Repository path: workshop/translations/black-cat/R04-v1/translation.md · rendered 2026-09-09

T-black-cat-R04-v1 — Edgar Allan Poe, "The Black Cat" (1843), the second cat

Pair. English → Japanese. The project's first translation out of English, and its first lead translation into a language other than English.

Twenty prior lead translations, nine source languages, one target: English every time. That is the exact shape of the gap this artifact exists to fill. TH-20260724-translation-distance-axes C1 claims that grammaticalised categories the target lacks are lost and repaid, if at all, in the lexicon — and since the 2026-07-25 retraction its two surviving instances both have English as the target, so the theory page's own note records that "the regularity is about category mismatch, not about which language is strange" has become "a conjecture the evidence does not reach". Filling it is named there as "the highest-value single addition to C1", and has been the top item of ARM-breadth since S015.

Why this source, and not a fresh one. Poe's "The Black Cat" is already a cell in this project: A-baudelaire-chat-noir reads Baudelaire's 1857 French against it. Translating the same story into Japanese gives one English source into two targets — one near/near with maximal lexical inheritability (French), one far/far with none (Japanese) — which is a within-source control the project cannot buy any other way. C1's gap statement asks for the English-as-source case "in a pair not conflated with lexical overlap"; English→Japanese is that pair, and having the French cell already built is what makes the contrast free.

Source. Edgar Allan Poe (1809–1849), "The Black Cat," Saturday Evening Post, 19 August 1843. Public domain. Text as stored at wiki/base/anchors/A-baudelaire-chat-noir/black-cat-poe-1843.txt (Project Gutenberg #2148, retrieved 2026-07-24, S013), whose header carries a text-witness caveat about a line in the cellar scene — that line is outside this span.

Span. Paragraphs 14–20 of the stored file (0-indexed from the provenance header, so paragraph 1 is "For the most wild…"): 880 source words, 4,913 characters, 7 paragraphs. The span is the complete arc of the second cat, from first sighting in the gin-shop to the gallows on its breast — self-contained, and chosen because it is where English's obligatory categories are load-bearing rather than incidental (see the log, §A).

Declared purpose (purpose-fit). A reading edition in modern literary Japanese — である narration, first person 私, 新字新仮名, leaning on 漢語 where Poe leans Latinate and dropping to 和語 where he drops to the Germanic floor. Deliberately not the prewar 翻訳調 register, and the reason is methodological rather than aesthetic: the only free published Japanese rendering of this story is from the 1930s, and if this translation adopted that register, any later measurement of convergence between them would be measuring register imitation. Choosing a different register is the cheapest available way to keep that measurement interpretable. It buys the interpretability at a price — period is now confounded with translator in any comparison, exactly as it is in A-beowulf-ingeld — and the price is stated here rather than discovered later.

Contamination: suspected, declared before measurement, with the exposure written out.

• The work is famous and the Japanese rendering is on Aozora Bunko, which means it is mirrored across the open web many times over. The prior for having seen it is high. Every measured lead translation so far has been English output compared against English comparators; this is the project's first contamination question about non-English output, and the existing instrument (tools/dependence_check.py) tokenises on whitespace and cannot answer it. A character-level instrument was built for this artifact and is run after this file is committed, not before.
• One admitted priming event, and it is in the span. Before translating, this session read the first ~60 lines of A-baudelaire-chat-noir, which quote "hogsheads of gin, or of rum" → "tonneaux de gin ou de rhum". That is one phrase of paragraph 14, seen in French. It cannot contaminate a Japanese-against-Japanese measurement, but it is a reading event on the span under note (abm) and is declared rather than argued away. No other sentence of paragraphs 14–20 was read in any translation before this file was written; no Japanese rendering of any part of this story was read.
• No Japanese comparator was fetched, opened, or listed by content before this file's freeze commit. What was established beforehand was existence only, from Aozora's title index: a row giving title, author, translator and card number. Metadata, not prose.

