Repository path: workshop/translations/maqamat-sanaa/loci-frozen.md · rendered 2026-09-09

The sound-figure loci of «المقامة الصنعانية» — enumerated from the Arabic, frozen first

Why this page is committed before anything else. ARM-persian-hands step 2 asks whether the record established on four English Gulistans — not one full rhyme in 128 chances at a Persian prose rhyme — holds on a second author in a second language. If the places that count as Arabic sound figures were chosen after the English was read, the answer is guaranteed: whatever the hands answered would become the loci. So the loci are enumerated from the Arabic alone, classified, and committed before the lead has written a word of English and before any published hand is opened beyond the priming declared below. Data: loci_saj.json, loci_verse.json.

The criterion, the STEM/AFFIX split, and the outcome codes are imported from workshop/translations/gulistan-bab1/loci-frozen.md so that the two censuses are commensurable. Only what the new material forces is re-derived, and each such change is named in §5.

1. The work and the copy-text

«المقامة الصنعانية» — the first of al-Ḥarīrī of Basra's fifty Maqāmāt (c. 1110), the work that fixed rhymed prose (سجع, saj') as the show-piece genre of Arabic literary art. 590 Arabic tokens: 139 prose cola in three blocks and 9 verse lines in two pieces.

Copy-text, provenance and its single-witness limit: collation.md.

Segmentation is the copy-text's own. This text is punctuated at every colon boundary — a full stop, question mark or exclamation at the end of each saj' member — so the 139 cola are read off the page rather than judged. That is a property of this transmission and is recorded as such.

2. PROSE-SAJ — the rhymed-prose locus (66; primary pool 64)

A saj' locus is two or more rhyme-bearing words in parallel syntactic position in consecutive cola of one Arabic period, sharing a rhyme, in a block typed PROSE — the Gulistan page's criterion verbatim.

Pool: 66 loci over 137 of the 139 cola — 53 STEM, 11 AFFIX, 2 JINAS-TAMM.

The two cola that belong to no locus are «ثمّ أنْشَدَ:» and «أنْشَد:» — "then he recited". That is the whole of the unrhymed prose in this maqāma: six words. It is the single most important fact about these materials and it is stated here, before any English is read, because it decides §4.

• STEM (53) — the rhyme reaches into the root consonants and is a fact about those two words: «الوِفاضِ» / «الإنْفاضِ», «القَيْظِ» / «الغيْظِ», «ندَمُكَ» / «قدَمُكَ».
• AFFIX (11) — the shared rime is a class-general ending that costs the author nothing and gives English nothing to inherit: «الألْوانِ» / «الأدْيانِ» (the plural -ān), «أتيتُ» / «رأيْتُ» (the 1sg perfect -aytu), «نعْلَيْهِ» / «رِجلَيْهِ» (the dual -ayhi). Registered as a secondary, S1, on the Gulistan's reasoning and to test its result there (STEM 0.133 against AFFIX 0.059).
• JINAS-TAMM (2) — S39 «تَسْتَهْدِيهِ» / «تَستَهْديهِ» (seek guidance of / seek a gift of: the same word twice, in two senses) and S42 «الصَّدُقاتِ» / «الصَّدَقاتِ» (bride-prices / alms: one consonantal skeleton, two vowellings). These are puns, not rhymes, and tools/rhyme_pairs.py returns NONE for identical bearers by rule. They are excluded from every primary rate and reported separately in prose, a decision fixed here so it cannot be made after the answers are seen. Primary pool: 64.

3. VERSE — the line (9)

Two mono-rhymed pieces: V-A, b001–b003, rhyme -ābah (3 lines) and V-B, b004–b009, rhyme -īṣah (6 lines, opening line internally rhymed, muṣarraʿ). Every line carries the piece's rhyme, so the source side is constant and anything that varies is the hand — the same property the Gulistan verse pool had.

ANSWERED for a verse line = the English line-end rendering it chimes (STRICT or NEAR) with the English line-end rendering an adjacent line of the same piece. Mono-rhyme depth — how many consecutive English line-ends share one rhyme, against the Arabic's 3 and 6 — is a secondary, recorded per piece; RS-20260822c §6 found no Gulistan hand attempts the source's chain.

