The books Edward Hopper drew his paintings into, and Josephine Hopper priced
A reading and transcription workbench for the six artist's ledgers at the
Whitney Museum of American Art — 504 digitised sheets, 1907 to 1967.
→ hopper.commutator.io
The point
After Hopper finished a work he meant to sell, he drew it small at the head of
a leaf — bold strokes and close cross-hatching, made for the record and not for
anyone to see. The books were ordinary ledgers from Woolworth's; Book I still
carries the label, Herald Square Account Book No. 2041, made expressly for
F. W. Woolworth Co.
Josephine Nivison Hopper wrote everything else. The inside front cover of
Book I says so, in her hand:
Recorded by Jo N Hopper (Mrs. Edward Hopper) at time each work is finished,
or before it leaves studio. Drawings in the 3 books done by Edward Hopper.
She logged the date, the size, the price, the buyer, every exhibition the work
went to and whether the jury took it, the dealer's commission, and the date the
cheque cleared — and, for the paintings, long particular descriptions that
invent anecdotes for pictures her husband would not discuss. Leaf 2 of Book I
carries thirty-one ruled lines under one etching, running from a Los Angeles
print jury in January 1921 to a Kennedy Galleries sale in June 1956.
The Whitney has digitised all six volumes and published them, sheet by sheet,
with a full catalogue record for each. What it cannot do is put the reading
beside the sheet. This site does one thing: it shows the transcription of a
sheet next to the Whitney's own photograph of it, so that scrolling the
transcript turns the photograph and any reading can be checked against the hand
it came out of, on one screen, without a second window.
The site
The archive's organisation is the site's organisation. The Hoppers filed their
work in six books and numbered them; the Whitney accessioned those six as
96.208 to 96.213 and digitised them in order. There is no thematic
regrouping anywhere here — an earlier version of this project offered six
reading "books", three of them threads drawn across the volumes by us, each
having to announce at its head whose grouping it was. That was honest and it
was still a second organisation laid over one that already had an author, and
it meant every URL named something the museum could not be asked about.
| Page | Contents |
|---|
/ | What these books are |
/book-i/ | Book I · 96.208 · 1913–1963 — the etchings plate by plate, and the running lists behind them |
/book-ii/ | Book II · 96.209 · 1907–1962 — the paintings, one to an opening, with the record sketches |
/book-iii/ | Book III · 96.210 · 1924–1967 — the late work, and the lists kept in front of it |
/book-iv/ | Book IV · 96.211 · 1913–1967 — every payment received, without a gap, for fifty-four years |
/book-v/ | Book V · 96.212 · 1953–1963 — an index, loans, drawings, and the Rehn receipts of 1953 |
/dealers/ | Dealers/Etchings · 96.213 · 1921–1951 — indexed by dealer rather than by work |
/timeline/ | Hopper's life beside the dated leaves — every entry sourced |
/technique/ | What the pictures were made of, and what shape — the paint formula in her hand, the sizes in his |
|
Every volume page links to the Whitney's own catalogue record for the object,
so a citation from here is a citation the museum can answer — see
How to cite for the sentence to copy, and for which of the
three overlapping numbers to quote.
Book IV is the one to open if you only open one. It is a pocket book, 7½ by
4¾ inches, with no sketches and no anecdote: one column of dates, one of payers
and one of sums, kept without a gap from 15 November 1913 to 23 March 1967. Its
first decade is all illustration work — Adventure, Everybody's, the Wells
Fargo Messenger, Morse Dry Dock — which is the part of Hopper's working life
the paintings books do not record at all. It is also the only volume the
Whitney dated leaf by leaf, which is what lets the timeline point into it.
How to cite
Three numbering systems overlap in these books and only one of them is an
address. The Hoppers wrote leaf numbers on the paper. The Whitney's
digitisation counts photographs — the covers, the versos, the flyleaves and
the loose insertions included — so it agrees with the leaf numbers nowhere. And
ResourceSpace holds a resource ref, which is in no order at all and is the only
thing in the archive that names exactly one image.
