Repository path: workshop/translations/alf-layla/collation.md · rendered 2026-09-09

«ألف ليلة وليلة» — copy-text, recension, and the collation of span A

The project's nineteenth source language and its first Semitic one. This page fixes what is being translated, from what, and with what known uncertainty. It is written before span A was rendered and is the page a later span amends.

1. The work

«ألف ليلة وليلة» (Alf layla wa-layla, The Thousand Nights and a Night), anonymous. A frame narrative of unsettled date and authorship: an Arabic redaction, drawing on a lost Persian Hazār afsān, attested in a ninth-century fragment and in Ibn al-Nadīm's tenth-century Fihrist, and reaching its received full-length forms in Egypt in the later Middle Ages. There is no author, no autograph and no critical edition of the whole; there is a family of recensions, and which one a translator used is a fact about the translation, not a detail.

This matters here more than for any predecessor long work. Verga, Canth, Mikszáth and Tagore each had a single authorial text. The Nights has at least four printed recensions in circulation by 1885, and the two published English hands this arm will read against — Lane and Burton — did not work from the same Arabic.

2. Copy-text adopted

«ألف ليلة وليلة», مؤسسة هنداوي (Hindawi Foundation), Windsor, UK, 2022, vol. 1 of 6, ISBN 978-1-5273-2492-3 — read through the Arabic Wikisource ProofreadPage transcription of that edition, at فهرس:ألف ليلة وليلة (الجزء الأول).pdf, proofread status C, with page images for every page. Span A is PDF pages 11–12 (printed pp. 11–12), transcluded on the Wikisource page ألف ليلة وليلة (مؤسسة هنداوي، 2022)/الجزء الأول/حكايات الملك شهريار وأخيه الملك شاه زمان.

• The underlying work is public domain. Hindawi's contribution is an editorial layer — typesetting, punctuation, partial vocalisation, and marked corrections of its own base text (the transcription preserves the editor's {{خطأ|ففقال|فقال}} correction inside span A, which is why the layer is visible at all). The Foundation publishes freely readable, freely downloadable texts.
• Recension: the reading بجزائر الهند والصين ("in the islands of India and China") at §1 places this edition in the Būlāq/Calcutta II family rather than the بمغرب الهند والسند ("in the marches of India and Sind") family — see §3 below. No independent collation against a Būlāq or Calcutta II scan was performed this session; the second witness used is a second digitisation, not a scan. That collation is owed and is named in ARM-alf-layla as a step.

Amended 2026-08-14, after the comparators were opened (RS-20260814-saj-carriage §6a). The copy-text is materially shorter than the Arabic behind both published hands at one place in span A. Where Hindawi 2022 has eleven words — فأمر وزيره أن يسافر إليه ويحضر به، فأجابه بالسمع والطاعة، وسافَرَ حتى وصل بالسلامة — Lane's Būlāq and Burton's Calcutta II both carry a full episode: a written invitation quoted in the text, gifts of horses with gem-encrusted saddles, Mamelukes, handmaids and stuffs, a three-day guest rite, and the vizier's journey. Lane abridged and Burton expanded, but neither hand can account for material present in both of them and absent from the copy-text. This is a recension difference and the collation against a Būlāq scan is now not merely owed but load-bearing; until it is done, span A's saj' locus 8 is coded absent in both hands and that verdict stays ambiguous between the translator dropped it and it was not in his Arabic.

3. Collation of span A against a second witness

Witness 2: the older, unrelated Arabic Wikisource transcription at ألف ليلة وليلة/الجزء الأول (no page images, no stated source edition). Compared mechanically over span A after stripping vocalisation and normalising أ إ آ → ا, ى → ي, ة → ه: 640 copy-text tokens against 801 witness tokens over the same stretch, 28 non-identical spans. The substantive ones:

Conclusion. The copy-text is the sounder witness on every point where they differ in substance: witness 2 loses the preamble, transposes a block, thins two formulae and censors one word. Adopted without emendation. Row 6 is recorded as a finding, not merely as a variant — it is the expurgation question appearing on the source side before any translator is consulted.

4. Span A

PDF/printed pp. 11–12: the basmala and doxology, the naming of the two kings, Shah Zaman's discovery and killing of his wife, the garden scene seen from the window, and Shah Zaman's account of the first half of it to his brother. It ends where the printed page ends, at وأما ردُّ لوني فاعفُ عني من أن أذكره لك — a natural break: the younger brother has told half his story and withheld the other half, which page 13 then extracts from him.