Regime. R04 v0.1 (lead close translation, source-only). Cost $0; never ledgered (charter §3, A4).

Evaluation. None designed. The lead never judges its own translation (charter §5). The log below is frozen at this file's first commit; the anchor A-sasaki-kuroneko was built afterwards and cites this log, never the reverse.

Extent. 880 source words → 2,246 non-whitespace Japanese characters, 7 paragraphs, one-to-one with the source's. Counted in Python, not with wc (note (abl)).

A convention. The stored English marks italics as _word_. The Japanese does the same, and _…_ here means 傍点 — the standard Japanese emphasis device. Poe's small-capital GALLOWS has no case-based equivalent; it is handled at log point 21.

黒猫（抄）

ある夜、悪所というのも生ぬるいような店で、私が半ば正体を失って坐っていると、ふと何やら黒いものに目を惹かれた。それは、ジンだかラムだかの巨大な樽——その部屋の調度といえばまずそれくらいのものだった——その一つの鏡板の上にうずくまっていた。私はもう幾分か前からその樽の上をじっと眺めていたのであって、いま私を驚かせたのは、そこに載っているものにもっと早く気づかなかったという、まさにその一事であった。近寄って、手で触れてみた。黒猫であった——たいそう大きな——プルートーに優るとも劣らぬ大きさで、ただ一点を除いてはどこもかしこもよく似ていた。プルートーの身体には白い毛が一本もなかったが、この猫には、輪郭こそさだかでないが大きな白い斑があって、胸のあたりをほとんど覆い尽くしていた。触れると、すぐさま起き上がり、高く喉を鳴らし、私の手にすり寄って、目をとめてもらったのがさも嬉しいという様子であった。これこそ、私の探し求めていた当のものであった。私はただちに亭主に買い取りたいと申し出た。ところが亭主のほうは、それは自分のものではないと言い——そんな猫は知らぬ——見たこともない、と言うのだった。

私はなおも撫でてやっていた。やがて帰り支度をすると、その獣はついて来たそうな素振りを見せた。私は好きにさせ、道々、時折かがんでは軽く叩いてやった。家に着くと、たちどころに居ついてしまい、たちまち妻の大のお気に入りとなった。

私はといえば、ほどなく胸のうちに嫌悪の兆すのを覚えた。予期していたのとはまるで逆であった。だが——どうしてそうなのか、なぜそうなのかは知らぬ——私に対するあからさまな懐きようが、かえって胸を悪くさせ、苛立たせたのである。その嫌悪と苛立ちとは、じわじわと嵩じて、憎悪の苦さにまで昇りつめた。私はその生きものを避けた。ある種の羞恥と、かつての残虐の記憶とが、手を上げることだけは押しとどめていた。幾週かのあいだ、私は打ちもせず、そのほか手荒に扱うこともしなかった。けれども次第に——ごく次第に——言うに言われぬ嫌悪をもって眺めるようになり、疫病の息から逃れるように、そのいまわしい気配から黙って逃げ出すようになった。

この畜生への私の憎しみに輪をかけたのは、疑いもなく、連れ帰った翌朝の発見であった——プルートー同様、これもまた片方の目を失っていたのである。だがこの一事は、かえって妻の情をつのらせるばかりであった。妻は、すでに述べたとおり、かつては私の身上でもあり、私のもっとも素朴でもっとも清らかな喜びの源でもあった、あの人情というものを、人並みはずれて濃く具えていたのである。

ところが、この猫への私の嫌悪が増すにつれて、私への肩入れのほうも増してゆくかに見えた。読者に呑み込ませるのは難しかろうが、それはしつこく私の足もとについて回った。私が坐ればきまって椅子の下にうずくまるか、膝に飛び乗って、いまわしい甘え方で私を覆い尽くした。立って歩き出せば足のあいだに入り込んで転ばせかけ、さもなければ長く鋭い爪を衣服に食い込ませ、そうやって胸まで攀じ登ってくるのだった。そのようなとき、私は一撃のもとに叩き殺してやりたいと願いながら、なおそれを差し止められていた。ひとつには先の罪の記憶によって、だが主としては——この際いっそ白状してしまうが——その畜生に対する絶対の恐怖によってである。