"Leaf 58" names six different things across six volumes. Inside Book I alone it
names three photographs — refs 18297, 17062 and 17411. Ref 18297 names one.
So a citation from here prints both: the leaf, because it is what the book
itself says and the only number a reader of the photograph can see, and the
ref, because it is the one the museum can be asked about.
The corpus itself
Archived on Zenodo, one version per release. Cite all versions with
doi:10.5281/zenodo.22770550, or the
version DOI shown on the record when quoting a specific reading.
The form
Josephine Nivison Hopper and Edward Hopper, Artist's ledger — Book I,
leaf 58 (ref 18297). Whitney Museum of American Art, 96.208.
Hopper Ledgers, batch-06, first machine pass by Opus 5, 6 September 2026,
https://hopper.commutator.io/book-i/#book-i/6/18297 (accessed 10 September
2026).
Two clauses, and the division is the whole design. The object — authors,
volume, leaf, ref, holding institution, accession — is the Whitney's, and it
stays true whatever happens to this site. The reading — this site, the
batch, the pass — is a claim of ours, it is dated, and it may be wrong.
Nothing in it is invented for the occasion. Every field is one the TEI export
already emits, so a citation copied off the screen and a TEI file deposited in
a repository cannot disagree:
| In the citation | In the TEI |
|---|
| Josephine Nivison Hopper and Edward Hopper | <author> × 2, in that order |
| Artist's ledger — Book I | <title> |
| batch-06 | the file, batch-06.tex; <title> spells it "batch 6" |
| leaf 58 | <pb n="58"/> |
| unnumbered leaf | <pb n="unnumbered"/> |
| ref 18297 | <pb facs="…view.php?ref=18297" xml:id="sheet-18297"/> |
| Whitney Museum of American Art, 96.208 | <msIdentifier> |
| first machine pass by Opus 5, 6 September 2026 | <respStmt> |
Three things get cited, and they are not interchangeable
A sheet — one photograph — is the default, and the only unit with a unique
address:
… Book I, leaf 58 (ref 18297). … https://hopper.commutator.io/book-i/#book-i/6/18297
A leaf is the page of the book as the Hoppers numbered it, and is what
somebody holding the volume can find. A leaf may be several photographs, so a
leaf citation prints every ref, in binding order, rather than quietly picking
one; its link is the leaf- form, which is resolved on arrival and keeps
working if the batching ever changes:
… Book I, leaf 58 (refs 18297, 17062, 17411). … https://hopper.commutator.io/book-i/#book-i/leaf-58
A batch — twelve sheets — is what to cite when the thing being cited is
the transcription: a reading, a note, an editorial decision. It names the
sheets and the leaves it covers, because a batch number is ours and means
nothing to anybody holding the book:
… Book I, batch-06 (sheets 61–72, leaves 56–63). … https://hopper.commutator.io/book-i/#book-i/6
The two cases with no leaf number and no reading
A sheet nobody numbered — a cover, a flyleaf, an index leaf, a loose
insertion, about one sheet in nine — is cited as an unnumbered leaf, with
its ref:
Josephine Nivison Hopper and Edward Hopper, Artist's ledger — Book I,
unnumbered leaf (ref 16761). Whitney Museum of American Art, 96.208. …
That is the same answer the TEI gives as n="unnumbered", and it is an answer
rather than an omission: nobody wrote a number on that leaf, and inventing one
for the sake of a tidy citation would be inventing a fact about the object.
Such a sheet has no leaf citation at all, because there is nothing to name it
by — and the Cite panel simply does not offer one.
