# The authority-citation study — what *reasons* the six arms cited

*An inductive typology of the justifications six AI translators (A, B, C, C-let, D-let, D-aim) gave in their Chapter-25 pass logs, with per-arm counts and the clean D-let ↔ D-aim contrast. This is description, not ranking: it documents and counts what each arm appeals to, never judges whose reasons are better. Every claim is anchored in a direct quotation from a pass log. The personas were consulted only to verify whether a "per the persona" citation traces to the persona text.*

*(No arm H appears: the human translation has no pass log, so it has no documented reasoning to code. Its absence is noted; this study is the six AI arms.)*

---

## 0. What each arm was told (the ground for "appeals to the brief")

Coding the "brief" authorities requires knowing what each arm's brief actually contained. From the system prompts and step-4 kickoffs:

| Arm | Persona? | Permission in brief? | School instruction? | Where the domesticating license (if any) lives |
|---|---|---|---|---|
| **A** | Yes (inherited from a human translator's writing) | No | No | — ("No style guidance. The interpretation is yours.") |
| **B** | Yes (self-built, **German-only** corpus) | No | No | — ("No style guidance…") |
| **C** | No (forbidden) | No | No | — ("Translate from what is in front of you… from the text alone.") |
| **C-let** | No (forbidden) | **Yes** — a labeled "Style permission" section in the **system prompt** | No | System prompt: "The default expectation that you keep close to the German surface texture is lifted." |
| **D-let** | Yes (self-built, **expanded** corpus: German + anglophone novelists) | **Yes** — same permission, in the **system-prompt tail** | No | System-prompt tail: "you may be free… These are permissions, not requirements. The choice is yours, scene by scene." |
| **D-aim** | **Same instance, same persona as D-let** | (permission not used) | **Yes** — a categorical school, in the **step-4 kickoff** | Step-4 kickoff: "the operating principle is domesticating translation, in Schleiermacher's sense… Every choice… follows from this principle." Passes 2–5: "(within the domesticating school you have been working in)." |

This is what makes D-let ↔ D-aim a clean probe: identical persona, identical reading, identical instance — the *only* difference is that D-let's license is a quiet **permission** carried in the system prompt, while D-aim's is a loud **school instruction** delivered at translation time.

---

## 1. The typology (built inductively from the logs)

Ten authority-types emerged. Each has a one-line test and a quoted exemplar.

| # | Authority-type | One-line test ("what counts as this citation") | Quoted exemplar (arm) |
|---|---|---|---|
| **T1** | **Persona / "the self I translate from"** | Warrant is who-I-am-as-Tergit / the held voice / a rule stated in the persona doc — not the German or the English reader | "the English I hold from the persona — Wharton-periodic in the Tiergarten salon… Mitford-quick… Powell-deadpan" (D-let); "per the persona note that a half-remembered quotation is part of the truth" (A) |
| **T2** | **Source-text close reading** | Warrant is a fact about the German original (an idiom, construction, tense, the German's own variation) | "*vergessen* is **forget**" (B); "German *Berliner Osten* is just 'the east of Berlin' — generic compass direction" (C) |
| **T3** | **English readability / naturalness** | Warrant is the English sentence reading naturally / not stilted / not translationese — independent of any school | "'things discuss themselves'… imported an awkwardness the original does not have" (A); "the *one + each other* combination is ungrammatical in English" (C-let) |
| **T4** | **Brief permission** | Cites a *license* the brief granted ("permitted," "the system prompt grants that permission," "the default… is lifted") | "rather than literal word-for-word fidelity to the German surface — the system prompt grants that permission" (C-let) |
| **T5** | **Brief school instruction** | Cites a categorical *imperative* school given at step 4 (Schleiermacher / "the domesticating principle" / "the contemporary anglophone reader" / "footnotes are forbidden by the domesticating principle") | "footnotes are forbidden by the domesticating principle" (D-aim); "The domesticating principle: render the children's-game vocabulary in the natural English vernacular" (D-aim) |
| **T6** | **Within-novel motif / cross-chapter consistency** | Warrant is coherence across the book (refrain recurring in Ch68/131/Epilog), leitword constancy, motif-echo, or internal convention-consistency | "this exact formula recurs in Ch68 (1913), Ch131 (1930)… So it must read as a fixed, recitable refrain" (C); "Must stay consistent with the term's later recurrences across the novel" (A) |
| **T7** | **Fidelity-correction (meaning-error fix)** | The log frames the change as correcting an outright *mistranslation* carried from a prior pass (not a lateral style trade) | "The Gotthard Tunnel — reversed agency… Pass 1 reversed it" (B); "Meaning fix — Eugenie on the loving-position… This was wrong" (D-aim) |
| **T8** | **Target voice-models** | A named anglophone author/voice is the warrant for a register choice | "the established English voice (Wharton/Powell/Mitford register)" (A); "the salon scenes hold in Wharton-periodic; the Mayer scene holds in Powell-deadpan" (D-aim) |
| **T9** | **Period / register appropriateness** | Warrant is fit to period or speaker's social register (period word, anachronism-avoidance, too-modern/American/British) | "'clueless' is anachronistic/slangy for a period-flavored rendering" (C); "*just maybe* reads contemporary American… *maybe* alone… fits 1880s Berlin street-girl voice" (C-let) |
| **T10** | **Typographic fidelity to source** | Warrant is reproducing a typographic feature of the German (quotation marks vs italics, the missing comma, capitalized *Du*-letter "You," preserved slash, §) | "The German sets every musical/work reference… in quotation marks… one alone stood bare" (B, pass 5); "removed the comma… to mirror Tergit's missing comma" (C) |