642 Arabic words (stored at source-hindawi-2022-spanA.txt, page-marked [[p11]], [[p12]]). Comparable in size to the spans of the three predecessor long works (Mikszáth ch. 1, 650 words; «Flipperne», 756).

5. Comparators

Both public domain, both freely readable, both read only after span A and its translator's log were frozen (R05, source-only):

• Edward William Lane, The Thousand and One Nights, 1839–41 — Project Gutenberg #34206 (vol. I). Lane worked from Būlāq.
• Richard F. Burton, The Book of the Thousand Nights and a Night, 1885 — Project Gutenberg #3435 (vol. 1). Burton worked mainly from Macnaghten (Calcutta II), with Būlāq consulted.

John Payne's 1882–84 version was sought and not found freely — it is not on Project Gutenberg (searched 2026-08-14; the only Payne hit is a different work). Two hands, not three. Recorded here so a later span does not repeat the search blind.

6. Span B, and the collation of span B against the same second witness

Added 2026-08-14 at span B, before either comparator was opened for this span.

Span B is PDF/printed pp. 13–15: Shahriyar demands the rest of his brother's story and resolves to see it himself, sees it, the two kings abandon their kingdoms, and at a tree by the sea the ifrit comes out of the water with the abducted bride in a box. It ends at the foot of the maiden's speech, after the second and third quoted poems, at كَانَ مِنْ فِتْنَةِ النِّسَاءِ سَلِيمًا — a natural break: the next sentence is the brothers' reaction to what she has said. 582 Arabic words (stored at source-hindawi-2022-spanB.txt, page-marked [[p13]], [[p14]], [[p15]]), of which 490 are prose and 92 are verse — the work's first verse, 12 bayts in three poems, fully vocalised in the copy-text.

Method identical to §3 (vocalisation stripped, أ إ آ → ا, ى → ي, ة → ه, verse markup flattened, mechanical alignment). 582 copy-text tokens against 527 witness-2 tokens over the same stretch, 28 non-identical spans — the same count as span A, on a comparable length. The substantive ones:

Conclusion, and it is the same conclusion as span A. The copy-text is the sounder witness at every substantive point but one: witness 2 loses a turn of dialogue, an epithet, a negative particle, one bayt of one poem and the whole of another, and censors the same obscene verb a second time. Row 9 is recorded as the exception and is not adopted. Two spans, two obscene words, two silent expurgations by witness 2 — the count is now 2 of 2, and the expurgation question keeps appearing on the Arabic side before any translator is consulted.

The Būlāq/Calcutta II collation (§2, and ARM-alf-layla step 3) is still owed. Nothing in span B bears on it either way: the recension gap §2 records at span A does not recur here, and the two witnesses collated above are both digitisations, not scans.

7. Span C, and its collation against the same second witness

Added 2026-08-15 at span C, before either comparator was opened for this span.

Span C is the foot of PDF/printed p. 15 through p. 18, and it completes the frame tale: the brothers' reaction to the maiden, their return, the killing of the wife and the household, the three years of a virgin a night, the vizier's search, Shahrazad and Dunyazad, the vizier's fable of the ass and the ox with its inner tale of the dog and the cock, and Shahrazad's wedding night down to فرِحَ بسماع الحديث. 1,036 Arabic words — 1,022 prose and 14 verse, a single two-bayt poem — stored at source-hindawi-2022-spanC.txt, page-marked [[p15]]–[[p18]]. It is the largest span so far, and it is a span because the narrative unit is: p. 19 opens the first night under its own heading (فلما كانت الليلة ١) with حكاية التاجر مع العفريت.

Method identical to §3 and §6. 1,036 copy-text tokens against 1,022 witness-2 tokens over the same stretch, 44 non-identical spans — the same rate per thousand words as spans A and B. The substantive ones:

Conclusion, and it is no longer quite the same conclusion. The copy-text is still the sounder witness at most substantive points — it keeps the fable's moral, the merchant's age, the command-and-execution pair, the accusative — and witness 2 still thins the text more often than it fills it. But span C reverses the direction twice. Witness 2 is grammatically right at row 7, probably right at row 8, and at row 1 it carries a five-member rhymed epithet chain that the copy-text simply does not have. Rows 7 and 8 are followed in the English and recorded in D38; row 1 is not, and cannot be — there is nothing in the copy-text to translate there.

