Repository path: workshop/translations/monelle-paroles/contamination.md · rendered 2026-09-09

«Paroles de Monelle» — contamination, measured

Run after both renderings and both logs were committed and frozen (commit fcdd9c2), so no measurement could prime a translation, and before E-20260807f was designed. Tool: tools/dependence_check.py. Cells: /tmp/mono_cells.json construction is reproduced by ../../experiments/E-20260807f-carriage-or-strangeness/verify.py.

1. The priming event, declared rather than discovered

Before translating, the lead saw nine fragmentary OCR lines of the published English. They were returned by a grep -i Destroy over the downloaded comparator, run to establish that the comparator was reachable at all, and they are lines from what became S1 and S2. Verbatim, as they appeared:

—TI give you the word: Destroy! Destroy! · Destroy! Destroy within yourself and destroy all · —Destroy all good and all evil. Rubbish and · —Destroy the ancient habitations of men and · —Destroy, for all creation comes from destruc- · destroyed. So it is that the new art must be · —And form must be destroyed. · cophori; that you may destroy by fire and that the · young God that it may be destroyed by him.

Nothing else of the English was read until both artifacts were frozen. This is a real breach of R04 §Procedure 1 and it is reported as one.

2. The measurement

The longest shared run with Meloney, 16 tokens: "created let the old god offer his creation to the young god that it may be" — from Que l'ancien dieu offre sa création au jeune dieu afin qu'elle soit broyée par lui, where the two hands diverge only at the final verb (Meloney destroyed, lead crushed).

3. Declaration: contamination: suspected, and exactly what that means

The subject figure sits inside the range spanned by two pairs of demonstrably independent published hands, and well above the lower of them. 27 / 5 / run 16 against 6 / 0 / run 10 on one reference pair and 23 / 6 / run 17 on the other. The standing rule requires the 7-gram count and the run (CLAUDE.md; run length alone has failed as a proxy three times) and here they agree with each other and disagree with each reference pair in opposite directions. none is not defensible and high is not supported. suspected is the honest reading.

Two things the table shows that are not about this passage. (i) One of the two reference pairs — two published translators, six years apart, no plausible dependence — is itself flagged DEPENDENT? by the tool at 23 / 6 / 17. A tool verdict of DEPENDENT? on a 400-word cell is not evidence of dependence, which is the caution tools/dependence_check.py's own docstring gives and which this run instantiates. (ii) The lead matches itself across a single revision pass at 456 / 429 / run 133, which is note (bhb)'s figure again in the extreme: a re-rendering is not an independent second opinion of anything.

4. Why this does not disqualify the design it feeds, and where it does bite

E-20260807f's validity does not turn on the lead's independence from Meloney. All four arms are derived from T-monelle-paroles-R04-v1 and every contrast is within-lead and matched: the FORM factor is an operator applied to the lead's own English, the STRANGE factor is another, and no arm is being compared with a published translation. This is the boundary condition the standing rule states — the lead is never the independent third translator where a design's validity turns on independence** — and it is not met here.

Where it does bite, and the result page carries it as a limit: if the lead's CC English is partly remembered Meloney, then CC's formal carriage is partly Meloney's formal carriage, and the CARRIED pole of the FORM factor is less purely the lead's craft than the design's language implies. The measurement above bounds how much of CC that could be: five twelve-grams in 519 words.

5. Consulted

Logged in wiki/base/consulted.md. Marcel Schwob, Le Livre de Monelle (1894), Project Gutenberg

53374, read in French. William Brown Meloney V, trans., The Book of Monelle (Bobbs-Merrill, 1929),

archive.org bwb_C0-BOD-229 — published 1929, public domain in the United States since 2025; the 597-word passage is stored at comparator-meloney-1929.txt for reproducibility of the measurement.