extend the visual casebook rather than the fixed corpus analysis. Their official photographs and
object records have separate hashes, credits and rights statements. Museum-supplied recto crops retain the complete engraved
designs and captions; the wider captures with calibration targets remain available for inspection. The two objects count as two
witnesses, not as multiple witnesses for each digital rendition. They add no baseline text hits and no formal textual adjudications.
Their methodological contribution is narrower: preserve an association when it is warranted, preserve an unresolved identity
when it is not, and make the edition-specific evidence recoverable.
16

## Page 17

5 The wine-press: abundance beside unequal suffering
The large inventory in Milton names the earth-worm, beetle, emmet, centipede, spider, spinner, earwig, maggot, flea, louse,
bug, tape-worm, slug, grasshopper, scorpion, gnat, wasp, hornet and honey bee. It places them among seeds, roots, plants and
vertebrates around Luvah’s wine-presses. Source: plate 27 [29], E124, record 134; Archive copy A, object 24, physical text lines
13–22. Pinned XML .
The passage combines dance and celebration with disease, death and the violent labour of the press. The human grapes’
suffering makes a simple ecological harmony reading inadequate. The same assemblage can display abundant life and unequal
exposure to torment. This article accordingly asks who moves, sings, feeds or suffers in an animal scene, and whether those
positions can change. An occurrence count alone answers none of these questions. The newly inspected object24 lines 2–42
sharpen the distinction between positions around and in the press. Its opening joins war to printing; the late human grapes do
not participate in the surrounding dance. The nettle that stings and the creeping weeds also provide non-taxon controls inside
the same context. These verbs cannot be assigned to insects by an associated-imagery search category. ( Blake, n.d.d, object
24, physical lines 2–42)
The related Four Zoas inventory at E404, manuscript page 136, lines 29–34, belongs to another compositional context. It
should be connected to the Milton inventory for comparison while retaining both work identities. This is a different operation
from recognizing repeated copies of a single poem. The protocol therefore preserves source repetition and separately records
relations between passages. A future account of revision or reuse needs textual and manuscript scholarship beyond exact-text
deduplication.
The snail passage at E323 supplies a further counterweight. Worm, snail, emmet and beetle participate in mockery of Los.
Their smallness does not ensure innocence, nor does the presence of Los ensure a stable human superiority. The scene motivates
reading agency and evaluative position alongside animal name. Source: retained Erdman record 335, opening lines of E323.
This juxtaposition changes how the wine-press inventory can be used. The abundance of named life around the press does not
ensure that every participant occupies a position of reciprocal benefit. The inventory must remain beside the human grapes and
their suffering. ( Blake, n.d.d, object 24, physical lines 2–42) Nor does the mere presence of predation establish cruelty in the
same sense as a deliberately inflicted injury. The article can instead follow three separate relations: the provision that enables
another life, the feeding that uses a living body, and the aggregation that reorganizes its members’ powers. Their overlap is an
interpretive problem, not evidence that they are morally identical.
6 Recognition, generation and collective bodies
6.1 Thel: recognition and nurture through an infant analogy
The earlier bounded reading of copy B, objects 6–7, physical lines 3–14 and 2–18, remains a separate textual witness. It
established the movement from a bruising foot to positive nurture, with maternal and nuptial language qualifying any universal
social doctrine. The following F/O comparison extends that local question through independently identified material witnesses.
(Blake, n.d.f)
The Worm’s presentation in The Book of Thel brings analogy into the act of recognition. Thel asks what she sees, calls it weak,
compares it to an infant, and hears weeping where articulate speech is unavailable. In both reproduced impressions of Bentley
plate 6, an infant-like body rests among long leaves below a floating figure and beside a standing, gowned figure with extended
arms. The bodily scale and placement make vulnerability visible before they permit any zoological resolution. ( Blake, 1795,
object 6, physical lines 2–13; Blake, 1818, object 5, physical lines 4–15) The scene can awaken concern by translating another
life into a familiar human form. It can also make that concern conditional upon resemblance to a child. Recognition through
analogy is consequently an ethical achievement with a limit, rather than transparent access to the Worm’s own experience.
The response attributed to the Clay changes the situation. Thel’s statement that no one cherishes the helpless being is answered
by care already becoming active. In the following reproduced plate from copy O, a reclining figure and an infant appear below
the poem and in front of a larger seated figure. The spoken acknowledgment distinguishes a prohibition on wilful bruising from
the positive provision of milk and oil. ( Blake, 1818, object 6, Bentley plate 7, physical lines 3–19) A community of care is
measured here by what sustains a dependent life, not merely by punishment after injury. Yet maternal, nuptial and hierarchical
language shapes that provision. The passage cannot by itself guarantee equal participation or make an obligation to nurture
independent of its gendered terms.
Perkins places the infant Worm within a contemporary rhetoric that mobilized tenderness towards humble creatures, while
preserving the possibility of irony in the poem’s surrounding world. ( Perkins, 1999, pp. 5–6) That caution helps prevent the
scene from becoming a frictionless model of mutual flourishing. Care can be real within the local exchange even if its speakers
cannot settle Thel’s larger unease. The selected plates support an account of this exchange; they do not settle the meaning of
every stage of her journey.
The copies themselves also require different forms of attention. Rosenwald 1798A is copy F, object 6; Rosenwald 1798B is
copy O, object 5. Both are Bentley plate 6, corresponding to Erdman plate 4. Object 6 in O is instead Bentley plate 7. In
17

## Page 18

Figure 10: The Worm/infant comparison in two material witnesses ( Blake, 1795, 1818). William Blake, The Book of Thel (work
imprint 1789), Bentley plate 6 / Erdman plate 4. Left: copy F, Archive object 6, LOC Rosenwald 1798A; Archive printing 1795, LOC
ca. 1794 (1794 watermark). Right: copy O, Archive object 5, LOC digital Rosenwald 1798B; Archive c. 1818, LOC ca. 1815 (1815
watermarks). O is bound with Milton D under Archive catalogue no. 1810; that volume number differs from its digital subresource.
Library of Congress, Rare Book and Special Collections Division, Lessing J. Rosenwald Collection. Historical public-domain artwork;
LOC states no known US copyright or other restrictions, with user determination required. Whole scans proportionally resized; neither
physical scale nor photographic colour is calibrated.
the matched reproductions, F gives the standing figure a conspicuous yellow gown against a relatively continuous ochre upper
field, while O shows a paler gown and a more extensively lined and coloured setting. O’s painted enclosure is more pronounced.
These are observations about the captured images. Unequal reproduction conditions, paper condition and colour management
prevent attributing every tonal difference to an intentional revision. The repeated arrangement is an anchor for comparison,
not permission to merge the two impressions into a single generic image of Thel.
6.2 Polypus: joining and the powers of a body
The historical polypus should first complicate the word “individual. ” In the second edition of Zoonomia examined here, Erasmus
Darwin describes young polyps branching from a parent’s side and subsequently separating. He compares them with vegetable
buds and interprets their generation through a theory that assigns formative priority to the male. Elsewhere in the same
discussion, gnats and tadpoles become examples of bodily transformation, while the polypus remains associated with budding.
These statements belong to Darwin’s historical physiology, including assumptions that cannot be accepted as modern zoological
description. ( Darwin, 1796, vol. I, section XXXIX.II.3 and XXXIX.IV.2) They nevertheless show why generation could bring
together questions about attachment, separation, sexual difference and the production of new powers.
Dauphin’s extended Blake discussion traces the poetic reworking of this complex into vegetation and collective bodies. Her
reading connects the multiplication of organic forms to a danger of losing distinct identity. Her argument also insists that the
18

## Page 19

Figure 11: Care extends beyond the prohibition of injury ( Blake, 1818). William Blake, The Book of Thel (imprint 1789), copy O,
Archive object 6, Bentley plate 7 / Erdman plate 5; LOC digital Rosenwald 1798B, bound with Milton D, Archive catalogue no. 1810.
Archive printing c. 1818; LOC ca. 1815. A reclining figure reaches towards an infant-like body beneath the selected Clay/Worm
exchange. Library of Congress, Rare Book and Special Collections Division, Lessing J. Rosenwald Collection. Historical public-domain
artwork; LOC no-known-US-restrictions statement retains user-determination qualification. Whole captured scan proportionally resized.
This is the following plate, not a second copy of Bentley plate 6.
19

