Repository path: workshop/translations/verter/plan.md · rendered 2026-09-09

«Вертер» — the long prose work chosen, and the selection record (W3 step 5)

Retired at the close-out (2026-09-09, S257). The project was wound up before span A was begun; no sentence of «Вертер» was translated. This page stays as the record of the selection. Correction: the claim below that this would be the project's first South Slavic work is wrong — Lazarević's «Први пут с оцем на јутрење» (T-jutrenje-R06-v1, S162) and Konstantinov's «Бај Ганьо в операта» (S168) preceded it.

What this is. wiki/plan.md §W3 step 5: choose the long prose work under the criteria Tom set 2026-09-04 (re-centring W3 on narrative prose after the thread closed) — public domain; written after about 1880; a kind of book rarely translated professionally (a minor novelist, a serial, a provincial press) rather than a canon monument; 8,000–15,000 words; the original and a free published English both reachable where possible; contamination measured on two candidate spans before choosing (CLAUDE.md §Contamination, note (bmw)). This page is that record and the declared span/cadence plan step 5 asks for. It is not itself a translation: nothing below is internal-judgment-only-exempt, and the first R05 span (step 6) is a later session's work.

1. The choice

Laza Lazarević (Лаза К. Лазаревић, 1851–1891), «Вертер» ("Verter" — the Serbian title transliterates Goethe's Werther), 1881. Serbian → English. 14,293 source words, 336 paragraphs, no chapter divisions.

• First publication: serialized in four parts in the magazine Otadžbina ("The Fatherland"), Belgrade, from mid-June to September 1881 — verified independently in two searches (the Serbian Wikipedia biography's account, and a separate search returning the same account in different words: "Lazarević published the first part of 'Verter' in mid-June 1881 in 'Otadžbina', and the remaining three parts were published by September"). A genuine newspaper serial, not a book.
• Source, verified by fetch: Serbian Wikisource, https://sr.wikisource.org/wiki/Вертер (canonical raw wikitext at ?title=Вертер&action=raw, byte-identical to the rendered page — checked by diff). Stored cleaned (template header and Напомене/category footer stripped, one paragraph per line) at source-srwikisource.txt, 336 paragraphs, 14,293 words — counted by Python str.split() and cross-checked with LC_ALL=C.UTF-8 wc -w; plain wc -w in this container's default locale returns 2,872 for the same file — a silent locale bug on non-Latin script, now recorded in CLAUDE.md.
• Lazarević is a founding figure of Serbian realist short prose domestically (his own country's histories credit him with the first Serbian psychological short story) but has, so far as this session could establish, no free public-domain English translation of any story — the only located rendering of any Lazarević fiction into English is a 2020s commercial anthology, Great Serbian Short Stories (Mikasinovich, ed.), still in copyright. This is the precise shape the criterion asks for: significant at home, essentially untranslated in English.
• Length: 14,293 words sits near the top of the declared 8,000–15,000 range, comparable in scale to Verga's «Jeli il pastore» (11,474) and larger than Mikszáth's Part I (8,064).
• Breadth: the project's first Serbian text, and its first South Slavic language of any kind. Prior long works were Italian (Verga), Finnish (Canth), Hungarian (Mikszáth); prior short pieces span Russian, Polish, Ukrainian and Czech among Slavic languages, but no South Slavic language before this.

2. Candidates considered, and the contamination measurement on two of them

Four candidates were located and fetch-verified (not assumed from search snippets) before choosing. All are public domain, post-1880, non-canonical narrative prose freely readable in full online.

The two candidate spans actually measured, per the standing rule (CLAUDE.md §Contamination, tools/dependence_check.py):

(a) «Popa Tanda» — a real comparator existed, so a real measurement was run. A middle-of-text Romanian passage was chosen that this session had not yet read in English (the opening and closing pages of the Byng translation had already been read while confirming the extraction, so using them would repeat the accidental-priming problem RS-20260727-longwork-selection documented for Verga — this time avoided by choosing an unseen locus instead of merely declaring the priming after the fact). The passage was translated fresh, frozen below, and only then was Byng's English of the same passage located and compared.

Lead, frozen before comparison: "Like most people, Father Trandafir never stopped to think about what he was doing. He was a priest, and he was content. He liked to sing, to read the gospel, to teach the Christians, to comfort and give spiritual help to those who had gone astray. He gave no more thought to it than that. Had he ever been asked whether he grasped the high sanctity, the hidden meaning of his calling, he might have laughed quietly at all the things a man understands only in hard times..."

dependence_check.py, two cells (inputs and raw JSON output kept at selection-probe/ beside this page — popa-tanda-lead.txt, popa-tanda-byng1921.txt, popa-tanda-null-control-nichifor.txt, dependence-check-result.json; not filed as an R05 translation, since this is a selection-gate probe on a work that was not adopted):

Reading: 12 tokens against a same-anthology null floor of 3 is a real, above-floor convergence — well below Verga's 15/floor-5 (suspected) or Turgenev's 21 (canonical, unavoidable), and it sits on a plain declarative sentence ("he was a priest and he was content") of a kind independent translators converge on by chance more than on marked prose. It would be declared contamination: suspected were Popa Tanda adopted. This measurement was not wasted by the rejection — it is the project's first contamination reading on a 1921 obscure-press Romanian anthology and is available if a later session returns to Slavici or to "Out in the World."