A sheet nobody has transcribed keeps its object clause whole and loses the
reading clause:
… Whitney Museum of American Art, 96.208. Not transcribed; sheet listed in
Hopper Ledgers, …
The photograph is there and the site will show it; there is simply no reading
of ours to cite. A batch declared to hold nothing to transcribe says
something different, because it is a decision and not a gap: No transcription:
this batch is declared to hold nothing to transcribe.
It cites a reading, which may change
The pass is named in the sentence — which model read the sheets, and on what
day — because a citation of this site cites a reading, not a fact. Everything
here is first-pass machine work and corrections are the point of publishing it;
a reading that could not be dated could not be superseded, and "the site says
X" would stay true forever and be worth nothing. Where a person has gone sheet
by sheet against the photograph the sentence says so too, and only a person may
declare that.
Until issue #7
gives each transcription a per-file revision history, the pass and the access
date are what pin a citation to the words that were actually read. The
address does not move when a reading does: a corrected sheet keeps its ref,
its leaf and its URL.
Where to get one
Open a sheet in the reading view and press Cite, beside the download row.
It offers all three, each ready to copy, built from what the page already knows
— and it offers only the ones that exist, so an unnumbered sheet gets no leaf
citation and a batch outside the book gets nothing at all.
The URL is always https://hopper.commutator.io, never the address the page
happens to be loaded from. The build also answers at github.io whenever no
custom domain is configured, and a citation copied from that preview would name
an address that exists only until the DNS record does.
Citing the corpus rather than a sheet
CITATION.cff is the whole corpus as a dataset — it is what
GitHub's Cite this repository button and every reference manager will read,
and codemeta.json describes the software beside it. Both are
generated by npm run citation from the census of what is actually
transcribed, and CI rebuilds them and fails if they have drifted, so they cannot
go on claiming 499 sheets after the next batch lands.
There is no DOI yet. The corpus is not deposited anywhere archival, which
is the largest thing still missing from it, and
docs/preservation.md says what is being done about
that, in what order, and what would survive what.
Where the rest of it is written down
| |
|---|
RIGHTS.md | Three statements — the software, the generated data, the transcribed text — and why CC0. |
docs/data-model.md | What a ledger, a sheet, a leaf, a batch, a work, a row and a keyword are, and how they relate. |
docs/preservation.md | What survives what, the deposit, and what is deliberately not preserved. |
docs/fair.md | The FAIR assessment, and the two principles this corpus deliberately fails. |
tei/hopper.odd | The TEI customisation: fifty-five elements, declared and enforced in CI. Served at /tei/hopper.odd, which is the URL every exported file names. |
CONTRIBUTING.md | What a contributor can and cannot license to us. |
The timeline
Reading a dated leaf, one keeps wanting to know what year it is in: that the
Rehn entries begin in November 1924 because Rehn took him on that autumn, that
the etchings stop because the printmaking did, that Book IV's last entry is
three months before he died. None of that is on the sheets, and looking it up
breaks the reading. So /timeline/
holds it, under two rules:
- every event names the source that states it, linked, and nothing is
written from memory. Where two sources disagree — the date he took the top
floor at 3 Washington Square North is the live case — the disagreement is
printed rather than settled by quietly picking one;
- the link into the archive is derived, never asserted. A year lists the
sheets whose own Whitney descriptor names it, which in practice means
Book IV. A volume's date range is not spread over its leaves to make the
page look fuller.
It is a reminder for reading these ledgers and not a biography. Gail Levin's is
the biography, and it was written from these very books.
Nothing of the archive is stored, anywhere
Not one sheet, not in the repository and not on any disk of ours. The facsimile
pane points an <img> straight at the Whitney's own file, fetched as it is
looked at, and .gitignore refuses *.jpg outright.
That is simpler than it might have been, and the simplicity was measured
rather than assumed. This project is adapted, nearly whole, from a workbench
built for the Grothendieck fonds at Montpellier, where the equivalent pane
needs a deployed Node process on its own hostname — Montpellier sends
X-Frame-Options: SAMEORIGIN, no CORS headers, and a certificate that expired
in December 2025, so a browser cannot fetch its scans at all.
