Repository path: workshop/translations/hafez-shahed/source.md · rendered 2026-09-09

«حافظ» غزل ۱۵ — the Persian of record

Witness: Ganjoor, Qazvini–Ghani text of the Divan of Hafez, https://api.ganjoor.net/api/ganjoor/poem?url=/hafez/ghazal/sh15, fetched 2026-09-03 (S242); the raw response is at runs/RS-20260903b-line-end-order/raw/ganjoor_sh15.json. Public domain (Hafez, d. c. 1390).

Why this poem. It is the no-radif counterpart of S237's sh77. That rendering (T-hafez-daasht-R58-v1) had to put the same finite transitive verb last at all nine rhyming positions; this one has nothing that must stand last anywhere. The pair is what lets the same hand's line ends be counted under an obligation and without one.

Selection, mechanical, and one criterion declared late

The script is runs/RS-20260903b-line-end-order/select.py; its output is selection.json. From the frozen 495-ghazal census (runs/RS-20260829-radif-hands/census.json), restricted to the 173 Payne↔Ganjoor matches accepted at S236 (runs/RS-20260831b-radif-hands/match_accepted.json), take the lowest-numbered ghazal satisfying all of:

1. the census records no radif (radif == "") — the mirror of S237's rule 1;
2. 8 ≤ bayts ≤ 10 — S237's rule 4, unchanged;
3. the slug is not one of the ten the project has already rendered (sh3, sh4, sh77, sh115, sh118, sh124, sh130, sh131, sh176, sh187);
4. no hemistich of the poem is wholly in Arabic.

Twenty-one ghazals qualify on 1–3. Criteria 1–3 were fixed before anything was fetched, and running them alone returns sh13, which was fetched and read first. sh13 turned out to be macaronic: three of its sixteen hemistichs are wholly Arabic — «الصَبوح الصَبوح یا اصحاب»، «المُدام المُدام یا احباب»، «اِفتَتِح یا مُفَتِّح الاَبواب». Criterion 4 was written after seeing that, and this page says so rather than presenting a clean rule. The reason it is a criterion and not a preference: a wholly-Arabic hemistich is not a line whose English word order the translator chooses — it is a line whose language the translator chooses, which is the decision S240 and S241 measured and a different one from this. Three of nine rhyming positions of that kind would have removed a third of the poem from the measurement this limb exists to make. Criterion 4 is a hand coding of the Ganjoor text, recorded per candidate in select.py, applied in slug order; the loop stops at the first survivor, which is sh15 — 10 bayts, 11 rhyming positions, Payne's ode 90.

Payne's page was not opened before the rendering and its log were frozen and committed.

The text

Radif: none. Nothing repeats at the rhyming positions.

Qāfiya: the -ābat family — نقابت · آبت · خوابت · ثوابت · شرابت · صوابت · جَنابت · سرابت · شبابت · خرابت · عِتابت. Eleven Arabic-derived nouns in ‑āb, each carrying the enclitic ـت. At ten of the eleven the enclitic is the possessive your (your veil, your water, your sleep, your reward, your wine, your judgement, your threshold, your youth, your ruin, your reproach); at 7b it is the object pronoun you (به سرابت نفریبد — let it not deceive you with a mirage), which is the older attachment and is worth naming because it is the one place a translator can misread the rhyme word as a possessive.

What this means for a translator, recorded before a line was written. The rhyme is rich: the qāfiya proper is ‑āb, and آب (water) is literally inside نقاب (veil) at the maṭlaʿ. No English family reaches veil · water · sleep · reward · wine · judgement · threshold · mirage · youth · ruin · reproach, so the chime was expected to be refused at most positions and the enumeration is in the log. Nothing obliges the English line end here — which is the whole reason the poem was selected.