Repository path: workshop/translations/saadi-ghazals-forced/R54-v1/translation.md · rendered 2026-09-09

Three Sa'di ghazals, each rendered whole twice — the two halves of the radif, one policy each

Source. سعدی, «غزلیات» ۹۷, ۹۴, ۱۲۶, whole. Copy-text, its single-witness limit, and the rhyming structure frozen from the Persian before any English existed: ../source-frozen.md. 21 bayts, 24 rhyming positions, all three at depth 8.

Regime. R54 (../../../regimes/R54-forced-pair.md). Two renderings of record per poem:

• REP — the repeated tail kept at every rhyming position, the chime given up. The word before the tail is not chosen for sound; where two renderings are equally defensible and one would chime, the non-chiming one is taken.
• CHI — the chime kept at every rhyming position to the source's own depth, the repeated tail given up. No two positions end in the same word; the radif's sense is carried inside the line.

Contamination: suspected. Declared before any English was written, on the standing ground that Sa'di is a much-translated poet. The measurement is ../dependence-note.md, run after this log was frozen and committed. Priming declared: the lead has not knowingly read an English rendering of any of these three ghazals, and no published hand was opened for any of them before the freeze.

Cost: $0. Lead translation, never ledgered (charter §3, A4).

1. Why all three are the forced case — the enumeration, written before any line

RS-20260826b-radif §2 established, 6 of 6 on six poems, that an English radif can only be something an English clause can end on, and that the qāfiya slot immediately before it inherits that constraint. All three poems here pass the first test and fail the second, and they fail it for two different reasons. The distinction is not in framework/v0.2 §7.36 and is the craft finding of this limb.

۹۷ — radif دوست, and ۱۲۶ — radif تو نیست: the genitive inverts

In both, the radif stands in a genitive relation to its own bearer, and Persian and English order that relation in opposite directions. Persian ezāfe and the Persian possessive both put the head noun first and the dependent second, so the bearer lands immediately before the radif: جمالِ دوست, نظیرِ تو. The English of-genitive puts the dependent last but inserts a preposition and an article: the beauty of the friend, an equal of yours. The word immediately before the English radif is therefore a function word — of, the, my — at every position, and a function word cannot be a rhyme-bearer. The English possessive genitive reverses the order instead (the friend's beauty), which puts the radif at the front of the phrase and destroys the shape outright.

Three further frames were enumerated for ۹۷ and all three refuse under R44's sense clause:

1. …[adjective] friend — an attributive adjective can be rhymed. But Sa'di's bearers are head nouns (beauty, union, star, room, sapling, image, perfection), not modifiers of friend. Turning each into an adjective replaces what each line is about.
2. …[bearer]'s own friend — changes the possession relation, and is nonsense at 3b, where مجال دوست is room for the friend, not the friend's room.
3. Radif = friend bare, article dropped — not English.

For ۱۲۶ the corresponding frames are …is not yours (which asserts something the Persian does not: «قمر چون رخ منیر تو نیست» says the moon is not like your shining face, not that the face is not yours) and …[bearer] you have not, which reverses the scope of the negation at 3b, 4b and 7b.

Verdict written in advance for both: AB UNREACHABLE, structurally.

۹۴ — radif اوست: the slot is free and the rhyme family is empty

۹۴ is different, and the difference matters. With the English radif …of his, the word immediately before it is the possessed noun itself — that evening-dark tress of his, that ruby lip of his. The slot is a content word and is rhymeable. So nothing structural blocks AB here.

What blocks it is that the eight bearers would then have to fall in one English rhyme family: evening-dark tress · pine-swaying stature · standing-up · ruby lip · message · snare-like tress · desire · slave. The best family reachable was ‑AIR (hair for زلف, snare for دام, air for a bearing) and it covers three of eight; قیام, لب, پیام, کام and غلام have no member in it that renders them. Every other family tried covered fewer. Verdict written in advance: AB UNREACHABLE, by exhaustion of the rhyme family, not by the grammar.

So §7.36's rule — the radif's part of speech decides — is necessary and not sufficient. All three radifs here are things an English clause can end on (a noun, a possessive pronoun, a negated possessive predicate), and all three fail anyway: two because the construction the radif stands in is head-initial in Persian and head-final in English, one because eight senses will not rhyme. §7.36 is extended, not corrected.

