Repository path: workshop/canon/bettelweib-locarno/manifest.md · rendered 2026-09-09
Page metadata (front matter)
| type | manifest |
|---|---|
| id | canon-bettelweib-locarno |
| status | active |
| created | 2026-07-28 |
| updated | 2026-07-28 |
| links | workshop/canon/README.md, workshop/experiments/E-20260728c-length-matching/design.md, workshop/regimes/R06-lead-single-pass.md, workshop/regimes/R04-lead-close.md, wiki/arms/ARM-framework.md |
Source manifest — Kleist, «Das Bettelweib von Locarno», ¶1–3 (DE→EN)
This is a stored source span, NOT a ratified canon member. No canon-selection decision covers it and none is claimed; it was chosen by this session under ARM-framework step 3 for the reasons in §Selection. The whole story would in fact satisfy D-20260723-02 criterion 3 (960 words), and a later session may complete it; what is stored and translated here is the first three paragraphs.
Bibliographic
- Work: «Das Bettelweib von Locarno» — the whole story is 960 words in four paragraphs; the stored span is ¶1–3.
- Author: Heinrich von Kleist (1777–1811)
- First publication: Berliner Abendblätter, 10tes Blatt, pp. 39–42, Berlin, 11 October 1810, publisher Julius Eduard Hitzig.
- Register axis: late-Kleist reportorial narration — long hypotactic periods, heavy parenthesis, the whole of ¶2's apparition delivered in reported speech (Konjunktiv I) inside a single sentence. This is the property the span was chosen for: the syntax gives an English translator unusual latitude over sentence length, which is what
E-20260728cneeds to measure.
Source text (German Wikisource, page-level transcription of the 1810 printing)
- Page:
de.wikisource.org/wiki/Das_Bettelweib_von_Locarno,BEARBEITUNGSSTAND=fertig, transcluding four proofreadSeite:pages. - URLs retrieved 2026-07-28:
Seite:Berliner_Abendblätter_1810_039.jpg…_042.jpg, viaaction=raw. - Cleaning:
<noinclude>blocks,<section>markers,{{…}}templates and'''bolding removed; the four page transcriptions joined with a single space at each join, because all three joins fall inside a sentence (page 39 ends…indem etwas, das dem Blickand page 40 opensunsichtbar gewesen…). - Note (bch) applied. The failure that note records is a page-break marker splitting a paragraph and inflating the paragraph count while leaving the word count exactly right. The structure was therefore checked against the source rather than against the token total: the assembled text has 24 newline-separated blocks, of which blocks 1–5 are the story (title + four paragraphs) and blocks 6–24 are the rest of the newspaper page — an Arnim poem (
Räthsel auf ein Bild…) and aPolizeiliche Tages-Mittheilungennotice about eight missing baskets of peat. Those are not part of the story and are excluded. The story's own four paragraph breaks are each internal to aSeite:page, so no join creates or destroys one.
The span, defined by index
Blocks 2–4 of the assembled text, i.e. the story's paragraphs 1, 2 and 3 — from Am Fuße der Alpen… through …empfahl sich und reiste ab. Stored verbatim as source.txt in this directory.
- 361 words (Python
str.split(), per note (abl);wc -wis not used), 2,268 characters, 3 paragraphs of 163 / 132 / 66 words. - Boundary rationale: ¶3 ends the night's episode — the knight leaves — and ¶4 opens "Dieser Vorfall…" on a new time-scale and runs 599 words, longer than the three preceding paragraphs together. The cut is Kleist's own hinge, not a word count.
- Orthography preserved as printed 1810:
Schloß,Thür,Aechzen,gerathen,spuke. Not modernised.
Public-domain verification
- Author died 1811; life + 70 expired 1881 on any reckoning. Public domain worldwide. ✓
- The transcription is a Wikisource page-level transcription of an 1810 printing, itself PD; Wikisource text is CC-BY-SA / PD. ✓
Comparator reachability — the contamination gate, run before translating
Standing rule (CLAUDE.md, note (bcd)): measure contamination on the candidate material before designing the experiment. The gate could not be run: no freely reachable published English translation of this story was located. Three checks, all from actual tool results on 2026-07-28:
- Project Gutenberg catalogue search for
kleist— 16 results; every Kleist item is German or Finnish. No English Kleist volume. - Full-text check, The German Classics of the Nineteenth and Twentieth Centuries, vol. 4 (PG 12060, 1913, 933,980 characters — the volume whose Kleist section the catalogue points to).
Locarnooccurs once, at offset 557,481, inside a critical introduction sentence listing the author's works. The story itself is not in the volume: its own text names Locarno, so a translation present would give more than one hit. - Full-text check, Tales from the German (PG 32046, 1844, 1,364,998 characters):
Locarno0,Beggar0.
A web search returns only translations that are in copyright — Golffing (1960), Luke & Pasley (1978), Sublunary Editions (2021). None is storable and none was opened.
Which of the two cases this is (the note requires saying so): unmeasurable, not disqualified. The gate's binding clause is about works whose standard English translation is canonical; Kleist's is anthologised and the lead may well hold some of it, so contamination here is possible and unmeasurable, which is why the translations declare suspected rather than none. What makes the selection defensible anyway is stated in E-20260728c-length-matching §0 and repeated here because it is the load-bearing sentence: no evaluative weight of any kind is placed on this prose. The only quantities taken from these three texts are word counts and a token-level diff of two of them against each other. A remembered phrase would change none of them.
Selection
Chosen for one property: the syntax makes length a live choice. Kleist's periods can be rendered as one English sentence or as three, and the reported-speech apparition in ¶2 can be kept as indirect discourse or resolved into direct narration — differences of 10–20% in the English word count with no difference in what is being said. E-20260728c has to ask what a length constraint costs a translator, and that question is uninteresting on a source where the target length is nearly forced.
Second: the span is short enough to translate three times in one session (a draft, an unconstrained revision, and a length-matched revision), which the design requires and which a 2,000-word span would not permit.