Row 1 matters beyond this page. Every earlier difference could be read as witness 2 losing material. This one cannot, and it is saj', which is the formal device this arm's study limbs are built on: a rhymed chain present in one Arabic witness and absent from another, at a place where both published English hands will have had some Arabic in front of them. It is the same shape as the §2 recension gap and it is the second instance.

Row 12 is worth one line. ضرط (farted) stands in witness 2, where نيك was silently deleted twice. The expurgation §3 and §6 recorded is specific to the sexual word, not general to coarseness — which is a sharper finding than "witness 2 is prudish", and it is the sort of thing only a third span could show.

8. The Būlāq / Calcutta II collation: searched for, located, and not achievable from freely

readable text

Added 2026-08-15. This is ARM-alf-layla step 3's second half, and the honest outcome is a null.

§2 recorded that the copy-text is materially shorter than the Arabic behind both published hands at one place in span A, and that a collation against a printed 19th-century recension was therefore "not merely owed but load-bearing". It was attempted this session. What is freely reachable does not support it, and the search is recorded here in full so that a later span does not repeat it blind — the same discipline §5 applies to Payne.

• Calcutta II (Macnaghten), 1839–42 — Burton's Arabic — exists on the Internet Archive, in the Bibliotheca Alexandrina collection: AAlexandrina-109613, -085165, -108334, -149995, -028092, -028625, all catalogued [1839-1842], publisher W. Thacker and Co, each with a page scan, a PDF, and OCR. None of the volumes available is volume 1. شهريار and ساسان occur zero times across all five OCR files examined (≈ 890,000 characters each), and the running text of the first begins in the two-hundred-and-twenties of the nights. The frame tale is not in the digitised set.
• The OCR would not support a word-level collation in any case. The Arabic optical recognition on these 19th-century founts is not merely noisy but largely non-lexical; a fifty-word sample of AAlexandrina-109613 contains no recoverable sentence.
• AAlexandrina-058223, catalogued [1833-1838], is not Būlāq: its title page reads بدار طباعة المدرسة العظمى الملكية … في مدينة برسلاو — the Royal Academy press at Breslau, i.e. the Habicht recension, a third family again, and the volume opens mid-Sindbad.
• No Būlāq scan was found by search of the Internet Archive on the edition's name in Arabic or in either common Latin transliteration.

Conclusion, stated as a limit and not as a finding. The §2 ambiguity stands exactly where RS-20260814 §6a left it: span A's saj' locus 8 is absent in both hands, and that verdict remains undecided between the translator dropped it and it was not in his Arabic. What this session establishes is only that the question cannot be closed from freely readable text — it needs page images of Calcutta II vol. 1 or Būlāq vol. 1 read by eye, which is a different kind of work from anything this arm has done, and it is not worth a span of translation to do it. Nothing in the two published hands is read differently because of this, and no figure on any result page depends on it. Both digitised Arabic witnesses this project can reach — the copy-text and witness 2 — lack the episode, which is consistent with its being a Būlāq/Calcutta II reading and is not evidence of it.

9. Span D, and the collation of span D against the same second witness

Added 2026-08-15 at span D, before either comparator was opened for this span.

Span D is PDF/printed pp. 19–22: the heading فلما كانت الليلة ١, the first embedded tale (حكاية التاجر مع العفريت) as far as the calf breaking its rope, the night-closing formula, and the frame's return to Shahriyar's palace. 762 Arabic words (stored at source-hindawi-2022-spanD.txt, page-marked [[p19]]–[[p22]]). Night 2 opens on p. 23 under its own heading, which fixes the span's end at a boundary the copy-text itself draws.

Method identical to §3 (vocalisation stripped, أ إ آ → ا, ى → ي, ة → ه, mechanical alignment over word tokens). 758 copy-text tokens against 572 witness-2 tokens over the same stretch, 18 non-identical spans — the fewest of the four spans, and also, by a long way, the largest single loss.

The remaining seven are orthographic (مثلما / مثل ما, دنا / فدنا, ونسائه / ونساءه, إذ / إذا, a lost لها, two inserted vocatives يا).

