Repository path: workshop/translations/maqamat-hulwan/R43-v1/translation.md · rendered 2026-09-09

«المقامة الحلوانية» — al-Ḥarīrī's second Assembly, whole, R43 v1: restraint

Source. أبو محمد القاسم بن علي الحريري (Basra, 1054–1122), «المقامات», المقامة الثانية، الحلوانية. Copy-text, provenance and its single-witness declaration: ../collation.md. Loci enumerated from the Arabic by script and frozen 2026-08-27, before a word of this English existed: ../loci-frozen.md — 58 loci over 134 of the 140 prose cola.

Span. The whole maqāma — 140 prose cola and 17 verse lines, 709 Arabic tokens → 1,157 English words of prose.

Regime. R43 the prose holds — ../../regimes/R43-prose-holds.md, unchanged. Prose stays prose; the sense outranks the sound; at every frozen saj' locus the hand enumerates the ordinary English renderings of both rhyme-bearers and takes a chime only if one is already lying among them.

Why this artifact exists. ARM-worth-paying step 1 put three English policies on al-Ḥarīrī's first Assembly to blind readers and found that the ranking of the three inverts on one disclosed fact about the source — but the statistic that carried the page was invented after its gate failed, and its own §8 limit 1 says the first thing owed is that statistic registered in advance on material the hypotheses were not formed on. This maqāma is that material. R50 already existed on it; R43 and R48 did not. This is the restraint arm.

Translated by the lead agent as a labelled subject, in session, at $0 — not ledgered (charter §3, A4). No API call produced any part of this rendering. The lead does not judge it.

The verse is carried from T-maqamat-hulwan-R50-v1 unchanged, declared here rather than discovered later, exactly as T-maqamat-sanaa-R48-v1 carried Assembly I's verse from R43. All three regimes rhyme the verse and all three graded it; re-rendering seventeen lines to no purpose would have made the arms differ where the regimes do not. Every study segment built on this rendering excludes the verse.

Contamination: none against every published hand, and **measured DEPENDENT? against the

project's own earlier rendering of the same text** — which is the number on this page worth reading

No published English hand of this maqāma was opened by the hand that wrote this rendering. Preston 1850 and Chenery 1867 were sliced to files by a script in an earlier session and were passed to tools/dependence_check.py as paths; only its statistics were read.

clean against both published hands, below their overlap with each other, and contamination: none on the front matter, which is a statement about published hands and is what the field means.

The fourth row is the one that matters and it is not a formality. ORD and BAL are two different declared policies — restraint, and Preston's balanced period — executed by one hand on one text. On Assembly I, where BAL was written last, the same two arms share 19 twelve-grams and a 15-token run. Here, where BAL was written first and was open on disk while this rendering was made, they share 216 twelve-grams and a 48-token run: "and the glittering of his show, I looked hard into his features, and let my eye run over his marks, and there he was — our shaykh of Sarūj, whose pitch-black night had turned to moonlight, I gave myself joy of his coming, and hurried to take". Eleven times the twelve-gram overlap, from reversing the order of composition and nothing else. Method note (bhb) — the lead matches itself above its record against any published human hand — has never fired this hard, and it fires here on two renderings the regimes were written to make different.

The consequence is a design consequence and it is taken, not argued away: a forced preference between ORD and BAL on this text would be a near-duplicate control wearing the clothes of an experiment, and E-20260827b's design drops that pair by this measurement rather than by taste. ORD ~ RHY is high for the opposite and intended reason — R48 is R43 with a chime delivered at the loci and is supposed to be identical between them — and the design handles it by requiring every study segment to contain loci where the arms in fact differ.

The rendering

The prose, one record per colon and beside the other two arms, is ../cola.tsv, frozen with this page. The verse is in ../R50-v1/translation.md and is identical here.