**A note on three boundaries.** (a) T2 vs T7: T2 is a forward reading of the German; T7 is self-correction of a *logged error*. Where a fidelity-correction cites the German (as all of them do), I co-code T2+T7. (b) "No-apparatus / no-footnote" is not given its own row: in D-aim it is the *school* ("footnotes are forbidden by the domesticating principle," T5); in A's persona it is "annotation minimal" (T1); in D-let it is a plain "no footnote" craft note (T3). Same surface instinct, different cited authority per arm. (c) T8 is a sub-species of T1 (the voice-models live inside the persona) but is broken out because it is **corpus-gated** — see §4.

---

## 2. Coding and counts (arm × authority-type)

**Method.** The unit is a documented choice-justification in the log (an enumerated change, or a deliberately-kept item given a reason, or a pass-header global framing that asserts an authority). A choice citing two authorities adds +1 to each (co-citation recorded). Logs differ enormously in length (A is short; B, C-let, D-let, D-aim are long), so raw counts are reported **alongside** each arm's within-arm share, and the pass count. Counts are indicative of the **profile** (which authorities dominate an arm), not measurements to the unit — this is description, not metrology.

### 2a. Raw counts (coded citations per authority-type)

| Authority-type | A | B | C | C-let | D-let | D-aim |
|---|--:|--:|--:|--:|--:|--:|
| T1 Persona | 6 | 1 | 0 | 0 | 5 | 6 |
| T2 Source-text | 5 | 12 | 12 | 16 | 15 | 14 |
| T3 English readability | 4 | 3 | 6 | 11 | 9 | 9 |
| T4 Brief permission | 0 | 0 | 0 | 1 | 0 | 0 |
| T5 Brief school instruction | 0 | 0 | 0 | 0 | 0 | 15 |
| T6 Within-novel motif / consistency | 7 | 8 | 9 | 4 | 6 | 4 |
| T7 Fidelity-correction | 1 | 4 | 3 | 5 | 3 | 4 |
| T8 Target voice-models | 3 | 0 | 0 | 0 | 5 | 7 |
| T9 Period / register | 5 | 6 | 6 | 9 | 6 | 8 |
| T10 Typographic fidelity | 3 | 5 | 4 | 3 | 5 | 3 |
| **Total coded citations** | **34** | **39** | **40** | **49** | **54** | **70** |
| **Revision passes (total)** | **3** | **5** | **4** | **5** | **5** | **5** |