2. Writing order, recorded because R54 rule 5 requires it

Two of three had CHI first, not the balanced alternation rule 5 asks for. Three is an odd number and the imbalance is recorded rather than tidied: if second-drafting confers an advantage, it falls on REP twice and on CHI once. This is a declared limit and is carried into any design that uses these texts.

3. ۹۷ — radif دوست "the friend"

English radif in REP: the friend (the article is English's, not Sa'di's; the preposition before it varies with the sense, as the unmarked ezāfe does). Rhyme family in CHI: ‑AY.

۹۷ REP — the repeated tail kept

Blessed the morning whose first look falls on the beauty of the friend, and takes its fruit from the tree of hoping for union with the friend.

My luck had not been sleeping, since out of the morning's sleep I rose up under the happy-omened star of the friend.

Get out of my heart, grief of this world and the next: either the house is a place for your baggage, or room for the friend.

I mean to tear the root of the company of strangers out; in the garden of the heart I will let nothing stand but the sapling of the friend.

He gave me the honour and went. Being out of myself, I cannot tell whether that was, in my sight, the friend, or the image of the friend.

My wits did not stay, reason walked off, speech would not close; lucky the man who is rubbed out inside the perfection of the friend.

Sa'di, there is no veil. Keep your own mirror clean: how should a thing eaten with rust show the beauty of the friend?

The hand's reading. Radif held 8 of 8. Chime 0 of 8: the word before the tail is a preposition at every position (of ×6, with, for), and the content words that carry Sa'di's bearers — beauty · union · star · room · sapling · image · perfection · beauty — sit one to three words further back, where the Persian's chime does not fall and where none was sought. Sa'di's own ایطا is preserved: beauty at 1a and 7b renders جمال at 1a and 7b.

۹۷ CHI — the chime kept

A morning that looks on the friend's own beauty is a blessed day, and eats from the tree of hope that the friend will come and stay.

My luck had not been sleeping: out of the sleep of morning I got up into the friend's own fortunate ray.

Out of my heart, grief of this world and the next! Either this house is a room for your baggage, or the friend's: go away.

I mean to tear the company of strangers out by the root and leave in the garden of the heart no shoot but the friend's young spray.

He gave me the honour and went; and I was so far out of myself that whether it was the friend in my sight or his image, I cannot say.

My wits were gone, my reason walked off, my speech would not close. Fortunate the man rubbed out where the friend's perfection has its way.

Sa'di, there is no veil between. Keep your own mirror clean: how should a glass eaten with rust the friend's beauty display?

The hand's reading. Chime held 8 of 8 — day · stay · ray · away · spray · say · way · display, eight distinct bearers at eight positions, where Sa'di has seven distinct at eight. Radif 0 of 8 by construction. Two positions cost something and it is logged, not hidden: 2b fortunate ray renders طالعِ فرخنده‌فال, an ascendant star, by its ray — a metonym the Persian has not written; 6b has its way renders محو شود در کمالِ دوست, is effaced in the friend's perfection, and adds a faint sense of dominion that کمال has not got. Both stand as written; the hand does not revise to move a grade.

4. ۹۴ — radif اوست "is his"

English radif in REP: of his. Rhyme family in CHI: ‑ED.

۹۴ CHI — the chime kept (written first)

The sun is under the shade of a tress like evening on his head, and Tūbā of paradise goes slave to that pine-swaying tread.

Call it a stature? In truth it is a resurrection, for my own rising again is in his standing up, who was dead.

In this affair I am glad that the heart should die: the water of life is on his lip, and the lip is the ruby's own red.

Is it the breath of spring that reaches me, or the wind of morning? Is it the air of paradise going by, or a word he has said?

The heart was selling its airs — I am a clever bird — and look at it now, fallen in the end of that tress, where his snare is spread.

Helpless I have stayed, a whole day long, in his snare; and now I am fallen into an exile: it is there his wish has led.

Every moment the heart in my breast turns to blood with thinking: whose slave is he, whose slave Sa'di is? — and that is my dread.

The hand's reading. Chime held 8 of 8 — head · tread · dead · red · said · spread · led · dread, eight distinct at eight, matching Sa'di's own eight distinct. Radif 0 of 8. One position is logged as strained: 7b and that is my dread renders دل از اندیشه خون شود, the heart turns to blood from the thought, whose 7a already carries; dread names an emotion the Persian leaves as an image. It stands as written.