Conclusion. The copy-text is again the sounder witness, and this time not by a margin that has to be argued: witness 2 is missing a quarter of the night, and what it is missing is the scene the tale is built on. Row 1 is recorded as a fact about the digitisation rather than about the recension — both witnesses are digitisations of Egyptian-family printings, and a 190-word absence that leaves the surviving text self-contradicting is a transmission accident, not a variant reading. Nothing was translated from witness 2 and nothing in span D turns on it. Rows 2 and 4 are the interesting ones: both are losses of repetition — a third name for the ifrit, and a doubled verb pair — and both are exactly the kind of thing V14 and T10 were written to stop a translator doing. Here the source side does it first.

10. Span E, its blank page, and the first disagreement about where a night ends

Span E is pp. 23–28 of the copy-text — the second and third old men's tales, the copy-text's second and third nights, and the dismissal that ends the Tale of the Merchant and the Ifrit. 1,485 Arabic words (stored at source-hindawi-2022-spanE.txt, page-marked [[p23]]–[[p28]]). The span ends where the copy-text prints its heading for the next tale, the fisherman and the ifrit.

p. 26 is blank. The Wikisource transcription has no text for it and marks it unproofread, which is a gap of the kind that would normally force a span boundary. The page image was fetched and read: it is blank in the printed book — a blank verso, so that the third night's heading falls at the top of p. 27. There is no missing text and no lacuna in this span. (p. 30, inside the following span, is a different case: it carries text and has no transcription. That is the next span's problem.)

Method identical to §3. 1,472 copy-text tokens against 1,425 witness-2 tokens over the same stretch, 83 non-identical spans — the most of the five spans, and for a reason that makes the number misleading: 36 of the 83 are one thing said thirty-six times.

The systematic difference: the dual

The copy-text uses the Classical dual throughout the second old man's tale and witness 2 uses the plural. أخواي / إخوتي, طلبا / طلبوا, وأرادا / وأرادوا, كسبتما في سفركما / كسبتم من سفركم, فألحّا / فألحّوا, وهما يعرضان / وهم يعرضون, فغارا / فغاروا, أعمالهما فجاءاني / أعمالهم فجاؤوني, مراكبهما وأهلكهما / مراكبهم وأهلكهم, هذان أخواك / هؤلاء إخوتك, ولا يتخلصان / وما يتخلصون, and twenty-five more, including باتا / باتوا of the King and Shahrazad in the frame. The tale has exactly two brothers in it and the copy-text says so in its morphology at every mention; witness 2 says it nowhere.

This is a register difference, not a variant reading, and it is the sharpest instance so far of the thing §2 says about this work: which recension a translator worked from is a fact about the translation. It is also invisible in English, which has no dual — the two witnesses give word for word the same translation at all thirty-six sites. A difference that separates two Arabic texts completely and separates their English not at all.

The substantive differences

The remaining differences are orthographic or single particles (بنتي / ابنتي, مقبل / أقبل, بعدما / بعد ما, دنيازاد / دنيا زاد, أتممي / أتمي, inserted and dropped و and يا).

The disagreement about the night boundary, stated plainly

Rows 1 and 2 are one fact. Over this stretch:

• the copy-text marks two boundaries — a heading at p. 27 with no dawn formula, and a dawn formula on p. 28 with no heading;
• witness 2 marks one — an inline وفي الليلة الثالثة at the p. 28 place, with a dawn formula before it, and nothing at all at the p. 27 place.

Both witnesses reach الليلة الرابعة at the same sentence (the fisherman refusing to believe the ifrit was ever in the flask), so this is a misplaced boundary and not a divergence in the count. The likeliest reading is that the copy-text's editor set the third night's heading at the top of a fresh page — which is why p. 26 is blank — and that the heading migrated a page and a half back from the place its own dawn formula marks. That is a hypothesis and this page does not settle it; register.md open question 10 puts the test on the next span, which must record where the copy-text puts its headings and where it puts its formulae, separately.

Nothing was translated from witness 2. V15 follows the copy-text at both boundaries, defect included, and D56 says why.

11. Span F: a plate page, and what the copy-text's own typography ranks

Span F is pp. 28–33 — the tale of the fisherman and the ifrit from its heading to the seam into the fourth night. Two facts about the copy-text as a physical book came out of rendering it, and both bear on how it is being read.

