Repository path: workshop/canon/kusamakura-vii-bath/manifest.md · rendered 2026-09-09

Source manifest — 草枕 第七章（湯壺の場） / Kusamakura, chapter VII, the bath

This is a stored source span, NOT a ratified canon member. D-20260723-02's criterion 3 (whole-work translatability, ≤ ~10,000 characters) is not met by Kusamakura, which runs to 83,979 non-whitespace characters. What is stored here is a defined span of it, committed so that a translation of it is reproducible. No canon-selection decision covers it and none is claimed; it was chosen by this session under ARM-framework for the reasons in §Selection.

Bibliographic

• Work: 草枕 Kusamakura — chapter 七, the opening of the bath scene
• Author: Natsume Sōseki 夏目漱石 (1867–1916)
• First publication: 1906, 「新小説」
• Register axis: Meiji aesthetic prose; a first-person 余 narrator; dense Sino-Japanese diction and a quoted Chinese line, alternating with colloquial reflection and an interpolated six-line verse.

Source (Aozora Bunko)

• Card: 000148 / work 776 · file: files/776_14941.html (Shift_JIS, decoded cp932)
• URL: https://www.aozora.gr.jp/cards/000148/files/776_14941.html
• Retrieved: 2026-07-27
• 底本 (source edition): 「夏目漱石全集3」ちくま文庫、筑摩書房 · 初出: 「新小説」
• 入力: 柴田卓治 · 校正: 伊藤時也
• Cleaning: tools/fetch_aozora.py (ruby <rt>/<rp> and Aozora ［＃…］ annotations stripped; paragraph breaks preserved). Full cleaned text: 83,979 non-whitespace characters, 0 gaiji notes.

The span, defined by index

Lines 976–998 (0-indexed) of the cleaned full text, being the first twelve non-empty paragraphs of chapter 七 after the two-line opening 「寒い。手拭を下げて、湯壺へ下る。」 — from 「三畳へ着物を脱いで…」 through 「…太棹かとも思う。」

• 1,956 non-whitespace characters, 12 non-empty paragraphs, of which 6 are the lines of the interpolated 土左衛門の賛.
• Stored verbatim as source.txt in this directory.
• Boundary rationale: the span begins at the paragraph after the chapter's two-clause opening and ends at the end of the paragraph in which the distant shamisen is first heard — the last sentence before the narrative turns to the memory of お倉さん. Both boundaries are narrative breaks, not word counts.

Public-domain verification

• Japan (charter rule — author's death + 70 years): Sōseki died 1916; 1916 + 70 = 1986 → public domain in Japan (nominal onset 1 Jan 1987). Over-determined: already public domain under the former 50-year term from 1 Jan 1967, which the 2018 extension did not revive. ✓
• US: published 1906 → public domain in the US. ✓

Selection — why this span

Chosen under ARM-framework for the second regime comparison, against three constraints:

1. Source and a published English translation both freely reachable (charter §7, A8, and CLAUDE.md's materials instruction). Source: Aozora, above. English: Kazutomo Takahashi (1862–1931), Unhuman Tour, The Japan Times, 1927 — Project Gutenberg #73131, public domain. This is the pairing the standing instruction prefers and it is rare for Japanese prose: Gutenberg holds exactly two Sōseki works in English.
2. Comparability with the project's only prior regime comparison. E-20260724-r01r02-selfrevise ran on 蜘蛛の糸 and 夢十夜 1–2 — Meiji/Taishō literary prose, JA→EN, one of them by this same author. Holding the pair and the period roughly fixed is what lets the lead-executable comparison sit beside the API one.
3. Not the anthologised opening. 草枕's first paragraph is among the most quoted in modern Japanese literature. A mid-chapter span reduces — it does not remove — the chance that a lead rendering is remembered rather than made.

Contamination is measured, not asserted, on the translation artifacts, after they were frozen; see T-kusamakura-vii-bath-R06-v1 §Contamination and RS-20260727c-arm-identifiability. The measurement is against Takahashi 1927 only. Kusamakura's better-known Englishes — Alan Turney's The Three-Cornered World (1965) and Meredith McKinney's Kusamakura (2008) — are in copyright and not freely reachable here, so the measurement bounds contamination from one side only, exactly as the Verga measurement did against Dole rather than Lawrence (RS-20260727-longwork-selection §1).