## Page 20

scientific creature and its poetic transformations are not interchangeable. ( Dauphin, 2026, paragraphs 7–28) It is useful as an
account of transformation across discourses. It does not independently establish that any particular branching ornament in an
illuminated book depicts a polypus, nor prove a single direct route from Darwin to every occurrence in Blake. The reproduction
of a Thel infant and the mention of a prophetic Polypus remain different evidential objects.
In the selected Milton passage, the joining of every born man into one Polypus is situated within vegetated nerves, wailing
souls and changes in perception. The figure is named Orc in the same sentence; the surrounding lines relate Orc, Satan and
Luvah rather than leave a stable zoological emblem. ( Blake, n.d.d, object 28, physical lines 24–35) The problem is not simply
that many lives are connected. It is what connection does to their capacities to perceive, act and remain distinguishable. The
emmet’s helpers retain separate offices and answer a particular voice; the prophetic passage joins persons within a body whose
powers exceed any isolated participant. That contrast is locally supported, but the named transformations make a universal
opposition between good networks and bad organisms too blunt.
Erasmus Darwin’s later Temple of Nature further unsettles a simple opposition between scientific materialism and ethical atten-
tion. Its third canto brings human pride down towards kinship with emmets and worms after considering insect construction,
sensation and volition. Its fourth canto presents feeding across animal and vegetable life as pervasive destruction, extending
appetite to creatures conventionally imagined as gentle. [ Darwin (1803), canto III, lines 411–434; canto IV, lines 1–66] Kinship
and violence therefore inhabit the same natural history. Fara’s reading of these passages emphasizes the coexistence of ethi-
cal cooperation, progressive development and an account of death that regulates abundance. Her analysis also distinguishes
Erasmus’s historical explanations from the later theory of his grandson. ( Fara, 2020, paragraphs 21–23, printed pp. 10–12)
Dauphin’s account of the changing Polypus makes it especially important to distinguish the biological animal, the poetic figure
and later iconographic interpretation. The Archive’s copy-A Milton descriptions also retain alternative identifications: object 24
can be described through caterpillar/centipede and beetle/bee/hornet alternatives, while object 31 permits vegetable/Polypus
possibilities. A catalog’s alternatives cannot become several confirmed depicted species. Those copy-A descriptions have been
read; the corresponding Milton images have not been inspected in this expansion. ( Dauphin, 2026); Archive Milton A .
Historical lexicography remains an acquisition task. The comparison advanced here asks what joining permits and suppresses
in a particular scene. It does not equate Blake’s changing collective figures with modern political institutions.
6.3 Webs, prisons and the allocation of danger
The prison towers of The French Revolution make it difficult to attach one moral value to a creature or its structure. Within the
selected passage, a man is confined for prophetic writing, another has refused to sign papers of abhorrence, and a woman has
resisted a minister. The named towers assign institutional categories to persons whose acts the narrative also preserves. The
old man in the tower called Order occupies a den as narrow as a child’s grave, among webs and the slime of ancient horrors. The
surrounding snakes and scorpions, however, are described as harmless companions sharing his sorrowful breath. The grammar
does not extend that description automatically to every spider. It nevertheless prevents the tower’s animal occupants from
simply bearing the blame for its confinement. ( Blake, n.d.b, bounded Archive reading windows E29-17 and E29-27)
This is a local test of the relation between biological symbolism and institutional entrapment reported in Hoagwood’s review of
Baine. A web can make the prison’s accumulated enclosure visible without making a spider the author of the prison. Conversely,
retaining an animal sense does not turn the webs into a natural-history observation. Material habitation, political analogy and
the narrator’s allocation of suffering coexist. The adjacent eternal worm in the tower Bloody introduces another relation to
the confined body; its eschatological and bodily senses remain unresolved in the existing uncertain decision. The two retrieved
categories therefore lead into overlapping context, but neither their proximity nor their shared setting makes them one symbolic
function. ( Hoagwood, 1989, pp. 40–42)
The distinction matters beside the practical assistance in “A Dream” and the joined body of Milton. Association can answer
distress, accompany it without remedy, or reorganize the powers of those joined. The old man’s companions do not release him;
their harmlessness cannot establish the harmlessness of Order. Nor does the glow-worm’s office establish that every assigned
social place enables another life. What an association does must be read in its sequence and distribution of powers. This
comparison proposes no equivalence between these poetic institutions and a modern constitutional procedure, and it does not
resolve the unread Baine chapter. Its narrower result is that naming a web, finding an animal sense and identifying the agent
of an injury are different interpretive operations.
7 Seeing, injuring and judging
7.1 Microscopic display and the failure of demonstration
The failed demonstration in An Island in the Moon is sharper when placed beside practical microscopy rather than an undiffer-
entiated idea of science. In the second edition of Henry Baker’s The Microscope Made Easy , the observer must suit illumination
to the object, examine it in different positions and distinguish the image projected by a solar microscope from the body’s
natural colour. Baker also asks that a living animal be squeezed or disturbed as little as possible. His preparation instructions
combine methods of holding creatures alive with pinning or other restraint when size requires it. ( Baker, 1743, pp. 53–58) The
20

## Page 21

manual does not stand outside human use of animals; it makes the conditions of useful visibility explicit, including the tension
between preserving a creature’s behaviour and securing it for inspection.
In Blake’s chapter 10, by contrast, the speaker offers a louse, flea, butterfly or cockchafer before water, a smearing handkerchief,
broken apparatus and a panicked account of escaped pestilence displace the promised demonstration. The sequence is not a
successful observation followed by an abstract objection to empirical knowledge. It is a performance in which the promised
objects never become reliable shared evidence. ( Blake, n.d.a, E462, provider record 474, chapter 10 only) Baker’s practical
cautions make the failure legible without making his manual a demonstrated source for the satire. Both texts allow the means
of seeing to matter; they give those means very different dramatic and argumentative functions.
7.2 Senses, cells and the limits of emancipatory speech
In Visions of the Daughters of Albion , the question “With what sense does the bee form cells?” belongs to a sequence that
differentiates creatures by their habitations, pursuits and pleasures. Oothoon’s questions move through the chicken and pigeon,
bee, mouse and frog before the wild ass, camel and other beings. Shared organs do not provide a sufficient explanation of their
different activities. The sequence ends by making that inquiry prior to an account of human thought. The bee is neither a
specimen whose behaviour has been measured nor a replaceable decorative comparison. Its cell-making contributes a distinct
activity to an argument about the limits of a uniform account of sensing. ( Blake, 1793a, object 5, physical lines 3–14)
Hutchings reads this differentiated animal activity as a condition of human self-reflection and relates it to the poem’s challenge
to imperial uniformity. His comparison with Swedenborg distinguishes compelled behaviour from Blake’s language of creaturely
pleasure; the cited Swedenborg and Raine texts have not been independently read here. His later objection is equally consequen-
tial: Oothoon’s apparently generous sexual vision can reproduce capture and the appropriation of other people. Emancipatory
speech does not secure its own emancipatory effects. These are claims from selected paragraphs of Hutchings’s essay, not a
reading of his monograph or the complete essay. ( Hutchings, 2001, paragraphs 19–22)
For this article, the consequence is a test of who benefits from the analogy. The bee question opens a space for capacities
that the human observer cannot explain through an inventory of organs alone. Yet the next lines return to Oothoon’s desire
for Theotormon’s attention, before he begins his own questions. The passage’s sequence keeps an argument about animal
difference beside a specific, unanswered human address. ( Blake, 1793a, object 5, physical lines 15–26) This local transition does
not disprove the expansive claim; it prevents the claim from settling the scene. The later silk-net judgment discussed in sec. 8.7
supplies a separate, explicitly coded case of connective material proposed for capture. The present transcription reading creates
no new candidate decision, inspected image or complete-work credit.
7.3 Animal suffering and the force of an augury
“Auguries of Innocence” poses a further difficulty: what should follow when injury has been noticed? Perkins argues that its
animal auguries voice innocence confronting cruelty and threatening consequences that exceed ordinary causal explanation. He
also warns that treating a tormented creature solely as a metaphor can diminish attention to the animal’s suffering. These are
interpretive positions, not an uncontested identification of the poem’s voice. ( Perkins, 1999, pp. 7–10) They encourage keeping
literal injury present while examining the form in which a speaker makes it consequential.
The boy who kills the fly is threatened with the spider’s enmity. This is not simply a food-chain statement: the two-line address
assigns a consequence to a human act. Yet the creature associated with catching flies becomes the figure through which killing a
fly is answered. Elsewhere the caterpillar repeats maternal grief, and the prohibition on killing moth or butterfly is attached to
judgment. [ Blake (n.d.a), E490–491, records 502–503; “Auguries,” lines 33–40] The sequence crosses animal behaviour, familial
feeling and eschatological warning without making them interchangeable. A consistently gentle insect world would have no
place for the spider’s enmity; a purely predatory one would lose the address to the boy. The poems considered here make small
lives matter through those unresolved relations between attention, use, obligation and response. Their ethical force depends on
retaining the differences that a single animal-symbol key would erase.
The sequence also includes gnat, bee, cricket and emmet, but a lexical resemblance need not name an animal. The Mite at
E492 is an amount extracted from labour, not an arachnid. Broad retrieval preserves that negative control. Its retained locator
is record 504; this expansion adds no new reading credit for that page. The distinction between use, deliberate cruelty and
metaphor requires a speaker, a predication and a setting, not merely a noun. ( Blake, n.d.a)
8 How the collection shapes what can be counted
The shared-method implementation permits two further analyses of the same pinned bundle. The NFKC experiment retains the
original categories and recomputes all primary counts. A separate literal experiment applies explicitly declared word windows.
Both have immutable source-bound receipts; their different profiles and denominators remain separately named. The historical
baseline and its eleven authored decisions retain their original run identity. The recorded installed replay for application 0.3.0
left the five analytical baseline artifacts byte-identical; its producer receipt did not repin those decisions or add contextual animal
classifications. The historical 0.4.0 authoring rebuild consumed the retained analytical outputs under their exact hashes and
21

## Page 22

its then-current SDK pin. It did not claim a fresh mathematical replay or transfer those outputs to a new analytical producer.
The present 1.0.0 review edition preserves those producer identities while consuming the retained exports. Its schema 3 native
host and the source-support analysis below use the separately verified Blake SDK 0.7.0 pin, with distinct receipts for their
current execution.
8.1 A retrieval candidate is not an animal observation
The sensitivity inventory contains the 199 identified Blake-primary reading units, grouped by 33 recorded work identifiers.
These are a restricted subset of the bundle’s 2,067 captured units. The vocabulary still has 102 categories; its 2,498 primary
category-span candidates include direct names, compounds, products, structures and actions. The counting unit remains a
category attached to a source span. Two categories can refer to overlapping language, while one category can retrieve both
animal and non-animal senses.
This distinction matters for the readings already proposed in secs. 4, 5, 6, 7. The fly category cannot decide whether a
retrieved form names an animal or describes motion. The lexical relation between a compound and one of its components does
not establish two creatures. Likewise, the web and honey channels can help locate relevant discussion without establishing a
speaker, a maker or an animal subject. Normalization and explicit forms improve reproducibility of discovery; they do not
resolve these interpretive relations. The animal and non-animal contrasts already discussed in sec. 4 therefore remain review
tasks, rather than exceptions that can be subtracted automatically from a total.
8.2 What weighting changes and what it cannot correct
The pooled rate divides a category’s copy-retaining count by all 644,247 NFKC word tokens in the eligible inventory. The
work-balanced rate first calculates a rate within each positive-token work group and then averages the 33 rates with equal
weight. A work group with no occurrence contributes a zero rate. These quantities answer different questions: frequency in the
retained textual mass, and the average of the recorded work-group frequencies. Neither treats the 33 identifiers as a random
sample of independent works.
Two distinct denominators organize the analysis. Let 𝑛𝑔𝑐 be the copy-retaining number of category- 𝑐 candidates in recorded
group 𝑔, and 𝑇𝑔 its NFKC word-token exposure. On the positive-token groups 𝒢𝑇 = {𝑔 ∶ 𝑇 𝑔 > 0} , pooled and equal-group
token rates are
𝑟𝑐 = 10 4
∑𝑔∈𝒢𝑇
𝑛𝑔𝑐
∑𝑔∈𝒢𝑇
𝑇𝑔
, 𝑟𝑐 = 104
|𝒢𝑇 | ∑
𝑔∈𝒢𝑇
𝑛𝑔𝑐
𝑇𝑔
. (1)
A positive-token group with no match contributes a defined zero rate. A group with no token exposure has no defined token
rate. The present inventory has 33 eligible positive-token groups; the equation specifies the estimands without treating those
identifiers as independent sampled works.
Table 3: Primary retrieval rates per 10,000 NFKC tokens and concentration in recorded groups. Counts retain copies and
overlapping categories.
Category Retained candidates Pooled rate Equal-work mean rate
Largest work-group
share
worm 468 7.264 10.972 36.5%
fly 183 2.841 5.392 56.3%
spider 78 1.211 0.702 46.2%
polypus 73 1.133 0.466 54.8%
bee 44 0.683 0.358 40.9%
ant 37 0.574 0.263 37.8%
caterpillar 36 0.559 0.625 41.7%
butterfly 20 0.310 0.371 80.0%
Rates are category-span candidates per 10,000 NFKC tokens, rounded only for presentation. The worm and fly rates increase
under equal-work weighting, whereas polypus, spider, bee and ant decrease. This reversal prevents a single ordering from
serving as a context-free account of prominence. The largest recorded work-group shares further locate the dependency: thel
contributes 171 of the 468 worm candidates; songsie contributes 103 of 183 fly candidates; and milton contributes 40 of 73
polypus candidates. These are counts across retained source units within each group, not counts of distinct lines or independently
inspected passages. The 80% butterfly share belongs to thel and comprises 16 of 20 retained candidates. It does not, by itself,
establish that the work contains 16 different butterfly passages.
22