この恐怖は、身に加えられる害への恐怖というのとは、正確には違っていた——といって、ではどう言えばよいのか、私には言葉が見つからぬ。白状するのが恥ずかしいほどである——そう、この重罪人の獄舎にあってなお、白状するのが恥ずかしいほどなのだが——あの獣が私に吹き込んだ恐怖と戦慄とは、およそ考えうるかぎりもっとも他愛のない妄想の一つによって、いよいよ高められていたのだ。妻は幾度となく、先に述べたあの白毛の斑の恰好に私の注意を促した。それは、この見知らぬ獣と、私が殺したあの獣とを分かつ、ただ一つの目に見える違いであった。読者は憶えておられよう、この斑は、大きくはあってもはじめはひどく曖昧なものであった。それが少しずつ——ほとんど感じ取れぬほど少しずつ、そして長いあいだ私の理性が気の迷いとして退けようともがいた歩みで——ついには、厳密なまでに輪郭のさだかなものとなっていたのである。それはいまや、名を口にするのも身の毛のよだつ、ある物の形をなしていた——そしてこのゆえにこそ、私は何にもまして憎み、恐れ、_もし度胸さえあったなら_この化物を始末してしまいたかったのだ——それはいまや、そう、おぞましい——身の毛もよだつ物の——_絞首台_の形であった！——ああ、〈恐怖〉と〈罪〉との、〈苦悶〉と〈死〉との、悲しくも怖ろしい仕掛けよ！

いまや私は、ただの〈人間〉の惨めさを越えて惨めであった。しかも_一匹の畜生_が——その同類を私は侮って殺したのだ——_一匹の畜生_が、この_私_に——至高の神の似姿に象られた人間であるこの私に——これほど堪えがたい苦しみをもたらそうとは！ああ！昼も夜も、私はもはや安息という恵みを知らなかった！昼のあいだ、その生きものは私を一刻も独りにしてはおかず、夜のあいだは、言うに言われぬ恐怖の夢から一時ごとに跳ね起きては、_あの物_の熱い息が顔にかかり、その途方もない重み——振り落とす力もない、肉を得た悪夢が——永久に私の_心臓_の上にのしかかっているのを見出すのだった。

Translator's log

Written at translation time, frozen at this file's first commit, before any Japanese rendering of this story was fetched and before the anchor was designed. Decisions and alternatives, not quality claims; the one place a quality word appears it is marked. Twenty-four points.

§A — why this span, stated first because it is a decision and not a given

1. The span was selected for grammatical load, not for narrative interest, and the two happened to coincide. Paragraphs 14–20 are where the story's engine runs on the categories at issue: a new animal is introduced (definiteness), is or is not the old one (definiteness again, plus the pronominal one), has one of two eyes (number), is addressed as both he and it inside four sentences (animacy), and ends as _the thing_ (definiteness carrying the horror). Paragraphs 1–13 have the same categories but they are inert there. The alternative span considered and rejected was 22–27 (the murder and the cellar), which is rich in realia — the pair's cultural-mediation behaviour — and would have measured the wrong thing.

§B — the categories English has and Japanese does not

2. Articles: 880 words, and the count of decisions is one. English forced Poe to mark every count noun as definite or indefinite; Japanese did not let me mark any of them, and — this is the point — it also did not make me decide. The single site where I decided anything was It was a black cat—a very large one → 「黒猫であった——たいそう大きな——」, where I dropped 一匹の. 一匹の黒猫であった is available and is what a Japanese translator taught to mark indefiniteness would write. I refused it because Japanese numeral-classifier phrases are marked: 一匹の foregrounds singularity rather than merely permitting it, so using it to render a overshoots. The rest of the paragraph's articles cost nothing at all: the landlord→亭主, the animal→その獣, the house→家 — bare nouns, no choice presented.