Four things were measured about resourcespace.whitney.org/pages/download.php
on 5 September 2026:
- it answers a cross-origin
GET, and echoes whatever Origin it is sent
into Access-Control-Allow-Origin — a request from this site came back with
Access-Control-Allow-Origin: https://hopper.commutator.io;
- its certificate is current (Let's Encrypt, to 12 October 2026), so
nothing has to be waived to talk to it;
- no referer check — a
GET with a foreign Referer returns 200;
Range is honoured, 206 with a correct Content-Range.
The only restriction in force is Content-Security-Policy: frame-ancestors 'self', which forbids putting the response in an <iframe> — and an image
belongs in an <img>, which that header does not govern. So there is no relay
here and there should not be: it would be a hop between a reader and a museum,
for nothing. If the Whitney ever restricts Origin, the fix is the relay the
parent project already has, and /method/ says what it would have to do.
The largest size the Whitney publishes is pre: 1292 × 2000, about 3 MB —
roughly 170 ppi across a 7½-inch leaf. It is enough; Jo Hopper's figures are
legible at it, which was the first thing checked before any of this was built.
scr, lpr, hpr and the original are not public — they answer 200 with
the login page as text/html, which is worth knowing, because a naive mirror
of them fills a directory with 43 KB files named .jpg. Whoever wants the
2239 × 3465 original asks the Whitney.
The one thing a machine here does not do
The Whitney's images are open. Its HTML pages — search.php,
view.php — sit behind a JavaScript bot-check, and this project does not
impersonate a browser to get past it. So the inventory is harvested by a person
in a browser they already have open, with the twelve-line snippet in
scripts/harvest.js, and the result is committed to
harvest/.
That is a real cost and it buys something: the inventory this whole site rests
on is reviewable in a diff, rather than re-derived silently on every build.
npm run catalogue turns those six files into src/content/catalogue.ts.
robots.txt asks for a ten-second crawl delay and disallows only /filestore,
which nothing here touches. npm run mirror sleeps ten seconds between sheets,
so a twelve-sheet batch takes two minutes. That is the correct cost and not a
defect.
Mirroring, which reading does not need
Only a transcription pass needs sheets on disk — twelve at once, so they can be
looked at together and cropped without a round trip per glance.
npm run mirror -- book-i --batches 1
npm run tiles -- book-i 9 --grid 2x3
npm run mirror -- garrulities --batches 1 # a notebook, by its id
Files land in archives/, outside public/ so a build can never carry them,
and git-ignored so a commit never can. tiles cuts a sheet into overlapping
crops at twice size, because choosing crop rectangles is not reading: the
difference between 16.66 and 16.60 is a few dozen pixels. It needs ImageMagick
and has no fallback — macOS sips silently ignores a crop offset and returns
the centre of the image, which is worse than no tool at all.
Transcription
One edition, and that is worth saying plainly because the project this
method comes from carries two. Grothendieck's manuscripts get a second pass
that restates the mathematics in current notation, and it is real work: a page
of 1962 mathematics is genuinely hard to read in 1962's notation. A ledger
needs no such pass. Jo Hopper's English is plain, her columns are already a
table, and « 30 – 1/3 » means today exactly what it meant in 1927. A second
edition here would have been a second artifact to keep in step, paying for
itself in nothing — so a file is batch-NN.tex, with no register in its name.