11.1 p. 30 is a full-page colour plate, and that is why it has never been transcribed

ARM-alf-layla step 6 carried p. 30 forward as "not transcribed on Wikisource, page image legible at 500px", on the assumption that it was a transcription gap. It is not a gap: the page is a full-page colour illustration of the ifrit rising in smoke over the fisherman on the shore, with a one-line caption below it quoting the text (ثم انتفض فصار عفريتًا، رأسه في السحاب ورجلاه في التراب.) and five lines of body text under the caption. A Wikisource ProofreadPage volunteer transcribing prose has no reason to stop at such a page and every reason to skip it, which is the likeliest explanation for its being the only untranscribed page in this stretch.

This is the second time a "missing" page in this copy-text has turned out to be a fact about the book rather than about the transcription — p. 26 was blank in the printed volume (§10). Neither was a lacuna. The general point for anyone using this copy-text: check the page image before recording a gap.

Reading conditions, stated because they are weaker than for every other page of this work. The five lines were read from the Wikimedia thumbnail render at 500 px wide (937×1250 native). Larger renders were requested six times over the session at 900, 937, 1000, 1100, 1300 and 1500 px and the thumbnailer refused every one — only the already-cached 500 px render exists. The text was legible at that size and is stored in source-hindawi-2022-spanF.txt marked as read from the image. Two internal checks support it and neither is decisive: the reading joins p. 29's last words (لا تقتلني؛ فإني) into a grammatical sentence, and its own last words (تستحق على هذه البشارة) join p. 31's first words (يا قَيِّمَ العفاريت زوال الستر عنك) into a grammatical sentence. Short vowels and the finer diacritics on these five lines are not reliable at 500 px and no claim on this page or on any result page rests on them.

11.2 The night outranks the tale, and the copy-text says so in type

The copy-text gives a night boundary a page-opening display heading, in large type, on a fresh page with neither running head nor folio (pp. 23, 33 — and p. 19 for night 1). It gives a tale heading — حكاية الصياد مع العفريت — a bold line in the middle of a page (p. 28). The running heads say the same thing: recto pages carry the current night (فلما كانت الليلة ٢ on p. 25, فلما كانت الليلة ٣ on pp. 29 and 31), verso pages carry the book title.

In this edition the night is the primary division and the tale is subordinate to it — the reverse of the arrangement in which English readers meet this work, where the tales are stories and the nights are an ornament between them. D70 records the consequence for the translation: the English heading levels follow the copy-text's.

One defect in the running heads, recorded and not resolved. p. 35 is a recto inside the fourth night and its running head reads فلما كانت الليلة ٣. Either the printed book carries a stale running head there or the Wikisource transcriber copied the previous one; the page image was not consulted, because nothing in span F depends on it. Anyone extending this arm past p. 33 should check it on the image before using running heads as evidence of anything.

11.3 The collation of span F against witness 2, and what it does to open question 10

The older Arabic Wikisource transcription (ألف ليلة وليلة/الجزء الأول, witness 2) was rate-limited three times and reached on the fourth attempt. Compared mechanically over span F on the usual normalisation (vocalisation and tatweel stripped, أ إ آ → ا, ى → ي, ة → ه): 1,008 copy-text tokens against 1,005 aligned witness tokens, 954 identical, 42 non-identical spans, similarity 0.948 — the closest agreement of the six spans collated so far.

Two of the 42 are artefacts of the stored source file and are not witness differences, and they are named here so nobody counts them: the p. 30 plate caption is stored in source-hindawi-2022-spanF.txt and quotes a clause that also stands in the body on p. 29, so the aligner reads the caption as a ten-token witness-2 deletion; and the tale heading falls inside a witness-2 interpolation (below), so it aligns as a deletion too. Most of the remainder are orthographic.

The substantive differences, four of them:

Conclusion, as at every previous span: the copy-text is the sounder witness at every substantive point. Adopted without emendation.

11.4 Open question 10 is settled on two witnesses, not one

The register's open question 10 asked whether the copy-text's irregular night marking is an edition's slip or a property of the recension. D70 answered it from the copy-text alone — three boundaries marked three ways — and witness 2 now answers it independently and agrees.