Recorded because it is the exact converse of T-beowulf-ingeld-R04-v1's finding that "Modern English requires a determiner on nearly every count noun, so 'a ring-band', 'the ale-cup' … are all decisions the source did not have to make." Same story, opposite direction: here it is the source that made them and the target that need not. The category does not transfer and it does not generate work either. That asymmetry is the thing I did not expect and is what §D is about.

3. Number: four sites, one repaid by a lexical item, one by a demonstrative, two lost silently. it also had been deprived of one of its eyes → 「これもまた片方の目を失っていた」. Japanese 片目 / 片方の目 is a lexicalised one-of-a-pair — it says "one of two" in a way English needs a partitive construction for. I chose 片方の目 over the tighter 片目 because 片目 reads as a settled condition (one-eyed as a state) while Poe's clause is about a discovery. This is C1's "repaid in the lexicon" running backwards: the target that lacks the grammatical category has a word for the thing the category expresses, so the loss is zero here.

one of the immense hogsheads cost a restructuring rather than a word. Japanese cannot hang one of off a plural noun the way English does, because there is no plural noun to hang it off. I moved the cask out of its relative clause into a dash-parenthesis and wrote 「巨大な樽——…——その一つの鏡板の上に」, where その一つ recovers the partitive and, by recovering it, implies the plural. The alternative — 「巨大な樽の鏡板の上に」 — is shorter and loses both, and loses with them the picture of a room lined with casks. The other two sites went silently and neither presented a choice: these feelings of disgust and annoyance → 「その嫌悪と苛立ちとは」, dreams of unutterable fear → 「恐怖の夢」.

4. Subject pronouns: the largest single class, and almost all of it evaporated without a decision. Poe's span carries roughly forty third-person pronouns for the cat. Japanese zero-anaphora took nearly all of them: 「触れると、すぐさま起き上がり、高く喉を鳴らし、私の手にすり寄って」 renders Upon my touching him, he immediately arose, purred loudly, rubbed against my hand with no third-person expression whatever. 彼 was live and was refused everywhere: modern literary Japanese uses 彼 of a cat only in translationese, and using it would have imported the very category I was supposed to be losing.

5. And with the pronouns went the thing they were carrying — he against it, on one animal, inside four sentences. This is the site I expected to be the hardest and it turned out to be the one where the target simply refuses to play. Poe writes, of the same cat and within four sentences:

Upon my touching him, he immediately arose … I at once offered to purchase it of the landlord; but this person made no claim to it—knew nothing of it—had never seen it before.

The animal is a him while it is being caressed and an it the moment it becomes a thing to be bought. Japanese has no way to mark that at all, because it has no obligatory third-person pronoun to modulate. What I did was not a repayment; it was a substitution of a different device: I let the verbs carry it — 「すり寄って、目をとめてもらったのがさも嬉しいという様子であった」 uses 〜てもらう, a benefactive that construes the cat as an agent with a viewpoint, and then the purchase sentence drops to 「それは自分のものではない」「そんな猫は知らぬ」, where the cat is それ and そんな猫. So the animacy shift survives, but it has moved from the pronoun system to the honorific-benefactive system and the demonstratives, and it is weaker and later than Poe's.

Two things worth separating here. (i) The English contrast is real and is within one animal — this is not a revival of the him/it claim about the two cats that RS-20260725-anchor-verification retracted, which was a claim about a contrast between Pluto and the second cat; the retraction explicitly left the within-animal shift standing as narrative voice. (ii) The French cell and this one lose it in two different ways: French must say il because le chat is masculine, so French has a pronoun and grammatical gender overrides the choice; Japanese has no pronoun to have a choice about. One source category, two targets, two distinct mechanisms of loss.