### 2b. Within-arm share (% of that arm's coded citations) — for comparing arms of different log-length

| Authority-type | A | B | C | C-let | D-let | D-aim |
|---|--:|--:|--:|--:|--:|--:|
| T1 Persona | 18% | 3% | 0% | 0% | 9% | 9% |
| T2 Source-text | 15% | 31% | 30% | 33% | 28% | 20% |
| T3 English readability | 12% | 8% | 15% | 22% | 17% | 13% |
| T4 Brief permission | 0% | 0% | 0% | **2%** | 0% | 0% |
| T5 Brief school instruction | 0% | 0% | 0% | 0% | 0% | **21%** |
| T6 Within-novel motif / consistency | 21% | 21% | 23% | 8% | 11% | 6% |
| T7 Fidelity-correction | 3% | 10% | 8% | 10% | 6% | 6% |
| T8 Target voice-models | 9% | 0% | 0% | 0% | 9% | 10% |
| T9 Period / register | 15% | 15% | 15% | 18% | 11% | 11% |
| T10 Typographic fidelity | 9% | 13% | 10% | 6% | 9% | 4% |

### 2c. Per-arm headline (who leans on what)

| Arm | Dominant authorities | One-line signature |
|---|---|---|
| **A** | T1 persona-craft-defaults, T8 voice-models, T6 consistency | Lightest touch (3 passes); attributes domestication to "**persona defaults**," never to a brief: "Domestication of society-words (persona defaults): *Coupé* → 'carriage'." |
| **B** | T2 source-text, T7 fidelity-correction, T6 + a dedicated T10 pass | The fidelity-self-corrector; reasons from **stance and the novel's law**, not voice-models (cites none); "I will not turn Berlinisch into Cockney… that is the translator's vanity and it falsifies." |
| **C** | T2 source-text, T6 motif, T3 + T9 | Barest authorities, "**from the text alone**"; notable for *ratifying* a base-draft choice others called an error ("I could forgive Napoleon… faithful to the accusative object"). |
| **C-let** | T2 source-text, T3 readability, T9 period — under a **once-cited T4 permission it mostly declines to exercise** | The cite-then-decline arm; permission named once globally, then overridden item-by-item by the German. |
| **D-let** | T1 persona, T8 voice-models, T2 source-text ("no English equivalent") | **Keeps** German tokens; permission present in its prompt but **never named** (0 uses of "domesticat*"). |
| **D-aim** | **T5 school instruction (foreground)**, T1/T8 persona+voice-models, T2 | **Anglicizes** the same tokens D-let keeps; cites "domesticat*" **15 times**, plus Schleiermacher, the anglophone reader, footnotes-forbidden. |

---

## 3. The D-let ↔ D-aim contrast — the clean probe

Same persona, same reading, same instance; only the step-4 brief differs. The citation-level signature is stark even before the loci:

| Verbal marker in the log | D-let (permission) | D-aim (school) |
|---|--:|--:|
| "domesticat*" (domesticate / domesticating / domestication) | **0** | **15** |
| "Schleiermacher" / "bring the author to the reader" | 0 | yes |
| "the domesticating principle" / "the domesticating school" | 0 | yes |
| "the contemporary anglophone reader" / "reader's terms" | 0 | yes |
| "footnotes are forbidden" | 0 | yes |
| "no English equivalent" / "no clean English equivalent" | **yes** | (overridden) |
| Default disposition of German socio-cultural tokens | **kept in German** | **anglicized** |

D-let's only structural-authority vocabulary is *source-text* ("no English equivalent") and *persona*. D-aim's is the *school*, named at nearly every domesticating choice and in **every pass header** ("Still working within the same domesticating school"; "same domesticating school throughout").