One locus, the three arms side by side — S026, p064/p065, «فسلّمَ على الجُلاّسِ. وجلَسَ في أُخرَياتِ الناسِ»:

Compliance with the regime, measured

Every consecutive colon-end pair in all three arms was graded by tools/rhyme_pairs.py — ../build_cola.py, run before any study using these texts was designed. A locus counts as chiming if any pair inside it grades above NONE.

The two STRICT colon-end chimes in this arm are S011 recitation ⁄ penetration and S044 examination ⁄ estimation, and both are -ation suffixes — that is, both are the accident this project has now recorded on four works. Assembly I's R43 came in at exactly 2 STRICT as well, which is either a fact about English or a coincidence at n = 2 and is reported as the latter.

Translator's log — frozen 2026-08-27, before any evaluation of this rendering was designed

Decisions and alternatives. Not quality claims.

1. The rule of enumeration, as it was actually applied, and it is Assembly I's rule verbatim. R43 rule 3 says: write the ordinary English renderings of each rhyme-bearer the hand would accept on the sense alone, then look for a pair among them. The bearer is a word, not a colon. A pair obtainable only by inverting the clause so that some other word fell last, or by importing a word the Arabic has not got, was not counted — that latitude is R48 rule 3's and is exactly what separates the two arms. Where a chime was found among the ordinary renderings it was taken even when it does not land at the colon end, and the table below says which.

2. The yield is far lower than on Assembly I, and the reason is grammatical, not moral. Assembly I's R43 found a chime at 33 of 65 loci; this one finds 10 of 58. Assembly I is a sermon: its rhyme-bearers are second-person verbal nouns — your rebelling, your preening, your hoarding — and English has a productive -ing that turns any verb into one, so the pairs fall out of the morphology by the dozen. Assembly II is narrative, and its bearers are concrete nouns — amulets, turbans, a lair, a meaty bone, a bushy beard. English has no morphology that makes two unrelated concrete nouns chime. The reachability of a prose chime under restraint is a property of the source passage's grammar, not of the translator's diligence, and two executions of one regime on one author two hundred cola apart differ by a factor of three because of it.

3. The chimes that were taken, all ten.

Two further chimes exist in this rendering at the bearer level and do not reach the colon end, and they are reported because R48 rule 4's whole point is that a reader hears the end: S025 shaggy ⁄ shabby (English puts the noun last: a shaggy beard ⁄ a shabby look) and S053 blazing ⁄ glittering. The restraint arm's chimes are therefore, at two loci out of twelve, inaudible by grammar — nobody chose to hide them.

4. The three places the sense was genuinely uncertain. These are inherited from R50-v1's log §3, which settled them on this same Arabic, and are not re-litigated: p025 «ومُدراةٍ ودِرايَةٍ» (the comb-and-discernment jinās, rendered for sense with none of the play); p047 «عدَمُ العُراقِ» (a bone with meat on it, punning against العِراق Iraq in the next colon, kept literal because the literal image is what sets up a pun that cannot be carried); p131 «وقدْ أقْمَرَ ليلُه الدّجُوجيُّ» (his pitch-black night turned moonlit, i.e. his hair went white, kept as the image). One is settled differently here: R50 rebuilt p019 to end on prince because Sāsān and Ghassān chime and its rule 3 forbids a chime. R43 has no such rule, so the names stay where the Arabic puts them and the chime is inherited rather than made — the one place in this maqāma where a translator gets al-Ḥarīrī's sound for nothing, because the sound is carried by two names that must be transliterated whatever the policy.

5. What was given up, listed. The saj' at 48 of 58 loci. The jinās at p025 and p047. The Qur'anic tag at p097 («إنّ بعْضَ الظنّ إثْمٌ», sura 49:12), rendered flat. The intra-colon rhyme at p068 (yalīh ⁄ fīh), which ../loci-frozen.md §3 names as the machine rule's one visible blind spot and which no arm carries. Nothing was added: no fact, event, referent or evaluation the Arabic does not have.