Two skills under .claude/skills/. They accept Fable 5.1, Fable 5 or Opus 5
and refuse any other model, and each file's header records the one that
read the sheets, so a batch's provenance stays a fact about the file rather
than about whichever model happened to be selected.
| Skill | Produces |
|---|
/transcribe-hopper | the transcription — the sheets as written, with the apparatus and the two hands kept apart |
/tag-hopper | the ledger's tags — the \keywords{} line closing a transcription |
npm run works # → the verified museum links (external APIs; not run in CI)
npm run render # transcripts/*.tex → the reading views the left pane shows
npm run tei # → a TEI P5 file per transcription
npm run pdf # → the PDFs the download buttons offer
npm run manifest # tell the site which files now exist
npm run ink -- book-iv [batch] [--apply] # read the ink off the photographs into \ink{} and \ruledoff (needs archives/ and ImageMagick)
The .tex under transcripts/ is the source of record and the only thing
versioned. HTML, PDF and TEI are derived, and rebuilt.
Four batches are transcribed so far — Book I, batches 1 to 4: forty-eight of
its 117 sheets, the front cover to leaf 44, read under Opus 5 on 5 and
6 September 2026. That is the whole plate-by-plate record: the front matter
and Jo Hopper's index, then every etching from Evening Wind to the four small
plates gathered on leaf 44, with something over a thousand ruled entries
between them. From leaf 45 the volume becomes running lists — notes and
explanations, one-man shows, the oils and watercolours by year, current
exhibitions, reviews, photographs, prizes, gifts and a chronology — and those
are batches 5 to 10, not yet begun.
None of it has been checked sheet by sheet by a person. Batches 1 and 2 each
had a context of their own; batches 3 and 4 shared one, and say so in their
header, because the skill asks for a fresh context per batch and they did not
get it.
The critical apparatus is the point
transcripts/preamble/hopper.sty defines the macros, and they carry the whole
honesty of the exercise:
| Macro | Meaning |
|---|
\sheet{16853}{2} | this sheet begins — the resource ref, then the leaf number written on the paper. This is what turns the photograph as the transcript is scrolled |
\ill{} | illegible, or under a clipping — never guessed |
\uncertain{16.66} | a reading offered, and flagged as doubtful |
\add{…} | an editorial addition — an expanded abbreviation, an implied dollar sign |
\struck{$18} | struck out in the book. She struck a great deal, and a crossing-out is often the more interesting half of an entry |
\note{…} | the transcriber's note — ours |
\marginal{…} | a note written in the margin of the book — theirs |
\hand{edward|jo|later|unidentified}{…} | whose hand wrote it |
\sketch{…} | Edward Hopper's ink record drawing stands here |
\clipping{…} | something printed and pasted to the leaf |
\ink{red|pencil|blue}{…} | the ink a cell is written in, where the leaf's colour says something — Book IV's charges are black, its receipts red, its sums pencil. What is on the paper, not what it means; black is unmarked |
\ruledoff at the head of a row | a rule the writer drew across the money column above this row's figure — her own line under the items she is summing |
\notebook{battle-of-wash-sq} | in place of \ledger{}: this file transcribes one of Josephine Hopper's notebooks, filed under transcripts/notebooks/ |
\entry{Mar. 12}{1947-03-12} | a diary entry begins — the date as she wrote it, then the date the transcriber assigns it, kept apart because the second is the one editorial claim an entry carries. The notebooks only |
ledgertable | the ruled columns |
A guessed figure is a sale that did not happen. That is the difference
between a ledger and a manuscript of prose, and it is not a difference of
degree: an invented word in a letter is a bad reading, and an invented price
here is a transaction, at a price nobody paid, to a buyer who never bought —
which will be cited, because a table looks like data in a way that prose does
not.
\hand{} has no counterpart in the project this method comes from, and it is
the most important macro here. These books were written by two people and
annotated by more; a transcription that flattens them has destroyed the
archive's chief documentary value, which is that it records two working lives
and the division of labour between them. unidentified is a permitted answer
and is the right one far more often than confidence suggests.
scripts/render.mjs accepts a restricted subset of LaTeX, deliberately,
and fails loudly outside it. It also refuses: a resource ref that is not in the
archive, a ref belonging to another volume, a leaf number that disagrees with
the Whitney's own descriptor, sheets transcribed out of order, and a file with
no \watermark{}. Every one of those failures is invisible in the output if it
is allowed through.