6. Tense and the pluperfect: the one place where the loss is a real loss and I could not repay it. I had been looking steadily at the top of this hogshead for some minutes, and what now caused me surprise was the fact that I had not sooner perceived the object thereupon. The whole sentence is built on English's ability to hold three time-layers apart — a prior durative, a present-of-narration surprise, and a prior negative perfect. Japanese has 〜ていた and 〜た and little else at this depth. I wrote 「もう幾分か前から……眺めていたのであって、いま私を驚かせたのは……気づかなかったという、まさにその一事であった」, buying the layering with lexical time adverbs (もう, 幾分か前から, いま) and the のである frame. That is C1's mechanism exactly, in this direction: grammar-borne meaning transcoded into lexis, and it is intermittent, which is the half of C1 that says the transcoding costs systematicity. I did it here because the sentence collapses without it. I did it again, cheaply, at that humanity of feeling which had once been my distinguishing trait → 「かつては私の身上でもあり」, where the adverb かつて does the whole job. And I did not do it at _a brute beast_—whose fellow I had contemptuously destroyed → 「その同類を私は侮って殺したのだ」, where the pluperfect flattens into plain た and the sequence — that he killed the first cat before this one arrived — is left to the reader's memory of the story. Three pluperfects, two repaid by adverbs, one dropped, no rule governing which.

7. The pronominal one has no Japanese counterpart and forced an explicitation. the sole visible difference between the strange beast and **the one I had destroyed** → 「この見知らぬ獣と、私が殺したあの獣とを分かつ」. English's pro-form one lets Poe avoid repeating the noun; Japanese made me write 獣 twice. そいつ was live and was refused as too colloquial for the register; もの was live and refused as too abstract for an animal. The translation is more explicit than the source at this site, and it is a compelled specification of exactly the kind D-20260727-08 was about — except that here the target is the article-less language, so the direction is the reverse of every case in that decision's evidence table.

§C — what did not come from a category mismatch

8. The noun ladder, which is the hardest thing in the passage and is not grammatical at all. Poe escalates: a black cat → the animal → the creature → the beast → the monster → _the thing_. Six steps, each a step further from personhood, and they are the passage's argument. I fixed a corresponding Japanese ladder before drafting and held it: 黒猫 → その獣 → その生きもの → 畜生 → 化物 → あの物. The rung that cost most was beast: 畜生 is a curse word in Japanese and carries a contempt beast does not, but 獣 was already spent on the animal and 動物 is clinical. I took 畜生 and accepted that the contempt arrives a rung early.

Logged because of the うるはし result in A-yosano-yomogiu: a thread held 3/3 by the translator at zero lexical overlap and 0/3 at maximum. This is a zero-overlap pair and I held 6/6 — with the qualification that I decided to before drafting, having read that result. So this is not independent evidence about zero overlap; it is a translator applying a finding. Stated so nobody counts it twice.

9. hogshead — a measure I converted by deleting the measure. 巨大な樽 says "huge cask"; a hogshead is a specific volume (~238 L). ホグズヘッド was live (copy-opaque, cf. Shaw's 三途の河); 大樽 was live. I took 巨大な樽 and lost the measure entirely — which is neither Garnett's measure conversion (пуд→"forty-pound") nor copy-opaque retention, but a third thing: measure deletion. Proposed to the handling catalogue at goodness-senses.md alongside measure conversion and preserved-inert.

10. den of more than infamy. 「悪所というのも生ぬるいような店」 — 悪所 is period-flavoured for a place of vice and is doing the work of infamy; 生ぬるい ("lukewarm") is doing the work of more than. Rejected: 「悪名高いどころではない」, which is flatter and closer.

11. Poe's Latinate/Germanic contrast, which is the one place this pair behaves like the French pair. Poe's style is high-Latin diction over a Germanic floor, and Japanese has the same two-storey lexicon (漢語 over 和語) from an unrelated history. So unutterable loathing → 「言うに言われぬ嫌悪」 (和語 idiom + 漢語 noun), absolute dread → 「絶対の恐怖」 (漢語), the blessing of rest → 「安息という恵み」 (漢語 + 和語). The register contrast transferred at essentially no cost — which is not what a far/far, zero-inheritability pair is supposed to make cheap, and is worth putting next to A-baudelaire-chat-noir's finding that the high-overlap pair flattened exactly this contrast because French cognates of Poe's Latin words are ordinary French stock. Two-storey-ness is not the same variable as inheritability.