## Page 23

The leave-one-work-out calculations offer a related diagnostic. Removing one recorded group at a time gives pooled fly rates
from 1.411 to 3.828 and polypus rates from 0.581 to 1.288 per 10,000 remaining tokens. These are finite-inventory omission
ranges, not confidence intervals or uncertainty estimates for Blake’s oeuvre. Their value is to expose which claims about relative
prominence depend on the retained group composition. They support returning to the different settings of the short lyrics, the
inventories and the Polypus discussed in secs. 4, 5, 6; they do not establish a common function across those settings.
8.2.1 Compositions, omission and concentrated evidence
The retained installed composition run, qualified for application 0.3.0, independently recomputed the full captured atlas, primary
counted inventory and earlier literal grid before producing its composition and proximity exports. Those measurements keep
their recorded producer. The present native render binds the exact consumed exports and draws new introductory diagrams
from them; it does not recompute the composition analysis. Neither operation adds a contextual animal decision or image
identification.
The historical registry partitions into 82 name-family categories, four broad labels, 15 associated channels and one motion
channel. “Name-family” here means the registry groups for worms, insects, arachnids, myriapods, molluscs, crustaceans and
ambiguous life. It preserves historical retrieval policy rather than imposing a new biological classification. Thus it retains
blindworm, spinner, sponge, derivatives of worm, and the noun/verb ambiguity of fly. It must not be read as a set of adjudicated
animal nouns. The other channels retain generic labels, products, structures, parts and actions in their own declared groups.
For each name-family category, the complete matrix reports candidate counts, matched retained units, the eligible unit de-
nominator and NFKC token denominator in every recorded group, including zeros. Within-group composition divides each
category’s count by that group’s sum of name-family category candidates. Pooled composition divides pooled category counts
by the total name-channel mass. The equal-group composition averages the normalized profiles of groups with positive name-
channel mass. A group containing no name-family candidate has undefined composition; it remains a zero in coverage exports
and is excluded from this particular mean. This differs from equal-work token rates, whose denominator is text length and
whose zero-match groups can contribute defined zero rates.
We measure profile differences by total variation: one half of the sum of absolute differences between corresponding category
shares. The same distance compares the full pooled profile with the profile remaining after omission of each entire recorded
group. These distances express redistribution of category-candidate mass, not uncertainty about Blake’s oeuvre. Concentration
is also reported as the smallest number of descending-count groups needed to account for one half or four fifths of a category’s
mass, and as inverse Herfindahl concentration. The latter is an effective number of equally weighted groups, not a number of
independent works.
Categories and overlapping spans remain separate throughout. A compound and its component may each contribute a category
candidate; the denominator is a sum of category candidates, not distinct animal words or organisms. Exact-text sensitivity
comes from the shared SDK’s grouping of identical complete source units. It does not remove repeated poems embedded in
different whole texts.
The relationship analysis declares twelve exploratory pairs of singular literal terms and applies complete word envelopes of 8,
32 and 64 words. Whole-unit co-presence is a separate scope. Public SDK matching and proximity APIs retain original text
boundaries and count matching units before any display cap. Their 644,245 original-text word tokens remain distinct from the
NFKC denominator. These literal pairs do not expand to every form of their similarly named categories: emmet is narrower
than the ant family, and caterpillar does not silently include historical alternative spellings.
Definition 2 (Composition of retrieved name-family mass). Restrict categories to the declared name-family set 𝐶𝑁 .
Write 𝑁𝑔 = ∑ 𝑐∈𝐶𝑁
𝑛𝑔𝑐, 𝒢+ = {𝑔 ∶ 𝑁 𝑔 > 0} and 𝑁 = ∑ 𝑔 𝑁𝑔 > 0 . The within-group profile is 𝑢𝑔𝑐 = 𝑛 𝑔𝑐/𝑁𝑔 for 𝑔 ∈ 𝒢 +. The
pooled profile weights these profiles by retrieved candidate mass; the equal-group profile gives each nonempty group the same
weight:
𝑝𝑐 = ∑
𝑔∈𝒢+
𝑁𝑔
𝑁 𝑢𝑔𝑐, 𝑞 𝑐 = 1
|𝒢+| ∑
𝑔∈𝒢+
𝑢𝑔𝑐. (2)
A group with no name-family candidate has undefined composition and is omitted from this mean, while remaining visible as a
zero in the coverage exports. These profiles distribute category-candidate mass, including overlapping categories; they do not
distribute independently identified animals.
For two profiles on that same category set, total variation is
𝐷TV(𝑝, 𝑞) = 1
2 ∑
𝑐∈𝐶𝑁
|𝑝𝑐 − 𝑞𝑐|. (3)
Proposition 1 (Unequal group sizes do not by themselves force different compositions). Under Definition 2 , if all
nonempty groups share the same profile 𝑢, then 𝑝 = 𝑞 = 𝑢 , even when their candidate masses 𝑁𝑔 differ.
23

## Page 24

Proof. The pooled weights sum to one, so 𝑝𝑐 = 𝑢 𝑐 ∑𝑔 𝑁𝑔/𝑁 = 𝑢 𝑐. A veraging the same 𝑢𝑐 over the nonempty groups also gives
𝑞𝑐 = 𝑢 𝑐 for every category. Thus eq. 3 is zero.
The converse is not asserted: different group profiles can cancel in the weighted comparison. Proposition 1 instead isolates the
interpretive point. A weighting contrast concerns the joint pattern of group mass and lexical composition; equal weighting is
not automatically a correction toward a truer animal census. Its result is an observable sensitivity of this retained collection.
For a positive category, define its group shares by 𝑣𝑔𝑐 = 𝑛 𝑔𝑐/ ∑𝑔′ 𝑛𝑔′𝑐. Its inverse concentration is
𝑁eff,𝑐 = (∑
𝑔
𝑣 2
𝑔𝑐)
−1
. (4)
A zero-mass category has no defined concentration; Appendix A displays that state explicitly. This effective number describes
equally weighted groups with the same concentration, not independently sampled works or animals.
The 82 name-family categories yield 1,239 primary category-span candidates in 26 of the 33 recorded groups. Thirty-eight
categories have positive counts; 44 remain zero in this primary inventory. The associated channels contribute 1,087 further
candidates, the motion channel 156, and broad labels 16. Their sum reproduces the 2,498 primary candidates. Name-family
queries therefore account for 49.60% of the primary category-candidate mass, rather than all of the concordance. Exact whole-
text grouping reduces name-family candidates from 1,239 to 1,238. The changed category is mussel, from 13 to 12. That narrow
change does not establish that the remaining copies represent distinct contexts.
The pooled and equal nonempty-group name compositions have total variation 0.222387. The fly share rises from 14.7700%
pooled to 24.3951% under equal composition weighting, a change of +9.6251 percentage points. Worm falls from 37.7724% to
33.7732%, or −3.9992 points. Beetle falls from 5.0040% to 1.7664%, and Polypus from 5.8918% to 3.5986%. These changes an-
swer a different question from the earlier rate comparison: they redistribute a fixed name-channel profile rather than comparing
candidates per text token.
Two smaller categories expose the cost of normalizing sparse groups. Mussel rises from 1.0492% pooled to 7.6923% in the
equal-group profile. Its one candidate in ahania and twelve in america exhaust the name-channel mass in those two groups, so
each contributes a complete within-group share to that category. A bounded inspection of the retrieved Ahania context finds
the surface “muscles” among bodily formations, not a contextual mollusc identification. Similarly, the single mayfly candidate
contributes one half of the name-channel profile in but828; it is the normalized sequence “may … fly” in the captured Genesis
transcription. The surrounding reference is to fowl and flight. These inspected retrieval ambiguities explain why equal weighting
cannot serve as an automatic correction toward a truer animal census. They also caution that the recorded author/role field is
an eligibility rule, not a fresh attribution of every transcribed word to Blake. 1
The three largest omission effects are songsie (total variation 0.119982), thel (0.103325) and milton (0.079001). They remove,
respectively, 217, 203 and 168 name-family category candidates from the pooled profile. These groups do not simply contribute
the most textual mass: thel contains 23,019 NFKC tokens, whereas jerusalem contains 197,560 tokens and its omission
distance is 0.051693. Composition influence depends on a group’s lexical profile as well as its size. Worm remains the largest
pooled category after every single-group omission. That finite robustness of the leading category coexists with substantial
changes in the rest of the profile.
Concentration measures further distinguish visibility from breadth. Worm and fly each occur in 15 recorded groups, but their
effective group counts are 5.008 and 2.889. Five groups are needed to accumulate four fifths of worm candidates, compared
with four for fly. Polypus occurs in three groups and has an effective count of 2.030; two groups suffice for four fifths of its
candidates. These numbers locate concentrations in this retained collection. They do not count distinct literary episodes or
establish the independence of copies.
8.3 What exact-text equality leaves unresolved
Grouping complete source units by identical text removes one unit from this inventory: the two america records archive:ame
rica.n and archive:america.n--878c90904184 share 2,941 NFKC tokens. The exact-text inventory therefore contains 198
units and 641,306 tokens. The eight selected category counts in the table are unchanged, but their pooled rates change because
the denominator changes. Across all categories, the candidate sum falls from 2,498 to 2,489 through five changes:
1Fresh local inspection was limited to the SDK’s 120-character context on each side of two hits. Ahania: archive:ahania.a , reading-layer
locator lines 1–259, original character interval [5617,5624), source SHA-256 fa95577f5790fc5d2cba1ff08822d79b59d2399f352e2a7f477d89186839576
1. Genesis: archive:but828.1, reading-layer locator lines 1–230, interval [3089,3114), source SHA-256 ef98f4fa8c09feca8dee00110b7a844a7af1a1
bf8ddbc7edd9b5aefc96fd6967. The broad reading-layer line ranges are unit locators, not exact verse-line locators for these hits. Neither inspection
credits a complete work or design.
24

