What this declares, and why a subset is the honest form

TEI P5 has some six hundred elements. This corpus uses fifty-eight, and the fifty-eight are not a sample of the six hundred: they are the exact reach of a mechanical conversion from a small set of LaTeX macros. Every element here exists because a macro in the transcriptions maps onto it, and the mapping is one to one — the table in scripts/tei.mjs is the whole of it. Nothing in the export is a judgement the .tex did not already make.

Publishing the subset matters more here than the usual reason. A reader who is handed a TEI file with no customisation has been told that six hundred elements are possible and that none of them is promised, which is a way of saying nothing. The list below is a promise: these elements and no others, these attribute values and no others, and a build that breaks the promise fails rather than quietly widening it. scripts/tei-validate.mjs reads this file and enforces it against the export on every run, in both directions — an element used and not declared is an error, and an element declared and used nowhere is an error too, because a declaration nobody has read is a claim about the corpus that is not true.

The customisation is a pure subset. It removes; it adds nothing, it renames nothing, and it changes no content model. So a file from this corpus is valid against TEI P5 unmodified — which is the property that makes the export worth depositing — and the subset is a statement about what a consumer will actually meet, not a dialect they would have to learn.

The one thing that could not be reconstructed without it

These are two hands. Edward Hopper drew the record sketches and wrote the dimensions; Josephine Hopper wrote everything else, over fifty-four years, in four or five hands of her own that are datable. Which hand wrote a line is a reading — it is made by looking at the sheet, it is recorded in hand in the .tex, and it is the one part of the transcription that could not be recovered from the photographs by somebody else without doing the whole work again.

TEI has carried the notion of a hand since P3. That is the reason this export exists at all: handNote and hand on seg carry the attributions out of this repository and into a format that will outlive it. Everything else in the subset is there to hold them up.

The customisation What the division is on the page. One leaf of the ledger, headed as the transcription heads it. A leaf is the unit the Hoppers numbered. A division within a leaf that is not a work — a run of entries the transcription sets apart under its own heading. The record of one work, which is what a leaf of the paintings books is organised around. One entry of a notebook, headed by its date as written; the date the transcriber assigns it is the when of the date in the head. The notebooks only. Offered rather than read: the transcriber proposes a reading without having read it. What is on the paper that is not writing. Neither is reproduced here: the export records that it is there and what the transcriber saw in it. Edward Hopper's ink drawing of the finished work, made for the record. A cutting pasted to the leaf — a review, a photograph, a dealer's notice. The heading is the title of a work as the ledger writes it, and not the transcriber's description of a leaf. Present on the transcriber's own notes and absent on the book's. The distinction is the point: a note with place of margin is something written on the leaf, and an editorial note is not. The transcriber's observation about the leaf. How the writing appears. The three typographic values carry the transcription's own emphasis; the three inks record the colour on the leaf, which in Book IV is its status field — charges black, receipts red, sums pencil — and is recorded as what is on the paper, not as what it means. Black is unmarked. A rule the writer drew across the money column above this row's figure — her own line under the items she is about to sum. Present only where the leaf has it. The facet the tag declares, from the taxonomy each file's header carries. A tag that declares none is a tag and not a classification, and the attribute is then absent rather than empty.

Generating the schema from this file

This ODD is the source. The RELAX NG schema for the subset is derived from it, and is not kept in the repository, for the same reason no other derived file is kept: a copy that can drift from its source will.

To generate it, use the TEI's own stylesheets — Roma, at roma.tei-c.org, or OxGarage, or teitorelaxng from the tei-xsl package:

teitorelaxng --localsource=p5subset.xml tei/hopper.odd hopper.rng

Continuous integration does not run that step, and the reason is worth stating rather than hiding. Generating a schema from an ODD needs an XSLT 3.0 processor and the TEI stylesheet library — Saxon and some forty megabytes of XSL — to produce a schema whose only job would be to re-state constraints that scripts/tei-validate.mjs already checks directly against this file, in seconds, with no network. What CI does run is the part that cannot be shortcut: every exported file, and this ODD itself, validated against unmodified TEI P5 4.12.0 with xmllint --relaxng.

So the chain is: TEI P5 validity, proved against the Consortium's own schema at a pinned version and a checked digest; and subset conformance, proved against this file. A generated hopper.rng would sit between the two and prove neither of them again.

What this rests on TEI Consortium. , version 4.12.0, 28 July 2026. www.tei-c.org/Vault/P5/4.12.0/. The version this corpus is validated against, by the Vault URL rather than the moving one. Rahtz, Sebastian and Burnard, Lou. . In , pp. 193–196. New York: ACM, 2013. doi:10.1145/2494266.2494321. The account of the mechanism this file uses, by the two people who built it. Wilkinson, Mark D., et al. . 3, 160018 (2016). doi:10.1038/sdata.2016.18. Assessed against, and departed from in two places on purpose — see docs/fair.md.

One reference was carried for a while and is not here, because it does not exist: a book by Sebastian Rahtz called . Searching his bibliography turns up no such title. What he actually built is the thing this customisation stands on — the TEI stylesheets, OxGarage, and the ODD processing chain — and the paper above is where he and Burnard describe it. A plausible citation is worse than none, and this project would rather say what it checked.