Repository path: workshop/canon/torture-esperance/manifest.md · rendered 2026-09-09

Source manifest — «La Torture par l'espérance» — DISCARDED by its own contamination gate

• Work: «La Torture par l'espérance», in Nouveaux Contes cruels, Paris: Calmann Lévy, 1893, pp. 17–32 (first published 1888)
• Author: Auguste de Villiers de L'Isle-Adam (1838–1889); public domain.
• Source: French Wikisource, validated transcription of the Calmann Lévy djvu, fetched 2026-08-02 (S089). Stored at source-fr.txt, 1,942 French words, 27 paragraphs.
• Comparator (never stored, never displayed except as recorded below): the anonymous English "The Torture by Hope" in Masterpieces of Mystery in Four Volumes: Riddle Stories, ed. Joseph Lewis French, Project Gutenberg #29704, 1,865 words.

The gate fired, and the work was discarded

Same discard rule, frozen before the measurement.

The run: "the venerable pedro arbuez d'espila sixth prior of the dominicans of segovia" — twelve tokens of which nine are a proper name and an ecclesiastical title, i.e. the one string in the passage with almost no translational freedom. That is exactly the case tools/dependence_check.py's own documentation warns about and this project's standing caution names, and it is the honest reading of this number: the rule fired on a stretch of text where two independent translators would be expected to agree. The work was discarded anyway, because the rule was frozen and had already fired once that session, and re-reading a threshold after watching it fire twice is the move the verification discipline exists to stop.

The nulls are weaker here than in the «Menuet» cell: they are stories by other authors and other translators in the same anthology, not other work by the same translator.

One exposure, declared

While debugging this gate's heading extraction — the anthology's story titles are followed by an all-caps author line, which collapsed the extraction to zero words and produced a vacuous "clean" — roughly twenty words of the comparator's opening were printed to the lead's console. The gate translation was already frozen, so the 12-token figure above is a pre-exposure measurement; the exposed text belongs to the gate span and not to any locus; and no line of the work that was eventually translated («La Mort de Pan», a different author and volume) was ever exposed. Recorded because a contamination event visible only to the operator is exactly the kind that does not otherwise get recorded.