## Page 25

Figure 12: Which names dominate depends on how the sources are weighted. The sixteen largest share shifts compare the pooled
name-channel profile with the mean of 26 nonempty-group profiles (1,239 category candidates); all 33 groups remain eligible for
omission. The twelve largest omission distances describe redistribution of this finite inventory. Parentheses retain candidate counts.
Complete data and all omissions are supplemental; no sampling inference or animal adjudication is implied.
25

## Page 26

Table 4: The five category totals changed by exact equality of complete source-unit text. Partial repetition remains outside
this check.
Category Copy-retaining count Exact-text count
crawl 26 25
flight_verbs 156 153
insect 14 13
mussel 13 12
wing 532 529
The first column names the actual registry categories; the presence of insect and mussel means these changes should not all be
described as contextual-action changes. Exact-text equality detects an identical complete unit. It cannot detect all repeated
poems embedded in nonidentical copies, near-identical editions, shared passages within different works, or differences in editorial
boundaries. The small change in this check therefore supplies no evidence that the remaining copies or reading positions are
independent. In particular, unchanged counts do not turn the concentration results into an edition-controlled comparison.
8.4 Near words and shared source units
A separate experiment searches the raw singular literal pairs shown below. These are not expansions of the NFKC categories:
emmet is one explicit literal term, while the category named ant has a broader declared form inventory; fly in a literal pair does
not include every form in the fly category. The literal engine retains original-text word boundaries and has a denominator of
644,245 word tokens. The NFKC denominator is 644,247. These profile labels and denominators remain separate even though
the two experiments use the same 199 eligible source units in this bundle.
Definition 3 (Finite word envelopes and whole-unit co-presence). Fix the eligible source units, the original-text word-
boundary profile and two declared literal terms 𝑎, 𝑏. Let 𝑀𝑤(𝑎, 𝑏) contain the source units in which at least one occurrence of
each term fits within a complete envelope of at most 𝑤 word tokens. Let 𝑀unit(𝑎, 𝑏) contain units with both terms anywhere
in the unit, without a finite distance restriction. Each set retains source identities and copies. It counts matched units, not
independent scenes, semantic relations or distinct animals.
Proposition 2 (Enlarging a fixed-profile window cannot remove a matched unit). For permitted finite windows
𝑤1 ≤ 𝑤 2, with every other condition in Definition 3 fixed,
𝑀𝑤1
(𝑎, 𝑏) ⊆ 𝑀 𝑤2
(𝑎, 𝑏) ⊆ 𝑀 unit(𝑎, 𝑏). (5)
Proof. An envelope containing both terms within 𝑤1 tokens also fits within 𝑤2 tokens. Either envelope witnesses that both
terms occur in its source unit, which is the whole-unit condition. Taking set cardinalities therefore preserves the same ordering.
This nesting property explains why a larger window can reveal additional co-presence without demonstrating a stronger literary
association. Whole-unit scope remains a different retrieval condition, not a plotted point at an invented infinite window. A
change of corpus, literal forms or normalization profile falls outside Proposition 2 .
Table 5: Matched retained units under fixed literal profiles. Whole-unit co-presence has no finite word-distance restriction.
Literal pair 8 words 16 words 32 words 64 words Whole source unit
worm + fly 4 4 28 28 66
bee + flower 0 0 0 0 30
spider + web 13 13 13 13 33
emmet +
grass
27 27 27 27 36
Cells count matched source units, with copies retained. A finite window requires a complete word envelope containing both
terms. The last column removes that finite-distance requirement; it is a different scope, not a point at 200 words or an
extrapolated infinity. Thus the worm + fly increase from 28 units at 32 words to 66 whole units cannot be read as 38 additional
nearby encounters. The bee + flower contrast is sharper: no finite window through 64 words matches, while 30 complete source
units contain both literals. That result identifies a dependence on source-unit extent, not the absence of bee–flower relations
in Blake. Singular forms, unrepresented variants, source coverage and unreviewed senses all constrain recall.
The early plateaus for spider + web and emmet + grass likewise do not count 13 and 27 independent scenes. Repeated copies can
reproduce one textual relationship across several stored units. Establishing distinct passages, speakers, functions or material
witnesses would require the contextual and copy-specific work already specified in sec. 9. The contrast between a short window
and a whole-unit search makes that requirement visible before co-occurrence is mistaken for an interpretive association.
26

## Page 27

Figure 13: A frequent word, a concentrated source and a nearby pair answer different questions. A compares pooled and equal-work
category rates; B shows each category’s largest work-group share; C counts source units for literal singular pairs across finite windows
and whole-unit scope. Counts retain copies and overlapping categories. No panel measures independent reading positions or adjudicated
animals.
27

## Page 28

8.4.1 T welve literal-pair controls
The literal-pair audit makes source dependence visible alongside distance. Worm + fly matches four units from one group at
eight words, 28 units from three groups at 32 words, and 66 units from six groups at whole-unit scope. At 32 words its 28 units
divide into four jerusalem, eight songsie and sixteen vda units. Thus a comparatively large unit count can be supplied by
very few recorded groups. These cells identify where a contextual comparison should begin; they do not establish 28 distinct
nearby encounters.
The contrasts differ across pairs. Worm + beetle increases from four units at eight words to 31 at 32 words, with no further
increase at 64 words or whole-unit scope. Its 32-word matches comprise four milton, thirteen s-inn and fourteen songsie
units. Emmet + beetle matches only the four milton units through 64 words but reaches 31 units at whole-unit scope. A
shared source document therefore supports a much broader relation than a nearby word envelope in the latter case. Spider +
fly has no eight-word match, sixteen 32-word matches entirely in vda, and 24 whole-unit matches from three groups.
Other contrasts resist a simple relationship graph. Bee + beetle , gnat + fly , Polypus + worm and scorpion + centipede have
zero matches through 64 words but respectively four, five, eight and four whole-unit matches. Flea + louse matches five units
from two groups at every declared scope. Moth + butterfly matches one unit from one group throughout. Caterpillar + fly
returns zero even at whole-unit scope under the declared singular literal spelling, despite 36 primary candidates in the broader
caterpillar category. A zero in this experiment is consequently a result of a stated query, eligibility rule and source inventory,
not an absence claim about Blake or the scene discussed elsewhere in this article.
Appendix B prints all twelve pairs, including every zero, with matched units and contributing group counts. The full 82-by-33
matrix and twelve-pair source matrix remain vector/CSV supplements; neither is reduced to unreadable page labels here.
8.5 Zeros in a declared vocabulary
The new atlas recomputes the unchanged 102 categories and 376 explicit forms across 2,067 captured units. It returns 36,039
category-source-span candidates, including 2,498 in 199 primary reading units. All role/plane zero cells remain in the data.
The form audit records both casefolded surface-form counts and normalized-equivalent counts, separately for the primary scope
and all planes. 242 declared forms have no normalized-equivalent match anywhere in the captured inventory; 256 have no exact
casefolded surface match. A split or hyphenated form can be covered through an equivalent declared token sequence, so these
are different questions. 29 normalized alias groups expose this redundancy rather than multiplying one category’s hits.
The audit retains 36,039 distinct exact source intervals. 550 candidate pairs have intersecting intervals, including nested
compound/head matches. Equal intervals and intersecting intervals are different diagnostics; neither is a number of organisms.
No automatic taxonomic allocation is computed from these spans. The independently bound gap-probe registry tests new recall
leads while preserving the original vocabulary, baseline and query epochs. Even a zero-result gap probe would describe only
these bytes and forms.
The atlas in fig. 14 makes eligibility and repetition visible. Its first panel separates source-role/plane populations; its second
shows matched-unit fractions within every recorded primary work group. The adjacent equal-work mean gives each recorded
group one contribution, including zero-match groups. The pooled fraction weights retained units instead. This is a descriptive
comparison, not edition reconciliation or a comparison of independent works. Whole-text grouping leaves 198 primary units; it
does not remove repeated poems embedded in different copy texts. The snail’s primary zero and exploratory positives exemplify
why these populations cannot be collapsed.
Unverified Archive XML supplies 32,066 of all 36,039 candidates, or 88.98%, including 11,597 name-family, 12,485 associated,
5,993 motion and 1,991 broad-label candidates. Its 51 positive name-family categories exceed the primary reading scope’s 38, but
this additional retrieval reach includes markup and modern catalog language. It cannot enlarge the number of Blake-authored
animal attestations by simple addition. Mixed Erdman material, four mixed XML units and unverified reading documents
remain separately reported, with their own token and unit denominators.
tbl. 6 supplies the positive/declared category totals omitted from the figure: each cell counts observed categories against the
full channel vocabulary. It does not count reviewed animals. The underlying export is source-channels.csv.
Table 6: Positive/declared categories across source cohorts.
Source cohort Name families Broad labels Associated Motion
editorial/erdman 41/82 3/4 12/15 1/1
editorial/xml 42/82 3/4 12/15 1/1
primary/reading 38/82 2/4 11/15 1/1
unverified/reading 6/82 2/4 9/15 1/1
unverified/xml 51/82 3/4 13/15 1/1
28