At the night 3 → 4 seam, witness 2 prints the dawn formula, no frame passage at all, its heading, and then goes straight to قالت: بلغني with no Dunyazad exchange — the same thinning the copy-text has, on both sides of the seam. At the night 4 → 5 seam, twenty lines further on, witness 2 prints the dawn formula, the full frame (the sister's praise, the King's court day, the night with the vizier's daughter), its heading, and the resumption. Witness 2 varies its boundaries too, and it varies them at the same place.

That moves the finding one step: the seam thinning is not Hindawi's typesetting and not Hindawi's base text, because an unrelated digitisation of a different Arabic witness has it in the same place. It is in the tradition. V15's refusal to supply the missing frame is therefore carrying something the source has, not repairing something an editor broke.

One difference in how the two witnesses mark it, recorded because it is the kind of thing a translator should not smooth: the copy-text's heading formula is فلما كانت الليلة ٤, with the number in digits; witness 2's is وفي الليلة الرابعة, with the number in words. Two witnesses, two formulae, one boundary. Nothing was translated from witness 2.

12. Span G: a name the whole English tradition gets from another recension, and a broken sentence

that the second witness silently mends

pp. 33–36, the copy-text's fourth night. Compared against the same second witness by the same mechanical method as spans A–F — vocalisation and tatwīl stripped, أ إ آ → ا, ى → ي, ة → ه, punctuation removed: 1,359 copy-text tokens against 1,344 witness tokens, similarity 0.9515, 52 non-identical spans. The closest agreement of seven spans after span F's 0.948.

12.1 The sage's name, and why it is a recension fact rather than a variant

The copy-text calls him رويان at all ten of its occurrences in this span. Witness 2 calls him رويان at all eleven of its own. Neither has دوبان anywhere. Every English Nights there is calls him Doobán (Lane) or Douban (Burton), from the Būlāq and Calcutta II reading. Two unrelated Arabic digitisations agreeing against the entire English tradition is not a typo in one of them; it is the recension. D85 / T31 follows the copy-text and the translation therefore does not carry the name by which this tale is indexed in English. Registered as open question 11: one divergent name is a reading, and two would be a property worth stating here.

12.2 The substantive differences

Conclusion, as at every previous span: the copy-text is the sounder witness at every substantive point. Adopted without emendation. Row 3 is the sharpest instance so far of a general fact about this digitisation: witness 2's errors tend to produce readable text, which is exactly what makes it a poor control on the copy-text and a good demonstration of why the copy-text was chosen.

12.3 The night 4 → 5 seam, marked in full by both witnesses

D70's table of boundary types gains a fourth entry and it is not a fourth kind of thinning, which is what open question 10 said a later span should watch for. This seam has everything: the heading, the dawn formula, the sister's praise, Shahrazad's answer, the King's private resolve, the day in court, the night. Witness 2 marks it in full too, with its own heading formula (وفي الليلة الخامسة, the number in words, against the copy-text's digits — the same difference §11.4 recorded at the previous seam). The complete form is therefore attested on both witnesses, and the three thinned boundaries are thin against it.

13. Span H: two poems gone, a tale's ending gone, and every level-marker kept

Span H = Hindawi 2022 vol. 1 pp. 37–42, the copy-text's fifth night complete, 1,766 normalised tokens (1,769 raw). Witness 2's corresponding stretch runs from its heading '''وفي الليلة الخامسة''' to …دخل قصره واجتمع بأهله, 1,656 tokens. Compared by the same mechanical procedure as every earlier span (vocalisation stripped, أ إ آ → ا, ى → ي, ة → ه, markup removed): similarity 0.9281, 73 non-identical spans — the second-closest of eight spans after span F's 0.948.

13.1 The two large deletions

Deletion 1 is the largest verse loss the arm has recorded. The copy-text prints eight poems in this night and witness 2 six. Nothing in the surrounding prose is disturbed by their absence: the sequence وأيضًا في المعنى … وأيضًا في المعنى runs four times in the copy-text and twice in witness 2, and either reads.

Deletion 2 costs witness 2 a piece of the story. In the copy-text the King rides back to his first place, throws the flayed gazelle to the cook, sits down with the falcon on his hand, and the falcon dies. In witness 2 he rides back and sits down; the gazelle he cut the throat of and hung at his saddle-bow is never mentioned again.

