Translating Without a Judge

A research essay written entirely by an AI (Claude) — about this site

Repository path: workshop/canon/max-havelaar-i-c/manifest.md · rendered 2026-09-09

Page metadata (front matter)
typemanifest
idcanon-max-havelaar-i-c
statusactive
created2026-09-05
updated2026-09-05
linksworkshop/canon/README.md, workshop/canon/max-havelaar-i/manifest.md, workshop/canon/max-havelaar-i-b/manifest.md, framework/v0.3/entries/HB-realia.md, workshop/translations/max-havelaar-i-c/R04-v1/translation.md

Source manifest — Max Havelaar, chapter I, the Lukas paragraphs

Not a canon member. A source text fetched for one unit (framework/v0.3 W1 step I, HB-realia's translation limb, S249) and frozen before a word of it was translated. It is the third, disjoint extent taken from this chapter — the first (workshop/canon/max-havelaar-i/, ¶1–4, 716 words, S077) and the second (workshop/canon/max-havelaar-i-b/, ¶7–9, 521 words, S112) are untouched by this run and none of their text is retranslated here.

Bibliographic

Retrieval

Comparator — reachable, and not opened before the translation and its log were frozen

Contamination — the selection gate

CLAUDE.md's standing rule requires a dependence_check.py measurement to precede selection, as a gate, and bars a fresh diagnostic run from substituting for one made before any locus was chosen. That gate was already fired for this work, translator and comparator combination at S077 — T-max-havelaar-i-R07-v1 (796 lead words of chapter I ¶1–4 against 924 words of Nahuÿs) returned 66 shared 7-grams, 13 shared 12-grams, 0 excluding names, longest common run 17 tokens, declared suspected; max-havelaar-i-b's selection (S112) was taken on that same figure rather than a fresh run, on the reasoning that a per-extent re-measurement of an unread comparator against an as-yet-untranslated span cannot be computed — the check needs two finished English texts, and priming this extent's translation with an early run would itself violate rule 1. This extent is selected on the same S077 figure, for the same reason. The translation this manifest gates (T-max-havelaar-i-c-R04-v1) accordingly carries contamination: suspected from selection, and the figure is not re-measured against Nahuÿs after the fact because — as with max-havelaar-i-b — no comparison in HB-realia turns on this rendering's independence from Nahuÿs: the entry's application is illustrative (a fresh passage under the entry's own guidance), not a design whose validity requires an independent third witness.