## Page 29

Figure 14: Follow a name back to its source population. The atlas retains 2,067 captured units, 36,039 overlapping category-span
candidates; 199 primary reading units in 33 work groups. A prints raw counts with log(1+n) colour; B shows within-work matched-unit
fractions, with equal-work and pooled coverage at right. Work groups with zero matches remain. Exact whole-text grouping leaves 198
primary units. A large XML count may include apparatus or description; it cannot settle the fly’s status in a poem. No contextual
decisions transfer to this atlas; corpus images remain unreviewed.
29

## Page 30

Figure 15: Animal names and associated expressions travel through different source channels. Panel counts and within-cohort shares
distinguish 82 name families, four broad labels, fifteen associated channels and one motion channel. Every role/plane/author cohort
retains source-unit and NFKC-token denominators. These disjoint cohorts describe retrieval populations, with markup and editorial
language retained separately; they are not stages of complete-oeuvre coverage.
30

## Page 31

8.6 Readings, decisions and their separate denominators
The new casebook records 7 bounded interpretive cases and 4 complete short-poem transcriptions: “A Dream,” “The Fly,” “The
Human Abstract” and “The Sick Rose,” all in copy A. It separately records the inspected Milton and Thel passages and the
modern catalog alternative. Ordered line IDs and selected-text hashes close those reading claims. These are reading receipts,
not additional candidate decisions. The 11 historical decisions remain on their original run. 36,028 candidates were unreviewed
in that historical eleven-decision epoch. None has been reclassified or rebound to this atlas. The atlas still has zero authored
decisions. The additive baseline review described next reaches 47 decisions and leaves 35,992 candidates unreviewed. Across the
visual casebook, nine selected historical originals have been inspected. The corpus-wide image inventory remains unreviewed.
Four complete copy-Y poem transcriptions and selected Thel F/O passages belong to the additive expansion ledger; they inherit
no credit from the earlier copy-A and Thel-B readings.
This separation also sets a limit on the argument. The small lyric sequence can now be discussed through its ending, whereas
the Polypus and wine-press cases remain local passages. The contrast between practical assistance, conditional consolation,
secret destruction, nurture and bodily aggregation has support in these selected contexts. It has not been established as a
distribution of functions across all of Blake. The machine audit supplies the next places to read, and preserves the places where
it cannot yet support a claim.
8.7 Contextual judgment as a contestable record
Definition 4 (Authored judgment and unreviewed evidence). For one fixed finite candidate inventory 𝐻, an authored
adjudication is a partial function 𝑗 ∶ 𝐻 ⇀ 𝐿 , where 𝐿 contains the six dispositions literal animal, figurative animal, associated,
non-animal, editorial and uncertain. Its domain 𝑅 = dom(𝑗) contains submitted, valid decisions; 𝑈 = 𝐻 ∖𝑅 remains unreviewed.
Each decision identifies a reviewer and rationale; animal decisions also supply a verification locator. An uncertain disposition
is a recorded judgment, whereas an unreviewed candidate has no disposition. Source authorship and source role are separate
attributes and are not changed by 𝑗.
|𝐻| = |𝑅| + |𝑈 | = ∑
ℓ∈𝐿
∣{ℎ ∈ 𝑅 ∶ 𝑗(ℎ) = ℓ}∣ + |𝑈 |. (6)
The accounting identity eq. 6 makes the remaining denominator inspectable. It establishes neither that an interpretation is
correct nor that the selected decisions represent the unreviewed inventory. In particular, animal coding in an editorial or
unverified source does not make that source Blake-authored verse.
A new, declared batch of 36 candidate IDs was inspected in bounded original-source windows and coded by a research agent
on 29 October 2026. Selection preceded coding and deliberately included personified animals, concrete animal senses, products,
motion verbs and modern catalog descriptions. It was purposive rather than random. Each judgment names the reviewer,
rationale, original interval, inspected window, window hash and exact historical run pin. It therefore supports inspection and
disagreement, not a claim of specialist consensus, classifier accuracy or animal prevalence.
The actual review command validates 47 decisions after combining the new batch with the unchanged eleven historical decisions:
35,992 of the original 36,039 candidates remain unreviewed. tbl. 7 reports the combined dispositions. Literal and figurative
animal codes retain their separate meanings; the distinction concerns the selected use of a word, not a division between
scientifically true and false poetry.
Table 7: Authored dispositions in the purposive 47-decision study. Unreviewed candidates remain outside the disposition
categories.
Disposition Selected candidate IDs
animal literal 2
animal figurative 27
associated 4
non animal 6
editorial 4
uncertain 4
The new 36 IDs occupy 23 retained source units and 23 recorded work groups. The combined 47 occupy 28 source units and
25 groups. These are documentary counts, not independent poetic events. Two new same-source window pairs overlap: the
busy-bee and spider/friendship proverbs in The Marriage of Heaven and Hell (E29-06/19), and the dungeon spiders and eternal
worm in The French Revolution (E29-17/27). The Milton emmet returns to the wine-press inventory already represented by
historical Erdman seeds; the new “Auguries” fly window also contains historical spider/bee aphorisms. Distinct candidate
IDs and 36 distinct window hashes preserve these relations rather than converting repetition into independent observations.
Selection IDs and exact memberships are in the dated study ledger.
31

## Page 32

Figure 16: Purposive contextual review on the original baseline run: 47 category-source-span decisions, including the unchanged eleven
historical seeds and 36 new agent judgments. A retains every disposition cell, including zeros; B retains original source role/plane and
separates literal-plus-figurative animal coding from other dispositions. Four animal-coded candidates from the editorial source cohort
remain exploratory; catalog judgments in XML are a separate disposition and do not become primary text. These selected IDs occupy 28
source units and 25 recorded groups, with overlapping contexts and repeated poetic material. They are not 47 independent observations,
a random sample, an accuracy estimate or an animal census. Data: context-review-cells.csv and context-review-cohorts.csv.
32

## Page 33

Disambiguation changes what a retrieval can support. Four new primary-reading candidates use fly/flies as a verb: personified
Strife moves abroad, Hermes flies before Jupiter, Pestilence leaves a broken bottle, and the Genesis transcription speaks of
fowl flying. Nearby insects or buzzing do not turn the selected verb into an animal noun. Four additional matches belong to
modern XML design descriptions: birds, angels, Justice and an eagle supply the movement. These are coded editorial while
their original source role remains unverified. A semantic judgment does not promote a catalog description into Blake-authored
verse.
Positive animal coding also needs a rival reading. The bee’s lack of sorrow is an ethical proverb, not an assay of insect emotion;
the wise emmet’s dance in the wine-press belongs to a structure explicitly joined to war and printing. The selected dungeon
spiders can have a concrete animal sense within political allegory without supplying a natural-history observation. Conversely,
honey bees that sing in a transformed landscape, the Spider created to help the Daughters’ labour of pity, and the eternal worm
in a chained skeleton remain uncertain between available literal and figurative modes. Leaving those three new judgments open
is a result of contextual reading rather than a missing keyword rule.
The associated codes sharpen the article’s account of provision and capture. Honey on infant lips and the Lilly’s honey-bearing
perfume connect nourishment to vulnerable bodies without asserting a bee at the matched span. Oothoon’s silken nets, proposed
for catching girls, make a different relation possible: a connective material can enable confinement. The Web of Life in Gates
joins sexual strife, tears and kinship without identifying a spider. Thus neither material production nor a network is sufficient
evidence of care. What matters is who can act, who is sustained, and who becomes an object of another’s activity. The single
disposition labels expose semantic boundaries; they do not exhaust the passages’ literary functions.
These decisions remain attached to the original baseline run. The later atlas, its 102-category/376-form vocabulary audit,
and the 82-family composition epoch are unchanged and gain no retroactive adjudication. The new figure is a description of
authored decisions and their source channels, not a validation of all candidates. The next defensible extension is an independently
reviewed, explicitly designed batch across further works and senses, retaining the unresolved labels and the entire unreviewed
denominator.
8.8 The text occupied by a match and by a reading
The preceding counts locate candidates and judgments. They leave a different question open: how much of a retained source do
overlapping matches occupy, and what extent of contextual reading is actually documented? A new analysis uses the original
baseline hits and the unchanged 47-decision ledger. It does not rerun matching, merge copies, infer decisions from surrounding
text or alter source authorship. Its coordinates belong to the exact captured Passage.text, measured in Python Unicode
codepoints. They are not tokens, graphemes, verse lines, printed page area or a measure of interpretive attention.
Definition 5 (Source-local matched support). For the candidates 𝑇𝑠 in one retained source 𝑠, let 𝐼ℎ = [𝑎 ℎ, 𝑏ℎ) be candidate
ℎ’s original interval and let 𝐿𝑠 = |𝑥 𝑠| > 0 . Define the union of matched support 𝐺𝑠(𝑇𝑠), its length 𝑔𝑠(𝑇𝑠) and summed length
𝐵𝑠(𝑇𝑠) by the following expressions. Here [𝑎, 𝑏) denotes the finite set of integer codepoint positions 𝑎, … , 𝑏 − 1, so cardinality
is text length.
𝐺𝑠(𝑇𝑠) = ⋃
ℎ∈𝑇𝑠
𝐼ℎ, 𝑔 𝑠(𝑇𝑠) = |𝐺 𝑠(𝑇𝑠)|, 𝐵 𝑠(𝑇𝑠) = ∑
ℎ∈𝑇𝑠
(𝑏ℎ − 𝑎ℎ). (7)
All unions are taken within this source frame. A copy with identical text is still another source; its coordinates are never overlaid
on the first copy. The overlap area is the number of codepoints covered by at least two records. Excess length is 𝐵𝑠 − 𝑔𝑠. The
number of overlapping record pairs counts pairs with a positive-length intersection, rather than occupied positions.
These quantities answer different questions. Three identical intervals of length two have union length two, overlap area two,
excess length four and three overlapping pairs. The example is a mathematical control, not a corpus result. It shows why
subtracting the union from summed lengths cannot in general replace a calculation of overlap area. In every admitted source,
0 ≤ overlap area ≤ 𝑔 𝑠 ≤ 𝐵 𝑠 and 𝑔𝑠 ≤ 𝐿 𝑠. The reusable interval operations belong to public Blake; this application owns the
source grouping, eligible author/role/plane, channel meanings and literary consequences of the measurement.
Across the 2,067 retained units, the 36,039 candidates have summed length 193,634 codepoints and source-local union length
191,443. The overlap area is 2,191 codepoints; 550 record pairs overlap, agreeing with the retained atlas’s pair count. Here
excess length also equals 2,191, an equality specific to these inputs rather than a definition. The five nonempty source facets
remain separate in fig. 17. A large XML denominator includes markup, apparatus and modern descriptions, whereas the reading
projection has its own line-break and whitespace choices. The facet percentages consequently describe captured representations,
rather than comparable densities of animal reference in Blake’s verse.
The identified primary reading scope contains 2,498 candidates, 4,661,524 codepoints and 12,310 occupied codepoints. Its name-
family union is 6,443; associated terms occupy 5,177, motion forms 666 and broad labels 110. Name-family and associated
support intersect in 86 codepoints. The channel bars are therefore deliberately unstacked. They cannot be added as though
they partition the text, and none turns a motion verb or material association into an adjudicated animal.
33