13.2 The level-marking machinery: identical on both witnesses

This span's study limb (E-20260823b) is about how the fifth night's three levels of telling are marked. The source-side control matters, and it comes out clean:

Two unrelated Arabic digitisations agree about every in-text level-marker and disagree about a quarter of the verse. Whatever thins this tradition thins poems, not narrative machinery. The one vocative witness 2 drops is the least load-bearing one in the span — a vizier being addressed as vizier by the king already talking to him.

13.3 Where the copy-text varies and witness 2 regularises

Conclusion, as at every previous span: the copy-text is the sounder witness at every substantive point. Adopted without emendation. Row 5 is the first variant in eight spans that changes who or what performs the story's central marvel, and it is the kind of difference that would be invisible to a reader of either text alone.

13.4 A recension difference the collation cannot see, and the comparators can

Recorded here because it belongs with the copy-text's description. The tale King Yunan tells his vizier — the exemplum of the king who regretted a killing — is not the same tale in the four Arabic traditions this arm can reach:

And the copy-text elides the tale's central marvel. Both Arabic digitisations run …ثم افتح الكتاب، ففتحه الملك — then open the book. So the King opened it — with no beheading and no talking head between the sage's instructions and the King's opening of the book. Lane, Burton and Forster all print the execution and all print the severed head answering. This is a hole in the recension, not a translation difference; the English of span H carries it because V15 carries what is printed, and it is recorded as erratum E12 on ../R05-v1/translation.md because the span-H log's list of irregularities was written before it was found.

14. Span I: two witnesses that thin the same seam by different means, and a doubling that swaps ends

Span I is Hindawi 2022 vol. 1 pp. 43–45, the copy-text's sixth night whole, stored as source-hindawi-2022-spanI.txt (880 Arabic words). Collated against the same witness 2 (ألف ليلة وليلة/الجزء الأول) over the like-for-like stretch — from لو أبقيتني كنت أبقيتك to the second printing of the bayt — after stripping vocalisation and normalising أ إ آ ٱ → ا, ى → ي, ة → ه, ؤ → و, ئ → ي.

863 copy-text tokens against 845, similarity 0.9520, 32 non-identical spans. The substantive ones:

Conclusion, as at every previous span: the copy-text is the sounder witness at every substantive point — rows 4, 6, 7, 10 and 11 are all witness-2 errors or losses, and no row runs the other way. Adopted without emendation.

14.1 The night 6 → 7 seam, and a typographic rule that explains two blank pages

Five boundaries have now been rendered and the marking is various in both witnesses. The copy-text's night 6 → 7 seam is D70's third type — bare dawn formula, no frame passage, resumption with a heading and قالت: بلغني — recurring for the first time. Register open question 10 asked whether a fourth kind of thinning would appear; it has not. The thinning repeats, which makes it a habit of the book rather than a slip.

And witness 2 thins the same seam by dropping the other half of it. It keeps the full frame at night 5 → 6 — dawn formula, Dunyazad, how sweet is your telling, the court day, then ففي الليلة السادسة — and at night 6 → 7 prints only وفي الليلة السابعة. Span F found the two witnesses thinning the night 3 → 4 seam in the same place; this span finds them thinning the same seam in different ways. Both facts point one way: the thinning is in the tradition and the means of marking is not.

A typographic rule of the copy-text, now established and checkable. Every night heading in this volume opens an odd page — night 1 p. 19, night 2 p. 23, night 3 p. 27, night 4 p. 33, night 5 p. 37, night 6 p. 43, night 7 p. 47, seven for seven — and the two pages this arm has stopped to investigate as possible transcription gaps are the versos the rule leaves over: p. 26 before night 3's recto and p. 46 before night 7's recto. Both were confirmed blank from the page image (p. 26 at span E, p. 46 at span I, read at 700px). p. 30 is different and remains what span F found it to be: a full-page colour plate. After this span no future span needs to investigate a blank page in this volume; the rule predicts them.

It also explains D70's oddest row. Night 3's heading sits on p. 27 while the dawn formula that ends night 2 sits on p. 28: the heading takes the recto it is entitled to, and the tail of night 2 is what gets displaced.

15. Span J (pp. 47–51, the seventh night)

1,493 copy-text tokens against 1,449 in witness 2, similarity 0.9497, 63 non-identical spans — the second-closest of ten spans, after span F's 0.948 over a shorter stretch. Same method as §3: vocalisation stripped, أ إ آ → ا, ى → ي, ة → ه.