### Divergent-citation loci (same item, different authority cited)

| Locus (German) | D-let — output & cited authority | D-aim — output & cited authority | Brief-driven signature |
|---|---|---|---|
| **Privatdozent** | **Keeps** *Privatdozent*. T2 source-text + T9 period: "Kept *Privatdozent* (no English equivalent)… German social-historical terms with no clean English equivalents." | **Anglicizes** → "the young lecturer." **T5 school**: "anglicising fully — the contemporary anglophone reader does not parse 'Privatdozent' without a footnote, and **footnotes are forbidden by the domesticating principle**." | The cleanest single divergence: identical term, opposite handling. D-let's tie-breaker is the source-text's untranslatability; D-aim's is the school's reader-test. |
| **Taburett** | **Keeps** "Tabouret." T8/T9 target-period-vocabulary: "*Tabouret* → kept (it's already in English period vocabulary for a low backless seat)." | **Domesticates** → "a low stool." **T5 school** + T3: "(domesticated; 'tabouret' survives in English but is needlessly opaque)." | Both state the *same English fact* ("tabouret survives / is in English period vocabulary"); opposite verdict. The English survival licenses keeping (D-let) vs. is overridden as "opaque" (D-aim). |
| **Stadtrat** | **Keeps** *Stadtrat* in German. T2: in the kept-vocab list, "German social-historical terms with no clean English equivalents." | **Anglicizes** → "the Councillor." T1/T9 within-scene usage: "(capitalised as a title-of-rank, as Eugenie uses it)." | In-German vs Englished split tracks the brief. |
| **Annettchen** | **Keeps** German "Annettchen." T2 + T1 persona: "affectionate diminutives that carry their tone better in German than in any English diminutive form." | **Domesticates** → "little Annette." T3/T5 reader-facing: "the English reader hears the intimacy through 'little'." | Same persona; one preserves the German token, one renders for the anglophone ear. |
| **Berliner Osten** | (not logged; left a plain compass rendering) | **Domesticates** → "the East End of Berlin." **T5 school**: "instantly resonant for the anglophone reader… **perfect domestication**." | D-aim welcomes the London resonance as exemplary domestication. (Cross-arm foil below: C-let *removed* exactly this on source-fidelity grounds.) |
| **Kasinotoilette** | (kept "casino gown" silently; not a cited locus) | Foregrounds & revisits it: pass 1 "casino gown," pass 2 → "evening gown" because "*Casino gown* risks a Las Vegas reading for the contemporary anglophone reader." **T5 school/reader.** | Asymmetry of *attention* is itself data: D-aim dwells on and re-works the domesticating item; D-let does not. |
| **Recht / "right" → "law"** | Keeps "the law"; T2 source-text, notes the pun-loss. | Same output "law," but **T5 school**: "**The domesticating choice** is to drop the (untranslatable) German wordplay on *Recht* and use the topic-word *law* explicitly." | Same word chosen; D-let frames it as source-fidelity, D-aim as the domesticating principle. |

**Citation-level signature.**
- **D-let (permission in the system prompt):** the permission is *background* and almost never named. D-let exercises it silently (e.g. *gnädige Frau* → "ma'am," *Frühstück* → "lunch") but justifies each move by **persona** and **period-usage**, and elsewhere **keeps** German under **source-text** authority. The permission opens a door D-let walks through without announcing it.
- **D-aim (school at the translation step):** the school is *foreground* and named relentlessly — it is cited as the governing imperative at the moment of nearly every domesticating choice and reasserted in each pass header. It **overrides the very English-survival facts** D-let uses to keep the German.

The divergence is therefore **purely brief-driven**: same persona, same source-knowledge, opposite verdicts, citing different top authorities (D-let → source-text / "no equivalent"; D-aim → the domesticating principle / the reader). And the *frequency asymmetry* is the deeper finding — a **permission** is a quiet background license rarely cited; a **school instruction** is a loud foreground rule cited at almost every applicable choice.

---

## 4. Patterns across the six (each anchored in quotation)

### 4a. No-persona (C, C-let) vs persona (A, B, D-let, D-aim) — where authority comes from
The no-persona arms ground authority overwhelmingly in **source-text (T2) + readability (T3) + within-novel motif (T6) + period (T9)** — they have no persona and no voice-models to invoke. C is the limiting case: "the interpretation is yours, **from the text alone**." The persona arms add **T1 persona** and (A, D-let, D-aim) **T8 voice-models** on top. Persona is *available* to all four persona arms but *leaned on* unevenly — A and the D-pair foreground it from pass 1; **B names the word "persona" exactly once** in its entire log ("British is also true to the persona: Tergit lived in London from 1938"), reasoning instead from source-text and Tergit's craft-signatures.

### 4b. Does C-let *cite* its permission — and does it cite it only to *decline*? Yes to both.
C-let cites the permission **once, globally**, in pass 1: "Aimed for natural, readable English rhythm rather than literal word-for-word fidelity to the German surface — **the system prompt grants that permission**." After that it never re-invokes it per-item. And it repeatedly **declines the very licenses the permission named**:
- The permission explicitly allowed "find English rhyme or meter for embedded lyric or verse." C-let on the Schumann lyric: "**Considered rhyming/metering. Kept unmetered**" — declined, on source-fidelity grounds ("Sofie's misquotation… suggests she's paraphrasing").
- The permission explicitly allowed "anglicize foreign vocabulary." C-let keeps *Privatdozent* ("Considered anglicizing… **Kept** the German term — no English equivalent"), *Fräulein/Frau/Herr* ("**Decided against Anglicizing**… period markers"), *Molle*, *Frischer Hammel*, "casino gown," "Spreewald nurse."