## Page 34

Weighting also changes this measurement. Let 𝑆 contain all 199 eligible source units, including zero-hit units, and let 𝑆𝑤 be
the units in recorded work group 𝑤, for all 33 eligible groups 𝑊 . Write 𝑔𝑠 = 𝑔 𝑠(𝐻𝑠). Then
𝑝pool =
∑𝑠∈𝑆 𝑔𝑠
∑𝑠∈𝑆 𝐿𝑠
, 𝑝 unit = 1
|𝑆| ∑
𝑠∈𝑆
𝑔𝑠
𝐿𝑠
, 𝑝 work = 1
|𝑊 | ∑
𝑤∈𝑊
1
|𝑆𝑤| ∑
𝑠∈𝑆𝑤
𝑔𝑠
𝐿𝑠
. (8)
Table 8: Alternative weights on the same original-codepoint support. Zero-hit units and groups remain in scope, and copies
remain separate. These are representation-dependent occupancies, not animal prevalence or corrected independent editions.
Weighting of occupied primary text Occupancy Denominator retained
Pooled codepoints 0.264077% All primary codepoints
Equal source units 0.330366% All 199 primary units
Equal recorded work groups 0.347154% All 33 groups; units equally
weighted within each
Equal weighting changes which parts of the captured collection have more influence; it does not repair incomplete acquisition,
establish independent copies or remove polysemy. It also differs from the earlier name-family composition, which conditions on
retrieved category mass. Here a source with no hit contributes zero occupied support and still contributes its denominator.
Definition 6 (Documented contextual-window support). Let 𝑅𝑠 be the reviewed IDs in source 𝑠 and let 𝐷𝑠 contain the
retained window records with verified original endpoints. If record 𝑘 covers 𝐽𝑘 = [𝑢 𝑘, 𝑣𝑘), its documented window support is
𝑉𝑠 = ⋃
𝑘∈𝐷𝑠
𝐽𝑘, ∑
𝑠
|𝑉𝑠| ≤ ∑
𝑠
∑
𝑘∈𝐷𝑠
(𝑣𝑘 − 𝑢𝑘). (9)
Each record must reproduce its exact source slice, contain its selected hit and join to an existing valid decision on the original
run. A reviewed ID without equivalent window coordinates remains explicitly unavailable for this measurement. Its absence is
neither zero reading nor an inferred window.
The fixed 36-record witness inventory from the 29 October extension has summed window length 61,922 codepoints and source-
local union length 60,083 across 23 units. The full decision ledger reaches 28 source units; the eleven seed IDs lack equivalent
machine-readable window custody. fig. 18 shows every source unit containing a decision, including those unavailable windows.
The two enlarged pairs explain the difference between records and extent. In the selected French Revolution source, E29-17
and E29-27 occupy 2,417 codepoints jointly, rather than their summed 3,611. In The Marriage of Heaven and Hell , E29-06
and E29-19 jointly occupy 2,964 rather than 3,609. Their overlap does not make the bee and spider the same candidate or an
independent pair of literary scenes. Reinspection of one retained window adds no new area to this fixed inventory; separately
bounded poem transcriptions keep their own reading records rather than receiving invented source coordinates.
Proposition 3 (An ID partition need not partition text). For the disjoint reviewed and unreviewed ID sets 𝑅𝑠 and 𝑈𝑠 of
Definition 4, the matched support satisfies 𝐺𝑠(𝐻𝑠) = 𝐺 𝑠(𝑅𝑠)∪𝐺𝑠(𝑈𝑠). Its length is 𝑔𝑠(𝐻𝑠) = 𝑔 𝑠(𝑅𝑠)+𝑔𝑠(𝑈𝑠)−|𝐺𝑠(𝑅𝑠)∩𝐺𝑠(𝑈𝑠)|.
Disjoint IDs therefore need not imply disjoint occupied text.
Justification. Each hit belongs to exactly one ID set, so their intervals give the stated union. Finite-set inclusion–exclusion
gives its length. Different IDs can nevertheless have intersecting original intervals.
In these inputs, reviewed matches occupy 220 codepoints and unreviewed matches 191,240. Their supports overlap in 17 code-
points. Neither reading a larger window nor finding an unreviewed candidate inside it assigns that candidate a disposition. The
fixed review ledger, the intervals actually exposed to a reader and the literary argument therefore retain different denominators.
The resulting picture makes selection inspectable without turning it into an attention score. A narrow mark identifies where
a word was found; its window retains more context but can still stop before a poem’s resolution, a speaker transition or a
neighbouring design. In the overlapping French Revolution windows, the prison and its animal occupants already complicate a
single symbolic classification. Geometry shows where those readings share material; the narrative and its rival interpretations
determine what that material can support. The 35,992 unreviewed candidates remain available for further reading.
9 Attention, symbolic inheritance and unfinished readings
The software establishes complete matching of the declared forms within the pinned units. A comprehensive literary account still
requires candidate adjudication, work/copy reconciliation, examination of source boundaries, assessment of unlisted or unnamed
animals, and a systematic visual inventory. Long-range discussion can exceed the saved 120-character context envelope; review
must return to the complete source unit and, where necessary, its neighbours and physical object. Animal-coded candidates
from editorial source cohorts remain exploratory until the quoted text’s author and location have been established separately.
34

## Page 35

Figure 17: 36,039 category-source-span candidates across 2067 retained units. Primary: 199 units, 4,661,524 original codepoints.
Union length counts each codepoint once per source; overlap area and excess summed length have different multiplicity meanings.
Channel unions are unstacked. Declared positional examples are illustrative, not representative. Raw XML includes apparatus and
markup; source occupancy is not an animal census.
35

## Page 36

Figure 18: 47 authored decisions; 35,992 candidates remain unreviewed. 36 retained window events; 11 reviewed IDs lack equivalent
machine-readable window custody. All reviewed source units are shown. Ticks make selected starts visible; outlines preserve actual
window extents. Same-source enlargements retain exact original offsets. Repeated reading adds no independent scenes; no reading credit
is inferred from generated snippets.
36

## Page 37

Mike Goode’s discussion of digital fragmentation is pertinent to the research design: retrieving a textual fragment can detach
it from the verbal-visual work. The present concordance is therefore a navigation instrument. An argument requires returning
from its fragments to their material and textual relations. Only selected opening sections of Goode’s essay were read here.
(Goode, 2016)
The nine selected originals provide differential material witnesses, not a representative visual census. The unresolved corpus
image identifiers, other copies and unnamed forms still require systematic inventory. The additional image readings create no
textual decisions. The separate 36-case baseline batch supplies explicit, contestable decisions while preserving 35,992 unreviewed
candidates.
The symbolic rival remains active at the end of the inquiry. The maggot is explicitly emblematic, the Fly supplies a human
analogy and the Polypus names a mythic transformation. Their literary force does not depend on escaping those uses. The
relevant difference is whether the reading retains the particular action and its consequence: help is offered rather than accom-
plished, consolation can return to the speaker, and connected lives can acquire powers that alter their members. Returning
those differences to view is a contribution within the history of biological symbolism, rather than a claim to have replaced it.
The article’s provisional claim is consequently bounded but substantive: invertebrate figures let these selected Blake passages
test the relation between attention to a life and the forms that connect, classify or absorb it. The counterexamples are part
of that claim. Sympathy, mockery, appetite, predation, work and aggregation cannot be replaced by one stable animal-symbol
dictionary. The corpus application makes the further tests of this reading inspectable.
Appendix A: Complete name-family coverage
The table covers all 82 name-family categories in the 199-unit primary reading scope (33 recorded work groups). Counts are
category-source-span candidates; matched units and positive groups retain copies. Half-mass groups is the minimum number
of descending-count groups accounting for at least half of a category’s mass; a dash marks undefined concentration for a zero
category. Associated channels and broad labels are separate from this name-family table.
Table 9: All 82 name-family categories in the fixed primary scope, including zeros and undefined zero-mass concentration.
Category Candidates Matched units Positive groups Half-mass groups
worm 468 113 15 2
earthworm 22 22 3 1
tapeworm 4 4 1 1
silkworm 4 4 1 1
glowworm 27 27 2 1
bookworm 0 0 0 —
blindworm 16 16 1 1
cutworm 9 9 1 1
ant 37 37 5 2
fly 183 77 15 1
housefly 0 0 0 —
dragonfly 0 0 0 —
gadfly 0 0 0 —
mayfly 1 1 1 1
firefly 0 0 0 —
horsefly 0 0 0 —
bee 44 35 5 2
honeybee 14 14 3 1
bumblebee 0 0 0 —
wasp 4 4 1 1
hornet 4 4 1 1
beetle 62 31 3 2
chafer 2 2 2 1
cockchafer 1 1 1 1
scarab 0 0 0 —
weevil 0 0 0 —
ladybird 0 0 0 —
moth 18 10 3 1
butterfly 20 20 4 1
caterpillar 36 36 6 2
maggot 4 4 1 1
37