(b) «Вертер» — no comparator exists, so the standing alternative applies. CLAUDE.md calls a contamination: declaration without a measurement a placeholder; ARM-atelier-cycle (Canth, S085) establishes the honest alternative when no comparator is reachable on documented search routes: declare none, not measured, and record the routes rather than assume cleanliness. Four routes were checked and none returned a free English rendering of this story or of any Lazarević story: a Gutenberg-style bibliographic search, archive.org's full-text search API (Lazarevic Verter, 0 hits), general web search for "Lazarević" + "Verter"/"Werther" + Gutenberg/archive.org, and a search specifically for known Serbian-fiction-in-English anthologies (the only one found, Mikasinovich's, is a commercial 2020s edition, still in copyright). Contamination: none, not measured — no comparator reachable. This is the same position as Canth's, not a stronger or weaker one, and it does not license treating the work as an independence measurement for anything (CLAUDE.md §Contamination: obscurity is not protective, and here there is nothing to be protective of — there is no published rendering to be independent from or dependent on).

3. Why the missing comparator does not gate «Вертер» out

The assignment's own wording hedges only one of its five clauses: "the original and a free published English both reachable where possible." Word count, era, non-canonical status and public-domain status carry no such hedge. «Вертер» meets every unhedged criterion exactly — its 14,293 words sit inside the declared range with room on both sides, its four-part 1881 magazine serialization is about as literal an instance of "a serial" as this project is likely to find, and its author has never had a free English rendering, which is a stronger case for "rarely translated professionally" than Popa Tanda's single obscure 1921 anthology entry. Popa Tanda meets the hedged criterion and fails an unhedged one (6,663 words, 17% under the floor). Given the assignment's own emphasis — hedging the comparator and not the rest — «Вертер» is the better fit, and the comparator's absence is documented rather than glossed over.

Consequence for later steps. No later session may use this work as an independence measurement or a "two hands" comparison — there is no second hand to compare against, so the kind of finding ARM-longwork got from testing a marking policy against Dole, or ARM-legend got from testing one against Worswick, is not available here. What is available without any comparator — sustained voice across a long span, register drift, terminology consistency, translator fatigue — is the reason ARM-atelier-cycle (Canth) still counted as the deliverable's second long work despite having no English at all, and it is what this work is for.

4. Span plan (provisional — R05 §1 pins exact breaks at translation time)

No chapter markers exist in the source; it is 336 continuous paragraphs. The breaks below were chosen by reading the paragraphs near each target word-count and picking the nearest scene or time-shift ("On the third day after...", a fresh scene opening on a named event) rather than an arbitrary word cut, per R05 §1 — but as with Canth's eight-span declaration, these are a plan, not a freeze; a translating session may adjust a boundary by a few paragraphs where the log says why.

Six spans, sized 1,877–2,931 words — the same range Verga's five spans (1,889–2,806) and Mikszáth's Part I spans (2,389–2,811) held to. The cast (a Janko, a Marija, a Mladen, a Katanić, a Lieutenant Vasiljević, a postman Košutić all appear in the sampled paragraphs) is not mapped here — this session read only scattered paragraphs near each break to find a natural seam, not the story continuously, and a plot summary built on that sampling would risk asserting a relationship (e.g., which name is whose surname) the text does not actually support. Span A's translator reads the story whole before span A is drafted and corrects this table's break descriptions if a better seam is visible in context; nothing above is a claim about what the story is about, only about where the sampled paragraphs happen to fall quiet or change scene.

5. Cadence

At least one visit every four sessions. wiki/plan.md's rotation puts W3 up once every four sessions after step 4 closed (W1 → W3 → W1 → W2 → …), and ARM-legend's note (bjp) is explicit that a declared cadence should be the number the rotation can actually deliver, not an aspiration — a three-session cadence over a comparable rotation was measured unmeetable the day it was declared. Four matches what W3's own slot in the rotation supplies; under-visiting is the alarm, not a missed deadline.

Expected visits: 7 — spans A–F, then a closing craft report (step 7's first half; step 7's second half, a canon Japanese piece for the jury, is separate work).

6. Mechanics for span 1 (step 6, a later session)

R05-serial-long-work.md governs: draft from source-srwikisource.txt alone (no priming — this page's own §2(a) probe used a different candidate, Popa Tanda, precisely so that no part of «Вертер» itself is primed by anything read this session beyond the source text and the bibliographic facts above); self-revise; write the cumulative log D1…; open register.md with whatever terminology and voice questions span A raises; freeze before any evaluation is designed. Filed at workshop/translations/verter/R05-v1/translation.md, translated-by: lead, contamination: none (basis: §2(b) above).