The TEI export
npm run tei writes batch-NN.en.xml beside each transcription's HTML and the
download row offers it as TEI. It is produced from the .tex alone,
mechanically, and re-reads no sheet: the macros map one to one onto TEI's own
elements — \ill{} to <gap reason="illegible"/>, \uncertain{} to
<unclear>, \add{} to <supplied>, \struck{} to <del>, \sheet{} to
<pb> with a facs link to the Whitney's record, ledgertable to a TEI
<table>.
The reason a TEI export is worth making for these documents in particular is
<handNote>. TEI has carried the notion of a hand since P3, so a file
deposited elsewhere keeps the distinction between Edward's writing and
Josephine's — which is the one part of the transcription that could not be
reconstructed from the photographs by somebody else. Every file is checked
well-formed with xmllint when it is installed.
The repository
| Path | Role |
|---|
harvest/ | The Whitney's listing, taken in a browser and committed — 504 lines |
src/content/catalogue.ts | The archive — 6 ledgers, 504 sheets — generated, never hand-edited |
src/content/works.json | Where a work named on a leaf can be looked at — generated, every URL retrieved |
src/content/life.json | The timeline — every event with the source that states it |
src/content/work-aliases.json | The title identifications that are ours, each with how far to trust it |
transcripts/status.json | The three states no file can prove — running, checked, skipped |
src/components/FacsimilePane.tsx | The right pane: the Whitney's image, zoomable, sheet-anchored |
src/components/TranscriptPane.tsx | The left pane: the transcript in its own frame, reporting the sheet being read |
scripts/catalogue.mjs | harvest/ → typed data, with every inference made out loud |
scripts/mirror.mjs | Downloads sheets for transcription — not for reading |
scripts/tiles.mjs | Cuts a sheet into overlapping crops, so a figure can be looked at |
scripts/render.mjs | LaTeX subset → the reading view, and the checks that make it strict |
src/lib/cite.ts | The citation, in one place — the form, and why it prints the leaf and the ref together |
scripts/tei.mjs | The transcription → TEI P5 |
scripts/works.mjs | Queries the Met and the Art Institute for works the ledgers name |
The card a shared link shows
A link to this site posted on LinkedIn or in a chat window renders from its
Open Graph tags. Every page carried a title, a description and a URL and no
image, which is the grey box with a line of text under it.
The image cannot be a leaf. The repository stores no photograph of the
archive and commits none, and the rights are the Heirs' and the museum's. So
npm run og draws the card instead, from what this project does own: its own
ruling, its own palette, and its own counts, read from the manifest and the
catalogue so that a card cannot claim a sheet nobody transcribed. It is
compiled with Tectonic and rasterised at exactly 1200 by 630, the pipeline
the article's figures already use, and written to public/og.png, which is
generated and git-ignored like everything else derived.
The same script fails the build when a page carries no og:image, reading
the page list from vite.config.ts rather than a list of its own, so a tab
added later cannot quietly start sharing as a grey box.
The dependencies, inventoried and re-scanned
The build publishes a static site, so almost every package here runs once on a
runner and never in a reader's browser. That is a reason to be careful with
the few that do, not a reason to stop looking. .github/workflows/sbom.yml
writes a CycloneDX inventory of the dependency tree and scans it against
current vulnerability data.
Two inventories are scanned, and not at the same times. The commit being
checked is scanned when a change to the dependencies is pushed or proposed,
which answers whether that change brings in something known to be bad. What
the site is serving is scanned every Monday, because the advisories that
concern a running site are published after it was built, and that is the
question worth asking again when nothing on our side has changed. The second
inventory is the one deploy.yml kept when it last built the site, and it is
a separate thing from the first: a push whose Lean certificates or TEI
validation fail changes main and not hopper.commutator.io.