## Page 38

Category Candidates Matched units Positive groups Half-mass groups
grub 0 0 0 —
larva 0 0 0 —
chrysalis 0 0 0 —
flea 5 5 2 1
louse 5 5 2 1
gnat 5 5 2 1
midge 0 0 0 —
grasshopper 34 33 4 2
cricket 1 1 1 1
locust 0 0 0 —
earwig 4 4 1 1
bug 4 4 1 1
cockroach 0 0 0 —
roach 0 0 0 —
termite 0 0 0 —
cicada 0 0 0 —
mosquito 0 0 0 —
spider 78 43 8 2
spinner 4 4 1 1
scorpion 5 5 2 1
mite 1 1 1 1
tick 0 0 0 —
centipede 4 4 1 1
millipede 0 0 0 —
snail 0 0 0 —
slug 4 4 1 1
oyster 0 0 0 —
mussel 13 13 2 1
cockle 0 0 0 —
clam 1 1 1 1
whelk 0 0 0 —
limpet 0 0 0 —
nautilus 0 0 0 —
octopus 0 0 0 —
cuttlefish 0 0 0 —
squid 0 0 0 —
leech 0 0 0 —
crab 0 0 0 —
lobster 0 0 0 —
shrimp 0 0 0 —
prawn 0 0 0 —
crayfish 0 0 0 —
barnacle 0 0 0 —
woodlouse 0 0 0 —
polypus 73 9 3 1
hydra 0 0 0 —
coral 0 0 0 —
sponge 18 16 1 1
zoophyte 0 0 0 —
animalcule 0 0 0 —
medusa 0 0 0 —
Appendix B: Twelve literal-pair controls
Cells give matched units followed by contributing groups in parentheses. Eight-, 32- and 64-word complete envelopes differ
from whole-unit co-presence. All cells use 199 eligible units and 644,245 original-text word tokens; singular literal spellings do
not inherit expanded category forms.
38

## Page 39

Table 10: All twelve literal-pair controls. Each cell gives matched units and contributing recorded groups; copies remain
retained.
Literal pair 8 words 32 words 64 words Whole unit
worm+fly 4 (1) 28 (3) 28 (3) 66 (6)
worm+beetle 4 (1) 31 (3) 31 (3) 31 (3)
emmet+beetle 4 (1) 4 (1) 4 (1) 31 (3)
spider+fly 0 (0) 16 (1) 16 (1) 24 (3)
spider+beetle 4 (1) 4 (1) 4 (1) 4 (1)
bee+beetle 0 (0) 0 (0) 0 (0) 4 (1)
gnat+fly 0 (0) 0 (0) 0 (0) 5 (2)
caterpillar+fly 0 (0) 0 (0) 0 (0) 0 (0)
polypus+worm 0 (0) 0 (0) 0 (0) 8 (2)
moth+butterfly 1 (1) 1 (1) 1 (1) 1 (1)
flea+louse 5 (2) 5 (2) 5 (2) 5 (2)
scorpion+centipede 0 (0) 0 (0) 0 (0) 4 (1)
Appendix C: Material-witness concordance
Table 11: Nine selected material objects and their separate numbering systems. Unreconciled identifiers and conflicting printing
dates remain explicit.
Witness Bentley / Erdman Archive object Repository identity
Printing date
evidence
A Dream, Songs Y 26 / E16 26 Met 17.10.26 [1794], ca. 1825
The Fly, Songs Y 40 / E23 40 Met 17.10.40 [1794], ca. 1825
The Sick Rose, Songs
Y
39 / E23 39 Met 17.10.39 [1794], ca. 1825
The Human
Abstract, Songs Y
47 / E27 47 Met 17.10.47 [1794], ca. 1825
Thel F, III 6 / plate 4 6 LOC 1798A Archive 1795; LOC
ca. 1794
Thel O, III 6 / plate 4 5 LOC digital 1798B;
bound volume 1810
Archive c. 1818; LOC
ca. 1815
Thel O, Clay
exchange
7 / plate 5 6 LOC digital 1798B;
bound volume 1810
Archive c. 1818; LOC
ca. 1815
What is Man!, Gates
Children E
Museum plate 1;
later E260 text
separate
Not reconciled YCBA
B1978.43.1484
Catalogued 1793;
inscription 17 May
1793
Alas!, Gates Children
E
Museum plate 9;
printed 7;
handwritten 8; later
E263 passage 7
separate
Not reconciled YCBA
B1978.43.1492
Catalogued 1793
Appendix D: Sources and reading boundaries
The source ledger distinguishes complete short-poem readings from local passages and from whole works. The four complete
copy-A Songs transcriptions and selected Milton A / Thel B windows remain historical credit. This expansion adds four
complete copy-Y poem transcriptions and seven selected scans, plus local Thel F/O, Island and “Auguries” passages. It does
not credit a complete book, all copies or a comprehensive visual inventory.
Perkins (1999) and Srigley (1992) were read as complete journal articles in the retained agent research pass. The earlier Perkins
fetch failure remains in the frozen ledger as acquisition history. Essick and Viscomi (2002) are used only for selected opening
and registration sections (pp. 74–76, 90–92); Dauphin’s extended reading covers paragraphs 7–28 and 49–50 alongside the earlier
3–6 and 21–25. Fara is cited under the 2020 online publication date; its issue and PDF citation carry 2016 (Spring–Fall, nos.
66–67). Only paragraphs 15–23 were read. ( Perkins, 1999; Srigley, 1992; Essick and Viscomi , 2002; Dauphin, 2026; Fara, 2020)
Gould (1747) is limited to the title and printed pp. 66–68, 78–81; Baker (1743, second edition) to complete selected paragraphs
on pp. 53–58; Zoonomia to the 1796 second edition, I.XXXIX.II.3 and IV.2; and The Temple of Nature (1803) to III.411–434
39

## Page 40

and IV.1–66. These are historical comparisons, with neither whole-book credit nor a proved influence chain. Schwartz remains
publisher-abstract only; Goode remains selected opening sections. Reading a scholar’s article grants no automatic credit for its
cited books, manuscripts or images. ( Gould, 1747; Baker, 1743; Darwin, 1796, 1803; Schwartz, 2012; Goode, 2016)
The additive reading, image-provenance and claim ledgers preserve source URLs, selected paragraphs or ordered physical line
IDs, source hashes and rival readings. Original failed accesses and earlier decisions remain intact. New source captures and
binaries stay in private local custody.
The present edition adds two selected copy-E Gates objects, bounded ASU E260/E263/E268 passages, and Höltgen’s selected
Gates discussion; none receives complete-book or complete-article credit. The 36 new candidate windows carry separate agent-
review credit and do not extend these complete-poem claims. ( Höltgen, 2002; Blake, n.d.c)
The label n.d. marks captured online resources whose publication year was not verified. It does not erase a work’s imprint
or a material witness’s printing-date evidence; those remain separately specified. Bibliographic copy letters and proper names
retain their source capitalization.
The 30 October research pass verifies Baine’s 1986 book identity and its credit to Mary R. Baine for assistance, but obtains no
readable invertebrate chapter, introduction or conclusion. The insect chapter description and its reported book-page references
come from Hoagwood’s complete review, read at journal anchors 39–42. They are not direct book locators. Lincoln’s publisher
preview was read only on p. 117; p. 118 and Rothenberg’s review remain unread and supply no substantive claims here. The
title-page credit typography was not visually verified. ( Baine, 1986; Hoagwood, 1989)
Hutchings is limited to the title and opening and selected complete paragraphs 14–22, digital collected-PDF pages 71–76.
The corresponding Visions copy-A object 5 transcription, physical lines 2–26, was inspected separately, including its speaker
transition. It grants neither original-image nor complete-copy credit. Fresh reinspections of selected Songs Y and Milton A
transcription contexts corroborate their inherited local reading limits. No new image, contextual decision or historical influence
chain follows from these readings. The dated source/reading/access ledger preserves the captures, source hashes and unavailable
routes. ( Hutchings, 2001; Blake, 1793a)
The 1 November source-geometry pass rereads eight retained contextual windows, two complete copy-A poems and four bounded
Milton A / Thel B selections. Three further window reinspections check the enlarged bee/spider and worm/spider pairs. These
are additive agent reading events, not Daniel’s personal-reading credit. They change no contextual decision and inspect no new
original image. The analysis uses the fixed 36-record window inventory; the additional reading ledger is not silently joined to
its geometric denominator. Baine’s chapter remains unavailable and unread.
References
The following references identify the cited scholarship, historical witnesses and captured classification resources. The source
ledgers retain exact reading ceilings and snapshot hashes.
Rodney M. Baine. The Scattered Portions: William Blake’s Biological Symbolism . Distributed by the author, Athens, Georgia,
1986. ISBN 0935265104. URL https://library.vicu.utoronto.ca/special/bentley/catalogue/1516-1697.html . With the
assistance of Mary R. Baine. xx + 260 pages. ISBN-13 9780935265101.
Henry Baker. The Microscope Made Easy . R. Dodsley, London, 2 edition, 1743. URL https://wellcomecollection.org/works
/u4cdmgs8. ESTC T89680; Wellcome Collection EPB/B/11985. Complete selected paragraphs at printed pp. 53–58 visually
read from scans 0090–0095 in manifest b20642659. No whole book reading or demonstrated Blake source claim.
William Blake. Visions of the Daughters of Albion, copy A. William Blake Archive transcription, 1793a. URL https://raw.gi
thubusercontent.com/blakearchive/data/7063e9ea8b13d18bd5a585e87610164fd75d3f60/works/vda.a.xml . Pinned Archive
commit 7063e9ea8b13d18bd5a585e87610164fd75d3f60. Object 5, Bentley 6, Erdman 3, Keynes 3; physical text lines 2–26
inspected. No original-image or complete-copy reading credit. Local captured bytes reinspected 30 October 2026.
William Blake. Alas! (Plate 9). Yale Center for British Art collection record, 1793b. URL https://collections.britishart.yale.
edu/catalog/tms:2421 . For Children: The Gates of Paradise, copy E. Paul Mellon Collection, B1978.43.1492. Accessed 29
October 2026.
William Blake. What is Man!, Frontispiece (Plate 1). Yale Center for British Art collection record, 1793c. URL https:
//collections.britishart.yale.edu/catalog/tms:2412 . For Children: The Gates of Paradise, copy E. Paul Mellon Collection,
B1978.43.1484. Accessed 29 October 2026.
William Blake. The Book of Thel, copy F, object 6 (Bentley plate 6; Erdman plate 4). Library of Congress, Lessing J. Rosenwald
Collection, 1798A, 1795. URL https://www.loc.gov/resource/rbc.2003rosen1798A/?sp=6. Work imprint 1789. Archive dates
printing 1795; LOC catalogue describes ca. 1794 with 1794 watermark. Exact source disagreement retained. Whole captured
scan visually read; selected physical text only.
William Blake. The Book of Thel, copy O, objects 5–6 (Bentley plates 6–7). Library of Congress, Lessing J. Rosenwald
Collection, digital resource Rosenwald 1798B, 1818. URL https://www.loc.gov/resource/rbc.2003rosen1798B/?sp=5 . Work
40

