Repository path: workshop/translations/la-nuit/collation.md · rendered 2026-09-09

«La Nuit (cauchemar)» — copy-text, collation, and the search for an English comparator

Written before the revision was made and before any experiment on this translation was designed. It exists because S100 found that three apparently independent e-texts of Minna Canth's «Köyhää kansaa» were one transcription of one posthumous reprint wearing three coats, and the standing consequence — note (bif) — is that a copy-text is checked rather than assumed.

1. The copy-text

French Wikisource, Clair de lune (recueil, 1905)/La Nuit, fetched 2026-08-04 through the MediaWiki action=parse API. The page is transclusion of the proofread Page: layer of Maupassant — Clair de lune, 1905.djvu — P. Ollendorff, Œuvres complètes vol. 15, 1905, pp. 277–291.

Stored at source-fr.txt with paragraph indices added, 38 paragraphs, 1,873 French words, SHA-256 a2f61e673f816a57c0588963f1ca2a23fbea98cabf82add4d29438e9fbb0edaf.

It is a posthumous reprint — Maupassant died in 1893 — and it is at least the third printing: the story appeared in Gil Blas on 14 June 1887 and was collected in the Ollendorff augmented Clair de lune before this volume. Neither earlier printing was reachable, and that is a real limit on everything below. Two French printings were searched for and not found: Project Gutenberg's Claire de Lune (ebook 11199) is the Monnier 1884 first edition, which predates the story's addition to the collection and does not contain it (verified by grepping the fetched file for the story's fixed proper nouns: 0 hits); three volumes of Gutenberg's Œuvres complètes de Guy de Maupassant (13, 16, 11) were fetched and likewise carry no rendering of it. Gallica's Gil Blas run was not attempted; archive.org's full-text search returned a proxy error from this environment.

2. Collation against two independent transcriptions

collate.py, committed alongside, normalises apostrophes, ligatures, quotation marks and dashes, drops bare digits, and diffs the copy-text word stream against each witness. Both witnesses are stored under witnesses/.

The athena witness differs from the copy-text at six points and not one of them is a word. All six are œ written oe (sœurs, cœur ×2, œufs) and à written a — an unaccented-uppercase transcription convention. A witness that agrees with the copy-text on 1,821 words out of 1,821 is not independent evidence about the text; it is evidence that both derive from the same transcription lineage. That is exactly the S100 shape, caught early this time, and it is why a single corroborating e-text is worth nothing.

The free witness carries four substantive variants, and is therefore the only witness in this collation doing any work:

3. ¶30, «Je sommai» / «Je sonnai»

The context is a bell-pull: «Je tirai le bouton de cuivre, et le timbre tinta dans la maison sonore… J'attendis, on ne répondit pas, on n'ouvrit point la porte. Je sommai de nouveau ; j'attendis encore, — rien !»

sonnai — "I rang again" — is the reading the surrounding sentences make obvious, and it is what a compositor would produce from sommai or a transcriber from a worn nn. sommai — from sommer, to summon formally, to serve notice on — is odd used absolutely, and is the harder reading.

Two of the three witnesses read sommai, and one of those two is not independent (§2), so the witness count is really one against one. The copy-text reading is kept, on the ordinary editorial ground that the harder reading is the one a copyist corrects away from and not toward, and the variant is declared on the translation (D26). The English renders the copy-text: "I summoned again."

This does not settle the text. A reading of the Gil Blas printing or of the Ollendorff Clair de lune would settle it, and neither was reachable. What the collation establishes is that the question exists and which way this translation went.

4. The search for an English comparator — a negative result, recorded

CLAUDE.md's standing rule requires the lead's contamination to be measured on candidate material before a design is built on it. The measurement could not be run: no published English translation of this story was reachable. What was searched, all on 2026-08-04, all by direct HTTPS GET or documented API call:

1. Project Gutenberg, Complete Original Short Stories of Guy de Maupassant (ebook 3090, 2.8 MB, the thirteen-volume collection). Grepped for the story's fixed proper nouns — Crédit Lyonnais, Château-d'Eau, Halles, Column of July — and for the opening sentence. The single Credit Lyonnais hit is in «Garçon, un bock !». The story is not in it.
2. Six further English Maupassant volumes on Gutenberg (21327, 21655, 22069, 17375, 17376, 593, 7114, 3664): 0 hits on the same probes.
3. English Wikisource, The Complete Short Stories of Guy de Maupassant: the index carries no page whose title contains Night; a prefix search over the collection returns no such page.
4. archive.org full-text search: ia-fts.archive.org returned a proxy 502 from this environment.

So contamination: on T-la-nuit-R04-v1 reads suspected, unmeasured, and the artifact says so in those words. The declaration is a placeholder and is labelled as one.

5. What this page licenses

That the copy-text is a 1905 posthumous reprint; that one freely reachable witness is not independent of it; that the second witness disagrees at four points, one of which is a copy-text error now emended and one of which is an unresolved crux; and that no English comparator was reachable. It licenses nothing about the text of the Gil Blas or Ollendorff printings, which nobody here has seen.