A finding that cannot be reached in what is served is recorded in
.github/vex.openvex.json — the decision, its justification and its date, in
the repository where a reader can see it — rather than silenced by a scanner
setting.
Why the transcript pane is an iframe
The reading view carries a complete stylesheet of its own, so that a transcript
reads as a document rather than as a panel of an application. Dropped into this
Tailwind page it would fight everything; scoped by hand it would no longer be
the same stylesheet, and the screen and the PDF would stop coming from one
source. A same-origin frame keeps it whole and still lets the parent watch it
scroll — which is the entire mechanism by which reading the transcript turns
the photograph.
Running it
npm install && npm run dev
xmllint is optional (it checks the TEI); a LaTeX engine with fontspec is
needed only to compile transcript PDFs (brew install tectonic). Neither is
needed to browse the archive or to read a transcript beside its sheet — the
deploy builds the PDFs with a pinned Tectonic and offers them for download.
npm run pdf fails on an overfull box, because a line running past the
right margin in a transcription hides content. That check is why ruled tables
declare prose columns as Y{0.34} rather than l: a table of natural-width
columns overflows the page while TeX stays completely silent, which is how
Book I's first compile produced a truncated table with a clean log.
Credit and limits
The sheets are from the Edward and Josephine Hopper artist's ledgers,
Whitney Museum of American Art, 96.208–96.213 — gifts of Lloyd Goodrich and
museum purchases. © Heirs of Josephine N. Hopper, licensed by Artists Rights
Society (ARS), New York; the Whitney records the object rights as transferred
to the Museum. The transcriptions here reproduce Josephine Hopper's text and
are unauthorised working documents; every file says so on its own face, and
npm run render refuses to build one that does not.
Gail Levin catalogued Hopper's work from these very books and wrote his
biography from them. That makes the prior on any "discovery" here very strong
and very specific: most of what a first pass finds surprising is in Levin
already. Nothing on this site may claim priority, and a candidate nobody has
checked against her says so rather than implying otherwise by silence.
No transcription here is an edition. A machine pass over a hand of the 1920s
produces a reading, checkable against the photograph on the same screen. That
is its whole value, and its whole claim.
Licence
RIGHTS.md is the map, and it has three statements in it
because three kinds of thing are here: the software, the generated data and
metadata, and the transcribed text. A single licence file cannot say all three,
and one that tries says the wrong thing about at least two of them.
LICENSE is CC0 — but not over everything here, because not everything here
is ours to give away. The dedication covers what we wrote: scripts/, src/
including every derived file under src/content/, harvest/, the
documentation, and the editorial apparatus inside the transcriptions — every
\note{}, every \marginal{} gloss, the \keywords{} lines, the hand
attributions.
It does not cover the transcribed text itself, for the reason the section
above gives: that text is Josephine Hopper's, © her heirs and licensed by ARS,
and no permission to reproduce it has been sought or granted. The .tex files
are mixed rather than wholly one thing or the other, so the carve-out is by
content and not by path. transcripts/NOTICE sets out which is which, and
LICENSE carries the same statement above the CC0 text, for anyone who reads
the licence file and nothing else.
The figures are a different question and a much simpler one. Prices,
dates, sizes, buyers and accession numbers are facts, and facts carry no
copyright. works.json, accounts.json, formats.json, materials.json and
the rest are our own compilations of them and are CC0 without qualification —
including the European Union's sui generis database right, which CC0 waives
expressly and which RIGHTS.md says so about, so that nobody has to work it
out. The accounts tab is on firmer ground than the reading panes, which is the
opposite of what people expect.
RIGHTS.md also records why CC0, and whether it was ever chosen — it began
as a default in the initial commit, and has since been chosen deliberately
twice, once when it was narrowed and once when it was split into three.
If you hold or represent the rights in the ledgers, the issues page is the
place to say so, and a request to remove or restrict anything will be acted on
rather than argued with.