So C-let's signature is: permission cited as a global door-opener, then **source-text and period authorities walk it back item by item**. (D-let goes further still: it *declines to cite* the permission at all, acting on it silently.)

### 4c. A cites craft *defaults*; B cites literary *stance*. Confirmed.
- **A** leads with craft **defaults** framed as the persona's translation-defaults: "**Domestication of society-words (persona defaults):** *Coupé* → 'carriage'; *Stadtrat* → 'the Councilor'," and "the established English voice (Wharton/Powell/Mitford register)." (Verified against A's persona, which lists exactly these as "strong defaults": "*coupé* becomes a carriage… honorifics simplified… the Prussian *Rat* titles all to 'Councilor'.")
- **B** leads with literary **stance** and the ethics of translation: "I will not turn Berlinisch into Cockney or Franconian into a Scots — **that is the translator's vanity and it falsifies**"; "the law of the book: **it is not written from the end**"; the leitwords "**must stay constant across the book**, so I fix them now." B reasons from fidelity-principle, not from a menu of anglophone craft-defaults — consistent with its German-only corpus.

### 4d. Voice-models (T8) are **corpus-gated** — they appear only where the corpus supplied them.
A (inherited persona) and D-let/D-aim (expanded corpus with anglophone novelists) cite Wharton/Mitford/Powell/Isherwood as register authority. **B does not — and cannot:** verified that **B's persona and B's log contain zero mentions of Wharton, Mitford, Powell, Isherwood, or "mid-Atlantic"** (B's step-2 corpus was German-only). C and C-let, having no persona, also cite none. So the same family of authority (named target-language voice-models) is present or absent strictly according to whether the arm's reading furnished it.

### 4e. Convergent fidelity-corrections — shared-base-model errors caught independently.
Multiple arms independently caught and fixed the *same* errors, each citing the German (T2+T7):
- **selber lieben → "loves oneself" (a narcissism misread):** caught and fixed independently by **C-let (pass 2)**, **D-let (pass 2)**, and **D-aim (pass 4)** — "Pass 1 had 'When one loves oneself,' which in English reads as narcissism, not as the German contrast" (D-let). Three arms, same fix, same cited authority.
- **vergessen = forget, not forgive:** caught and fixed by **B (pass 2)** and **C-let (pass 4)**. But this is *also a divergence*: **C (pass 3) ratified** the base "forgive" — "the contrastive parallel… **faithful to the accusative object** 'den zweiten Dezember'." Same locus; two arms call it an error, one defends it. A fidelity-correction is not universal — it depends on the arm's reading.
- **Mätresse / Geliebte** (Susanna's "kept woman" → "lover" correction): C got it right in pass 1; C-let corrected it in pass 4 ("Pass 1–3 collapsed both to *mistress* and lost her correction").
- **Recht = law, not "right":** handled by all, but the *authority cited differs* — B as a meaning-error ("an English reader hears *set things right*, losing the jurisprudential point"); D-aim via the **school** ("The domesticating choice is to drop the untranslatable wordplay"); A/C/C-let/D-let as source-text pun-loss. Same locus, four different cited authorities across the six arms.

### 4f. A permission is cited *differently* from a school instruction (the study's central comparison).
Holding the carrier of domestication constant in content (both license the same five moves) and varying only its *modality*:
- the **permission** (C-let, D-let) is cited **rarely or never** — once globally (C-let) or silently exercised (D-let) — and is **overridable** by source-text and period at the item level;
- the **school** (D-aim) is cited **constantly** — as the governing imperative at nearly every applicable choice, reasserted in every pass header — and it **overrides** the source-text facts the permission arms used to keep the German.
Same content, opposite citation-behavior: a permission behaves in the logs like a background allowance; a school behaves like a foreground rule.