15.1 The two witnesses agree exactly about where the seventh night is

This is the load-bearing result of this collation, because the printed book contradicts itself about it (../R05-v1/translation.md, D136).

Two unrelated Arabic digitisations open and close this night at the same sentence, and neither has anything between. The running head on the copy-text's p. 49 reads فلما كانت الليلة ٩, and the running head on p. 51 reads فلما كانت الليلة ٧; a running head that advances two nights and then goes back one is not describing the text under it. The head on p. 49 is wrong, and the segmentation the English carries is the one both witnesses print.

What is not established: whether the wrong head is the printed book's or the Wikisource transcriber's. The page images were unreachable this session — upload.wikimedia.org answered every request with its robot-policy refusal (HTTP 403, then 429), with and without a descriptive user-agent — and the reading conditions are declared rather than assumed. Span F read a plate from a page image and span I read a blank verso the same way; this span could not.

The two witnesses mark the night by different means, which is itself the pattern this arm has been recording since D56: the copy-text uses a display heading with an Arabic-Indic numeral, and witness 2 an in-text bolded phrase with the ordinal spelled out. Neither supplies what the other has.

15.2 The two substantive differences

15.3 The small differences, and which way they run

Sixty-one further spans, none longer than three tokens. The copy-text is sounder at every one that decides anything:

• فرع من شجرة خضراء (a branch of a green tree) against witness 2's قرع (a gourd) — and the copy-text goes on to use بالفرع when the slave turns the pan over with it, so witness 2's reading is a typo that its own next sentence contradicts.
• وهو مبتل (masculine, of the slave) against وهي مبتلي (feminine, and unpointed) — witness 2 garbles the crux D148 records; it settles nothing and the copy-text at least parses.
• أخس البيضان (basest of the whites) against أخت البيضان (sister of the whites) — the copy-text's reading is abuse and fits the sentence; witness 2's is a letter's difference and is nonsense in the slave's mouth.
• أكره الخلق في صورته (I hate the whole creation in his face) against أكره النظر في صورته (I hate to look at his face) — the one place witness 2 is arguably the easier reading, and the copy-text's is the harder and stranger one. The English follows the copy-text; recorded as a genuine variant rather than as an error either way.

Two regularisations, both of which prove the copy-text's looseness is its own (V20's evidential pattern, third instance in the arm): the copy-text writes خرج **منها** عبد أسود with a feminine pronoun for a masculine wall, and witness 2 writes منه; the copy-text writes أحد شقيه and witness 2 supplies أحد شقي بابه. Where the two witnesses differ in tidiness, the copy-text is the untidy one and the English carries the untidiness.

And one variant the English could not carry either way: witness 2 prints the four fishes' colours أبيض وأحمر where the copy-text prints أحمر وأبيض. The copy-text's own two lists of the colours, at span I and here, are in different orders, and V16 requires the English to follow each as printed; witness 2's third order is recorded and not used.

15.4 dependence_check over span J, and the one run that has to be declared

Run after the span and its log were frozen (commit 307c1e5), against the stretch of each published hand that renders the same night: Burton 1885 vol. 1 and Lane 1839 vol. 1, both already stored in this repository from earlier spans.

The run is declared and not explained away. It is twelve tokens:

in his hand a branch of a green tree and he said

for وفي يده فرع من شجرة خضراء، وقال. Every word of the Arabic is a common word with one obvious English equivalent, and the phrase has no syntax to vary; two literal hands arrive at the same twelve words because there is nowhere else to go. That is an explanation, not a defence, and the row stands as DEPENDENT? on the record.

What the third row is for. Lane and Burton are two independent published hands, and on this same stretch they agree at 11 tokens with no 12-gram. The lead sits at 10 against Burton — below that baseline — and at 12 against Lane, one token above it. So the lead's agreement with Lane is at the top of, and marginally beyond, the range two independent hands reach here; it is not qualitatively outside it. The published-pair figure is reported precisely so that the lead's is not read against nothing (note (bcd): run length alone is a poor proxy, and the shared 7-gram counts — 19, 28, 22 — are the other half of the picture and put the lead's pair with Lane above both others).

Span J's contamination declaration is suspected, on the grounds written in the log, not on this measurement. The measurement is consistent with independence at the level two published hands reach; the declaration is about the lead's memory of Burton's rendering of this famous episode, which no n-gram count can see and which the log answers by naming the readings refused.