4. The control: why it is a chance rate and not a matched passage

The Gulistan census rested on 16 CONTROL-PLAIN stretches — unrhymed narrative by the same author on the same page — to catch a hand whose "answers" are merely its own habit of chiming or versifying. This work does not contain that control and cannot be made to. §2: the unrhymed prose of «المقامة الصنعانية» is six words long. Note (bqr) — at a matched shape there is often no plain version to compare against — fires here for a third time, on a third language, and this is the clearest case yet: the author's plain style is not scarce, it is absent.

So the control is declared, in advance, to be a chance rate, computed exactly as RS-20260822c §5 computed it after the fact:

CHANCE — every adjacent pair of clause-final words across the whole of a hand's rendering of this maqāma, graded by tools/rhyme_pairs.py. The fraction graded STRICT or NEAR is what that hand's English produces without trying, and every locus rate is read against it.

A hand whose locus rate does not exceed its own CHANCE rate has not been shown to carry anything. This is weaker than a matched control and is labelled as such wherever it is used.

4a. Amendment, 2026-08-24, made while translating and before any English was opened

S67 was missed by the enumeration and is added. The loci above were read off the copy-text's period punctuation. The opening period marks two of its members with commas instead — «لما اقتعدت غارب الاغتراب، وأنأتني المتربة عن الأتراب،» — and that pair was therefore not enumerated. The lead found it on the first sentence of the translation limb, with no published hand open. It is added as S67, flagged POST-FREEZE-ADDED and INTRACOLON, and every primary figure is reported both with and without it. The pool becomes 67 loci, primary 65.

No other comma-marked pair occurs in the maqāma; the copy-text's periods carry the rest.

5. What is new here, and why

1. JINAS-TAMM exclusion (§2) — the Gulistan span contained no total pun at a locus. This one contains two, and the instrument cannot grade them.
2. INTRACOLON — S24 («ممْلوكِكَ» / «مَليكِكَ») is one colon in this copy-text carrying both bearers, where every other locus spans two or more. Flagged, kept, and reported separately if it behaves differently.
3. The chance-rate control replaces the matched control (§4), under protest, with the reason written above.
4. COLA-KEPT — a code the Gulistan census did not have. RS-20260814 §1 found that Lane and Burton lose the saj' but not a single colon: what falls out of English is the sound, not the structure. That was two hands, eight loci, one work, and coded by the lead's reading. Here it is registered in advance over 64 loci and three hands: COLA-KEPT iff the English renders the locus as two or more distinct parallel members, in the same order, neither collapsed into the other. It is the study's one positive prediction.

6. The outcome codes, defined before any English is read

For each unit and each hand:

• FORM — VERSE or PROSE, read off the page image, not the OCR (the Gulistan round-1 critic's finding 2): indented, line-broken, capitalised at the line-head. OMITTED where the hand does not render the unit.
• CHIME — graded by tools/rhyme_pairs.py between the English words rendering the Arabic rhyme-bearers (PROSE-SAJ) or between the printed line-ends (VERSE). The per-bearer alignment — which English word renders which Arabic bearer — is committed to raw/census.json before the grading script is run (the Gulistan round-2 and round-3 critics' BLOCKING findings). The coder's own reading is logged beside the tool's and where they disagree both stand.
• COLA-KEPT — §5.4.

Derived, defined here so they cannot be defined to taste afterwards:

• ANSWERED = CHIME is STRICT or NEAR. STRICT and NEAR are never summed in a headline; the Gulistan's whole positive column was NEAR.
• GENRE-CHANGED = the Arabic unit is PROSE-SAJ and the English FORM is VERSE.

7. Declared priming, made here rather than discovered later

While establishing that three hands are freely reachable at all, the lead ran keyword greps over the three OCR files and saw four lines of English from Assembly I:

Nothing of Chappelow 1767 was seen. Its OCR was sampled at Assembly V, a hundred pages away.

Consequences, fixed in advance and exactly as the Gulistan page fixed them:

1. The published hands' figures are unaffected — what Chenery wrote in 1867 does not change because the lead read one line of it in 2026. PRIMED loci stay in the primary.
2. The primary is reported both with and without S01, S02, S57, and where the two disagree the smaller effect governs.
3. The lead's own rendering carries the priming as contamination, measured on the artifact.