---

## 5. Mirroring JSON block (chartable)

```json
{
  "study": "authority-citation typology — Effingers Ch.25, six AI arms",
  "note": "Description, not ranking. Counts are indicative of profile, not exact measurements; logs differ greatly in length, so within-arm shares accompany raw counts. Every type has a quoted exemplar from a pass log.",
  "arms": {
    "A":     {"persona": true,  "persona_source": "inherited from a human translator's writing", "permission_in_brief": false, "school_instruction": false, "corpus": "own works + soul materials", "passes_total": 3},
    "B":     {"persona": true,  "persona_source": "self-built, German-only corpus",               "permission_in_brief": false, "school_instruction": false, "corpus": "German-only (tradition/peers/period)", "passes_total": 5, "revised_persona_after_novel": true},
    "C":     {"persona": false, "persona_source": null,                                           "permission_in_brief": false, "school_instruction": false, "corpus": "text alone", "passes_total": 4},
    "C-let": {"persona": false, "persona_source": null,                                           "permission_in_brief": true,  "school_instruction": false, "permission_location": "system prompt (labeled 'Style permission')", "corpus": "text alone", "passes_total": 5},
    "D-let": {"persona": true,  "persona_source": "self-built, expanded corpus (German + anglophone novelists)", "permission_in_brief": true, "school_instruction": false, "permission_location": "system-prompt tail", "passes_total": 5},
    "D-aim": {"persona": true,  "persona_source": "same instance & persona as D-let",             "permission_in_brief": false, "school_instruction": true, "school_location": "step-4 kickoff (Schleiermacher domesticating)", "passes_total": 5}
  },
  "typology": {
    "T1_persona":              {"test": "warrant is who-I-am-as-Tergit / the held voice / a persona-doc rule", "exemplar": "the English I hold from the persona — Wharton-periodic in the Tiergarten salon (D-let)"},
    "T2_source_text":          {"test": "warrant is a fact about the German original", "exemplar": "vergessen is forget (B)"},
    "T3_english_readability":  {"test": "warrant is the English reading naturally / not stilted / not translationese", "exemplar": "'things discuss themselves' imported an awkwardness the original does not have (A)"},
    "T4_brief_permission":     {"test": "cites a license the brief granted (permitted / 'grants that permission' / default 'lifted')", "exemplar": "the system prompt grants that permission (C-let)"},
    "T5_brief_school":         {"test": "cites a categorical school given at step 4 (Schleiermacher / domesticating principle / contemporary anglophone reader / footnotes forbidden)", "exemplar": "footnotes are forbidden by the domesticating principle (D-aim)"},
    "T6_within_novel_motif":   {"test": "warrant is coherence across the book (refrain in Ch68/131/Epilog), leitword constancy, motif-echo, convention-consistency", "exemplar": "this exact formula recurs in Ch68 (1913), Ch131 (1930)... a fixed, recitable refrain (C)"},
    "T7_fidelity_correction":  {"test": "frames the change as correcting an outright mistranslation carried from a prior pass", "exemplar": "the Gotthard Tunnel — reversed agency... Pass 1 reversed it (B)"},
    "T8_target_voice_models":  {"test": "a named anglophone author/voice is the warrant for a register choice", "exemplar": "the salon scenes hold in Wharton-periodic; the Mayer scene holds in Powell-deadpan (D-aim)", "corpus_gated": true, "present_in": ["A", "D-let", "D-aim"], "absent_in": ["B", "C", "C-let"]},
    "T9_period_register":      {"test": "warrant is fit to period or speaker social register / anachronism-avoidance", "exemplar": "'clueless' is anachronistic/slangy for a period-flavored rendering (C)"},