12. chimaeras. 「他愛のない妄想」. 妄想 is 漢語 and clinical; キマイラ / 幻獣 would be copy-opaque and would make a reader think of the monster rather than the idea. Poe's word means the idea. Lost: the buried echo between a chimaera and the beast on the hearthrug.

13. felon's cell. 「重罪人の獄舎」. 独房 (solitary cell) is the natural modern word and is wrong: Poe's narrator is a condemned man in 1843, and 獄舎 keeps the period without archaising the sentence around it.

14. pertinacity, and a sentence I inverted. It followed my footsteps with a pertinacity which it would be difficult to make the reader comprehend → 「読者に呑み込ませるのは難しかろうが、それはしつこく私の足もとについて回った」. Japanese takes the relative clause badly at that length, so the concessive was moved to the front and pertinacity became the adverb しつこく. A noun became an adverb; the sentence's centre of gravity moved from the quality to the action.

15. Poe's long periods, kept rather than split. A-shaw-spider-thread records long source periods being split for English readability. Here the source is the long-period language and the target is one that tolerates long periods well, so I kept all of paragraph 19's chain of dashes intact, including the four-dash pile-up at the gallows. The one place I split was With my aversion to this cat, however, its partiality for myself seemed to increase → 「ところが、この猫への私の嫌悪が増すにつれて、私への肩入れのほうも増してゆくかに見えた」, where I turned Poe's absolute construction into a 〜につれて correlative because Japanese has no absolute construction and the alternative was a comma splice.

§D — three things I noticed while doing it, recorded because they bear on C1

16. The forced-explicitation half of C1 barely fired, and where it fired it fired at my discretion. Every prior lead log that records forced explicitation records it as pervasive and involuntary: T-genji-yomogiu-R04-v1 supplying "she", "he", "her father" at every clause; T-yanfu-yili-yan-R04-v1 supplying tense "on nearly every sentence"; T-beowulf-ingeld-R04-v1 on every count noun. Those are all translations into English. Working out of English into Japanese, the count is two sites — the pronominal one at point 7, and 畜生/the beast at point 8 — and both were choices I could have made otherwise.

17. The asymmetry is not symmetric, and the reason looks structural rather than accidental. A category the target obligatorily has forces a decision at every occurrence, because the slot must be filled. A category the source obligatorily has and the target lacks presents no slot at all — the information simply does not survive, silently, without asking. So "forced explicitation" and "silent flattening" are not two names for one phenomenon: the first generates decisions and shows up in a translator's log, and the second generates none and does not. (This is a description of what happened, not a claim that it generalises; n=1, one translator, one passage. internal-judgment-only.)

18. Which means the log corpus is biased in a way nobody has noticed, and this session is the first position from which it was visible. Twenty of the project's twenty-one logs are of translations into English, and RS-20260727-log-typology counted C4b — determinacy the target compels — at 14 decisions in 11 of 21 logs and treated that as a fact about translation. On this evidence it may be substantially a fact about English being the target. A corpus of logs made translating out of English would under-record it for exactly the reason at point 17. Recorded here, not asserted as a finding.

§E — form and small mechanics

19. である was chosen over だ and over です・ます. The narrator is writing a document ("to-day I would unburthen my soul"), not speaking; である is the register of a written confession. It also lets 〜のである carry the pluperfect scaffolding at point 6.

20. 私 throughout, and never dropped where Poe's I is contrastive. Japanese drops first-person subjects freely, and I dropped many. I kept 私 at every contrastive site — 「私はといえば」 for For my own part, 「この私に」 for for _me_ — because the whole passage is a man setting himself against an animal.

21. GALLOWS in small capitals had no case-based equivalent and was repaid with 傍点. _絞首台_. Rejected: 「絞首台！！」 (punctuation inflation), and 「ガロウズ」 (copy-opaque, and Japanese readers know 絞首台). The capitals in Poe are doing volume; 傍点 does emphasis. Not the same instrument, and the nearest one available.