## Page 41

imprint 1789. Archive dates printing c. 1818; LOC catalogue says ca. 1815 with 1815 watermarks. Bound with Milton copy D;
Archive repository catalog number 1810 differs from LOC digital subresource 1798B. Do not conflate these identifiers. Two
whole captured scans visually read; selected physical text only.
William Blake. Songs of Innocence and of Experience, copy Y: Selected Impressions and Transcriptions. The Metropolitan
Museum of Art; William Blake Archive transcription, 1825. URL https://www.metmuseum.org/art/collection/search/347935.
Four Met originals: 17.10.26, 17.10.39, 17.10.40, 17.10.47. Museum dating: [1794] printed ca. 1825. Images downloaded from
museum originals, not Blake Archive image servers. Complete corresponding short poems read in pinned Archive XML
7063e9ea8b13d18bd5a585e87610164fd75d3f60, songsie.y, objects 26, 39, 40, 47.
William Blake. The Complete Poetry and Prose: Selected Retained Reader Units. William Blake Archive reader API, retained
private corpus bundle, n.d.a. URL https://erdman.blakearchive.org/api/pages . Retained upstream response SHA-256
61c846e4b3122eb9eca6797805bf80d976a6712c179183504b1a465c2ef6f370. No live API refresh. E462 record 474 chapter 10
and Auguries E490–491 records 502–503 selected lines read. Source pages contain editorial material; article and line identities
remain separate.
William Blake. The French Revolution: Selected Retained Reading Transcription. William Blake Archive, pinned XML bb49.a,
n.d.b. URL https://raw.githubusercontent.com/blakearchive/data/7063e9ea8b13d18bd5a585e87610164fd75d3f60/works/b
b49.a.xml. Original reading-layer windows [3178,4982) and [3788,5595), witness records E29-27 and E29-17. Source SHA-256
3284bd7714e63e3f4b3ec496d80abf5e95c6c25515b238e0d1cde611e514a867. Bounded transcription reading, not complete-work
or image credit.
William Blake. For the Sexes: The Gates of Paradise. The Complete Poetry and Prose of William Blake, ASU Libraries
transcription, n.d.c. URL https://blake.lib.asu.edu/html/for_the_sexes.html . Selected frontispiece E260, emblem 7 E263,
and opening of the Keys E268. Accessed 29 October 2026.
William Blake. Milton, copy A, n.d.d. URL https://raw.githubusercontent.com/blakearchive/data/7063e9ea8b13d18bd5a585
e87610164fd75d3f60/works/milton.a.xml. Selected physical transcription lines in objects 24 and 28; modern object31 catalog
description separately read. No full-book or image-reading credit.
William Blake. Songs of Innocence and of Experience, copy A, n.d.e. URL https://raw.githubusercontent.com/blakearchive
/data/7063e9ea8b13d18bd5a585e87610164fd75d3f60/works/songsie.a.xml . Pinned XML; physical transcriptions of objects
27, 33, 41 and 48 read completely. No full-copy or image-reading credit.
William Blake. The Book of Thel, copy B, n.d.f. URL https://raw.githubusercontent.com/blakearchive/data/7063e9ea8b
13d18bd5a585e87610164fd75d3f60/works/thel.b.xml . Selected physical transcription lines in objects 6 and 7. No full-book,
other-copy or image-reading credit.
Erasmus Darwin. Zoonomia; or, the Laws of Organic Life , volume 1. J. Johnson, London, 2, corrected edition, 1796. URL
https://www.gutenberg.org/cache/epub/15707/pg15707-images.html . Accessed transcription identifies the second edition,
1796; dedication dated 1 May 1794. Selected section XXXIX.II.3 and XXXIX.IV.2 read. Not silently cited as the 1794 first
edition; no complete book reading.
Erasmus Darwin. The Temple of Nature; or, the Origin of Society: A Poem, with Philosophical Notes . J. Johnson, London,
1803. URL https://www.gutenberg.org/cache/epub/26861/pg26861-images.html . Selected canto III, lines 411–434, and
canto IV, lines 1–66 read in the captured transcription; no complete poem or notes reading.
Caroline Dauphin. A Genealogy of the Polypus, from William Blake to H.P. Lovecraft: The Root of All Evil? Lectures du monde
anglophone, (7), 2026. URL https://publis-shs.univ-rouen.fr/eriac/1132.html . Issue: Romantic Offshoots: Reassessing the
Legacy of British Romanticism. Extended bounded reading, paragraphs 7–28 with intervening quotations, plus 49–50. Earlier
paragraphs 3–6 and 21–25 remain historical credit in the frozen candidate; no full article or cited-book credit.
Robert N. Essick and Joseph Viscomi. An Inquiry into William Blake’s Method of Color Printing. Blake/An Illustrated
Quarterly, 35(3):74–103, 2002. URL https://bq.blakearchive.org/35.3.essick . Winter 2002. Bounded opening, Background
and Context paragraphs, and registration discussion at pp. 91–92 read. No complete article or embedded-image reading
credit.
Patricia Fara. Questions of inheritance: Erasmus and Charles Darwin. Romanticism and Victorianism on the Net , (66–67),
2020. doi: 10.7202/1069958ar. URL https://www.erudit.org/en/journals/ravon/2016-n66-67-ravon05340/1069958ar/ . Issue
labelled Spring–Fall 2016; Crossref records online publication 10 June 2020. PDF citation recommends 2016. Preserve both
dates explicitly. Selected paragraphs 15–23 read; no complete article reading credit.
Mike Goode. Google Blake. Romantic Circles Praxis, 2016. URL https://romantic-circles.org/praxis/pedagogical_blake/p
raxis.2016.pedagogical_blake.goode . Selected opening sections on fragmentation and pedagogy; no complete essay reading.
Metadata from the preserved source-intake ledger.
William Gould. An Account of English Ants . A. Millar, London, 1747. URL https://darwin-online.org.uk/converted/pdf/
1747_Gould_ants_CUL-DAR.LIB.233.pdf . Digitized Cambridge University Library copy, CUL-DAR.LIB.233, hosted by
41

## Page 42

Darwin Online. Title page and selected printed pp. 66–68, 78–81 read; OCR retained with original PDF. No complete book
reading or Blake influence claim.
Terence Allan Hoagwood. Rodney M. Baine, with the assistance of Mary R. Baine, The Scattered Portions: William Blake’s
Biological Symbolism. Blake/An Illustrated Quarterly , 23(1):39–42, 1989. URL https://bq.blakearchive.org/23.1.hoagwood.
Summer. Book review; complete review text read, with journal page anchors. Reviewer accounts do not confer book-reading
credit. Accessed 30 October 2026.
Karl Josef Höltgen. William Blake and the Emblem Tradition. Erfurt Electronic Studies in English , (2), 2002. URL https:
//webdoc.gwdg.de/edoc/ia/eese/artic22/hoeltgen/2_2002.html. Selected discussion of The Gates of Paradise and concluding
comparison with conventional emblems; no complete-article reading claim. Accessed 29 October 2026.
Kevin Hutchings. Gender, Environment, and Imperialism in William Blake’s Visions of the Daughters of Albion. In James C.
McKusick, editor, Romanticism and Ecology , Romantic Circles Praxis Series. Romantic Circles, 2001. URL https://ro
mantic- circles.org/sites/default/files/2024- 08/RC_Praxis_2001_Romanticism%20and%20Ecology.pdf . Selected
complete paragraphs 14–22, digital collected-PDF pages 71–76; title and opening paragraphs also inspected. No whole-essay
or monograph reading claim. Accessed 30 October 2026.
National Center for Biotechnology Information. NCBI Taxonomy: selected TaxId, rank and LineageEx records, n.d. URL
https://www.ncbi.nlm.nih.gov/taxonomy . Two retained EFetch XML captures; exact record URLs, IDs, hashes and limits
in research/reference-taxonomy-v1.json. Classification reference, not an inferred sequence tree.
National Library of Medicine. On what sequences do you base NCBI Taxonomy classification and the Taxonomy Common
Tree?, n.d. URL https://support.nlm.nih.gov/kbArticle/?pn=KA-03471 . Official knowledge article KA-03471; captured
HTML retained privately.
David Perkins. Animal Rights and “Auguries of Innocence” . Blake/An Illustrated Quarterly , 33(1):4–11, 1999. URL https:
//bq.blakearchive.org/33.1.perkins. Summer 1999. Complete article text read; cited sources not thereby read. Journal HTML
and PDF captured privately.
Janelle A. Schwartz. Art Thou but a Worm? Blake and the Question Concerning Taxonomy. In Worm Work: Recasting
Romanticism, pages 113–148. University of Minnesota Press, 2012. doi: 10.5749/minnesota/9780816673209.003.0004. URL
https://academic.oup.com/minnesota-scholarship-online/book/17469/chapter-abstract/175003012 . Publisher abstract and
bibliography only; full chapter not read. Metadata from the preserved source-intake ledger.
Michael Srigley. The Sickness of Blake’s Rose. Blake/An Illustrated Quarterly , 26(1):4–8, 1992. URL https://bq.blakearchive.
org/26.1.srigley. Summer 1992. Complete article text read. Proposed Paracelsian source relation retained as interpretation,
not independently proven influence.
42