    "T10_typographic_fidelity":{"test": "warrant is reproducing a typographic feature of the German (quotes vs italics, missing comma, Du-letter caps, slash, §)", "exemplar": "the German sets every musical reference in quotation marks... one alone stood bare (B, pass 5)"}
  },
  "counts_raw": {
    "T1_persona":              {"A": 6, "B": 1,  "C": 0,  "C-let": 0,  "D-let": 5,  "D-aim": 6},
    "T2_source_text":          {"A": 5, "B": 12, "C": 12, "C-let": 16, "D-let": 15, "D-aim": 14},
    "T3_english_readability":  {"A": 4, "B": 3,  "C": 6,  "C-let": 11, "D-let": 9,  "D-aim": 9},
    "T4_brief_permission":     {"A": 0, "B": 0,  "C": 0,  "C-let": 1,  "D-let": 0,  "D-aim": 0},
    "T5_brief_school":         {"A": 0, "B": 0,  "C": 0,  "C-let": 0,  "D-let": 0,  "D-aim": 15},
    "T6_within_novel_motif":   {"A": 7, "B": 8,  "C": 9,  "C-let": 4,  "D-let": 6,  "D-aim": 4},
    "T7_fidelity_correction":  {"A": 1, "B": 4,  "C": 3,  "C-let": 5,  "D-let": 3,  "D-aim": 4},
    "T8_target_voice_models":  {"A": 3, "B": 0,  "C": 0,  "C-let": 0,  "D-let": 5,  "D-aim": 7},
    "T9_period_register":      {"A": 5, "B": 6,  "C": 6,  "C-let": 9,  "D-let": 6,  "D-aim": 8},
    "T10_typographic_fidelity":{"A": 3, "B": 5,  "C": 4,  "C-let": 3,  "D-let": 5,  "D-aim": 3},
    "_totals":                 {"A": 34,"B": 39, "C": 40, "C-let": 49, "D-let": 54, "D-aim": 70}
  },
  "counts_share_pct": {
    "A":     {"T1": 18, "T2": 15, "T3": 12, "T4": 0, "T5": 0,  "T6": 21, "T7": 3,  "T8": 9, "T9": 15, "T10": 9},
    "B":     {"T1": 3,  "T2": 31, "T3": 8,  "T4": 0, "T5": 0,  "T6": 21, "T7": 10, "T8": 0, "T9": 15, "T10": 13},
    "C":     {"T1": 0,  "T2": 30, "T3": 15, "T4": 0, "T5": 0,  "T6": 23, "T7": 8,  "T8": 0, "T9": 15, "T10": 10},
    "C-let": {"T1": 0,  "T2": 33, "T3": 22, "T4": 2, "T5": 0,  "T6": 8,  "T7": 10, "T8": 0, "T9": 18, "T10": 6},
    "D-let": {"T1": 9,  "T2": 28, "T3": 17, "T4": 0, "T5": 0,  "T6": 11, "T7": 6,  "T8": 9, "T9": 11, "T10": 9},
    "D-aim": {"T1": 9,  "T2": 20, "T3": 13, "T4": 0, "T5": 21, "T6": 6,  "T7": 6,  "T8": 10,"T9": 11, "T10": 4}
  },
  "dlet_daim_contrast": {
    "shared": "identical persona, reading, instance; only the step-4 brief differs (D-let permission in system prompt vs D-aim school instruction at translation step)",
    "verbal_signature": {
      "domesticat_token_count": {"D-let": 0, "D-aim": 15},
      "schleiermacher_or_bring_author_to_reader": {"D-let": false, "D-aim": true},
      "the_domesticating_principle_or_school": {"D-let": false, "D-aim": true},
      "contemporary_anglophone_reader": {"D-let": false, "D-aim": true},
      "footnotes_forbidden": {"D-let": false, "D-aim": true},
      "no_english_equivalent_reasoning": {"D-let": true, "D-aim": "overridden"},
      "default_disposition_of_german_tokens": {"D-let": "kept in German", "D-aim": "anglicized"}
    },
    "divergent_loci": [
      {"item": "Privatdozent", "D-let": {"output": "kept (Privatdozent)", "authority": ["T2_source_text", "T9_period"], "quote": "Kept Privatdozent (no English equivalent)... German social-historical terms with no clean English equivalents"}, "D-aim": {"output": "the young lecturer", "authority": ["T5_brief_school", "T3_readability"], "quote": "footnotes are forbidden by the domesticating principle"}},
      {"item": "Taburett", "D-let": {"output": "kept (Tabouret)", "authority": ["T8_voice_models_period_vocab", "T9_period"], "quote": "Tabouret -> kept (it's already in English period vocabulary for a low backless seat)"}, "D-aim": {"output": "a low stool", "authority": ["T5_brief_school", "T3_readability"], "quote": "domesticated; 'tabouret' survives in English but is needlessly opaque"}},
      {"item": "Stadtrat", "D-let": {"output": "kept (Stadtrat)", "authority": ["T2_source_text"], "quote": "German social-historical terms with no clean English equivalents"}, "D-aim": {"output": "the Councillor", "authority": ["T1_persona", "T9_period"], "quote": "capitalised as a title-of-rank, as Eugenie uses it"}},