22. The four allegorical capitals — Horror, Crime, Agony, Death, Humanity, the High God. English marks personification with a capital; Japanese has no orthographic device that does this. I used 〈 〉 — 「〈恐怖〉と〈罪〉との、〈苦悶〉と〈死〉との」, 「ただの〈人間〉の惨めさ」. This is a manufactured marker: Japanese does not conventionally use 〈 〉 for personification, and a reader will read it as some kind of emphasis without knowing which. Alternatives rejected: 傍点 (already spent on italics at three sites in the same paragraph, so the two would collide), and 片仮名 (would read as foreignness, not as personification).

23. Poe's italics, all six of them, kept as 傍点. _had I dared_, _a brute beast_ ×2, _me_, _the thing_, _heart!_. One of them changed function: Poe's _a brute beast_ … _a brute beast_ is a repetition and an italicisation, and in Japanese 「_一匹の畜生_が……_一匹の畜生_が」 the repetition carries almost all the weight, with the 傍点 adding little. Form preserved, function partly evacuated — which is the preserved-inert outcome named at TH-20260724 C4, arriving here by a different route than in the French cell (there, italics on a French-flavoured word that is unremarkable in French; here, an emphasis device competing with a repetition that is already emphatic).

24. One thing I could not do at all, and it is a pun the project should not miss. Poe's word for the white mark is indefinite — "a large, although indefinite splotch of white", then "had been originally very indefinite; but … assumed a rigorous distinctness of outline". In English, indefinite is simultaneously the word for a blur and the name of the article system that the whole passage's reference-tracking runs on. The mark becomes definite as the cat becomes the thing. Japanese has no article system for the word to be about, so 「輪郭こそさだかでない」 / 「はじめはひどく曖昧なもの」 renders the blur and loses the second sense entirely. There was nothing to decide; there was only something to notice.

Appendix — contamination, measured after the freeze

Added after the log above was frozen at 505520a. Nothing above was altered.

The existing instrument tokenises on whitespace and is blind to Japanese, so tools/dependence_check_cjk.py was written for this artifact: character-level, whitespace and punctuation deleted, tokenisation frozen in the module docstring, eleven Japanese-shaped self-test fixtures (note (ii)). It requires reference cells, because Japanese prose shares long strings with itself.

The subject clears every reference and that settles nothing, because all three references hold content apart while the subject pair shares a source text. The control that would separate remembering 佐々木 from being constrained by Poe is a third free Japanese rendering of this passage, and there is none.

The three shared regions: 「白い毛が一本もなかったがこの猫」 (15) · 「見たこともないと言うのだった私は」 (16, 14 within a paragraph — note (ss)'s boundary caveat, third session running) · 「そして長いあいだ私の理性」 (12). The second contains a shared explicitation: Poe writes no verb of saying at had never seen it before, and both translators supplied one.

Note (bce)'s probe test was run and has no power on this cell. All 28 locatable decision sites fall outside the shared regions — but the regions cover 2.12% of the span, so P(0 inside | independent placement) = 0.377. Recomputed on S037's Bécquer cell (11.7% coverage, 19 sites, 1 inside): P(≤1) = 0.168. Suggestive there, uninformative here; the note is amended to require the null.

Verdict: contamination: suspected stands and is not cleared. This translation is admissible as practice and as process evidence, and explicitly not as an independence measurement — the project's third such (T-jeli-il-pastore-R05-v1, T-rayo-de-luna-R04-v1). No claim in A-sasaki-kuroneko §§1–4 depends on it.

One post-freeze reading event, declared. tools/fetch_aozora.py prints the first 80 characters of the fetched text; that is paragraph 1 of the Japanese and is outside the span. It occurred after the freeze commit and could not have affected anything above.

Files: dependence/ (cells, texts, result.json, runs_vs_decisions_cjk.py).