Ulysses: a first reading, and the apparatus built around it

The Difficulty Curve

Dimension 11 · Type III, Case study (primary evidence from one reader) · Build status: built · Panel score: not yet scored

First in the build order, the only dimension budgeting all four registers, and the first dimension in the project to be built to every register it budgets — HTML, diagram, presentation and, on 19 July 2026, the document PDF. That PDF is the first the project has produced by its own pipeline and the first to meet base document 00's definition of the register: contents to section to back, cross-references that open the reading record at the cited page, and a bookmark outline. Its nine-kind taxonomy of difficulty is the source of dimension 15's prompts, which is what anchor A15 required.

The claim
Where a first reader of Ulysses actually struggles is measurable from a complete dated record of one accompanied read — and it does not match the folklore. The difficulty is real, but it moves: period reference in the front half, form in the middle, voice and referent at the end; and question density tracks engagement at least as much as struggle.
The unit
the logged question — 290 extracted, 245 placed against episodes, page-anchored across sixteen exchanges of an accompanied first read
What it shows
This reader did not break at Proteus, where folklore says readers break — episodes 2–3 drew zero placed questions. His hardest sustained stretch was Sirens through Oxen of the Sun (episodes 11–14), where the record's only survival language appears. The three most-questioned episodes — Circe (40), Penelope (38), Ithaca (33) — are the ones he loved loudest and read fastest: the curve of questions is a curve of engagement wearing the costume of difficulty.

The artifact

Registers

The rule that built it

The inclusion rule
"245 questions could be placed against the book's eighteen episodes (290 were extracted in all)"; "Episode assignment is conservative": "7 questions spanning episodes or addressing the book or edition as a whole were excluded from episode counts" — the conservative-assignment rule. Stated in artifacts/difficulty-curve/curve.html (the placement rules in the caveats and ).
What was rejected under it
No "What was left out" section; the exclusions are recorded in the caveats with counts and grounds: 45 of 290 extractions not placed against an episode (curve.html); the retrospective episodes-2-3 question "left out of episode counts under the conservative-assignment rule" (same file); 7 cross-episode/whole-book questions excluded (same file); 38 duplicate extractions merged (same file); 4 extraction entries failed strict verbatim verification — kept in counts, barred as peak quotes (same file). No separate rejection file exists in data/.
What kind of thing it claims
A REPORTED DISTRIBUTION — under joint session 09's discriminating test (an inventory can only be attacked at its inclusion rule; a reading at its claim): Its claim field (data/dimensions.json) says "it does not match the folklore," but the artifact binds that in its own voice to n=1 — "That is all this proves: this one reader, forewarned... did not break there" (curve.html) — and disarms the count as a measure: "A question is not a unit of difficulty... Counts measure conversation, not struggle" (same file). What the record sustains is where 245 placed questions fell and how their kinds migrated — a distribution of one accompanied read; session 09 names dim 9 the only attackable claim among the dimensions it graded (docs/panel-session-09.md).

In the reading

Sources

Confidence

n = 1, an accompanied read, stated in the document's own voice throughout (anchor A10). A question is not a unit of difficulty; the artifact's caveats section carries every known way the dataset can mislead, including four quote-verification failures out of 290 extractions and the undercounted early episodes.

All dimensions · by episode · the working document