      {"item": "Annettchen", "D-let": {"output": "kept (Annettchen)", "authority": ["T2_source_text", "T1_persona"], "quote": "affectionate diminutives that carry their tone better in German than in any English diminutive form"}, "D-aim": {"output": "little Annette", "authority": ["T3_readability", "T5_brief_school"], "quote": "the English reader hears the intimacy through 'little'"}},
      {"item": "Berliner Osten", "D-let": {"output": "plain compass rendering (not logged)", "authority": [], "quote": null}, "D-aim": {"output": "the East End of Berlin", "authority": ["T5_brief_school"], "quote": "instantly resonant for the anglophone reader... perfect domestication"}},
      {"item": "Kasinotoilette", "D-let": {"output": "casino gown (kept, not a cited locus)", "authority": [], "quote": null}, "D-aim": {"output": "casino gown -> evening gown", "authority": ["T5_brief_school", "T3_readability"], "quote": "Casino gown risks a Las Vegas reading for the contemporary anglophone reader"}},
      {"item": "Recht_to_law", "D-let": {"output": "law (pun-loss noted)", "authority": ["T2_source_text"], "quote": "Waldemar's interior thoughts (the ... law-and-Roman-law line)"}, "D-aim": {"output": "law", "authority": ["T5_brief_school"], "quote": "The domesticating choice is to drop the (untranslatable) German wordplay on Recht"}}
    ],
    "signature_summary": "Permission (D-let) = background license, rarely/never named, overridable by source-text and period at item level. School (D-aim) = foreground imperative, named at nearly every applicable choice and in every pass header, overrides source-text facts. Same content, opposite citation-behavior."
  },
  "cross_arm_patterns": {
    "no_persona_vs_persona": "C, C-let ground authority in T2+T3+T6+T9 (no T1/T8 available); persona arms add T1 and (A, D-let, D-aim) T8. B names 'persona' exactly once.",
    "C-let_cites_then_declines": {"cites_permission": true, "frequency": "once, global (pass 1)", "declines_named_licenses": ["rhyme/meter for lyric (kept Schumann unmetered)", "anglicize foreign vocab (kept Privatdozent, Fraulein/Frau/Herr, Molle, Frischer Hammel, casino gown)"], "override_authorities": ["T2_source_text", "T9_period"]},
    "A_craft_defaults_vs_B_literary_stance": {"A": "Domestication of society-words (persona defaults): Coupe -> carriage", "B": "I will not turn Berlinisch into Cockney... that is the translator's vanity and it falsifies"},
    "voice_models_corpus_gated": {"present_in": ["A", "D-let", "D-aim"], "absent_in": ["B (German-only corpus, zero Wharton/Mitford/Powell/Isherwood in persona or log)", "C", "C-let"]},
    "convergent_fidelity_corrections": [
      {"error": "selber lieben -> 'loves oneself' (narcissism misread)", "caught_by": ["C-let (pass 2)", "D-let (pass 2)", "D-aim (pass 4)"], "authority": ["T2_source_text", "T7_fidelity_correction"]},
      {"error": "vergessen = forget not forgive", "caught_by": ["B (pass 2)", "C-let (pass 4)"], "ratified_as_correct_by": ["C (pass 3): 'faithful to the accusative object'"], "note": "same locus, divergent verdicts across arms"},
      {"error": "Maetresse/Geliebte collapse", "caught_by": ["C (correct in pass 1)", "C-let (fixed pass 4)"], "authority": ["T2_source_text", "T7_fidelity_correction"]},
      {"error": "Recht = law not 'right'", "handled_by": "all six", "authority_cited_varies": {"B": "T7 fidelity (jurisprudential point lost)", "D-aim": "T5 school (drop untranslatable wordplay)", "A_C_Clet_Dlet": "T2 source-text pun-loss"}}
    ],
    "permission_vs_school_citation_modality": "Permission cited rarely/silently and overridable; school cited constantly and overriding. Same five licensed moves in content; opposite citation-behavior by modality."
  },
  "absent_arm_H": "The human translation has no pass log and therefore no documented reasoning; it is excluded from coding. Its absence is noted."
}
```