۹۴ REP — the repeated tail kept (written second)

The sun is under the shade of that evening-dark tress of his, and Tūbā of paradise is a slave to the pine-swaying stature of his.

Call it a stature? In truth it is a resurrection, for my own rising from the dead is in that standing-up of his.

In this affair I am glad that the heart should die, for the water of life is on the ruby-coloured lip of his.

Is it the breath of spring that reaches me, or the morning wind? Is it the wind of paradise going by, or a message of his?

The heart was selling its airs — I am a clever bird — and look at it now, fallen in the snare-like tress of his.

Helpless I have stayed, a whole day long, in his snare, and now I am fallen into an exile that is a wish of his.

Every moment the heart in my breast turns to blood with thinking whose slave he is himself, I who am a slave of his.

The hand's reading. Radif held 8 of 8. Chime 0 of 8: the word before the tail is of at every position. Here — and unlike ۹۷ and ۱۲۶ — the bearer noun sits one token further back and is a content word (tress · stature · standing-up · lip · message · tress · wish · slave), which is exactly the enumeration in §1: the slot was free and the rhyme family was empty. None of these eight chimes with another, and none was chosen for sound.

5. ۱۲۶ — radif تو نیست "is not your —"

English radif in REP: of yours. Rhyme family in CHI: ‑AIR.

۱۲۶ CHI — the chime kept (written first)

On earth there is nothing at all with you to compare, and the moon is not that lit face of yours, so fair.

I will not give my heart to the cypress for its height: it has not that stature of yours, that lovely air.

In the whole city, you with the bow for an eyebrow, I know of no one your arrow will spare.

Nobody else carries the hearts of people off: there is not a heart your own hands do not ensnare.

If you take up with another as good as me, what can I do, when for me there is no second of you anywhere?

It is plain that a heart of that iron kind is not fit for a bosom like silk to bear.

The whole world has gone off to its love-making; Sa'di's name is not in your mind — you are not aware.

The hand's reading. Chime held 8 of 8 — compare · fair · air · spare · ensnare · anywhere · bear · aware, eight distinct at eight, where Sa'di has seven distinct at eight (نظیر twice). Radif 0 of 8. Sa'di's ایطا is not preserved here and could not be: rule 4 forbids two positions ending in the same word, so نظیر at 1a and 5b becomes compare and anywhere. That is a real loss and it belongs to the CHI policy, not to this poem.

۱۲۶ REP — the repeated tail kept (written second)

On earth there is not so much as an equal of yours, and the moon is not that lit face of yours.

I will not give my heart to the cypress's height and stature: it is not that lovely tallness of yours.

In the whole city, O you with the bow for an eyebrow, I know of nobody who is not the game of that arrow of yours.

Nobody else carries off the hearts of people: there is no heart anywhere that is not a captive of yours.

If you take up with an equal of mine, what can I do, when in the world there is nothing that is the equal of yours?

It is plain enough that a heart of that iron kind is not fit for that silk breast of yours.

The whole world has gone off to its love-making — and Sa'di's name is not in that mind of yours.

The hand's reading. Radif held 8 of 8. Chime 0 of 8: of at every pre-tail position. Sa'di's ایطا is preserved — equal at 1a and 5b renders نظیر at 1a and 5b, which the CHI rendering could not do.

6. Grading — the hand's reading, and the tool's

tools/rhyme_pairs.py over the declared bearers of each rendering, run after the readings above were written, per R44. The bearer of a position is the last token of the line, backed up one token where that token is a stopword — the tool's own rule, unaltered. Output: grading.json.

Three things the tool says that the hand did not.

1. The REP arms grade IDENTICAL at 28 of 28 in all three poems — the mechanical confirmation that the radif is held, and the mechanical confirmation that no chime is present, in one number.
2. Not one of the 84 cross-position pairs in the three CHI renderings is unrelated. 68 are STRICT and 16 are NEAR-N1, and every NEAR is the tool's rime-riche guard, not a failed rhyme: ray/spray, away/way, the four ‑ED bearers that share /r/ (tread · red · spread · dread), and every pair involving anywhere. The hand counts all sixteen full rhymes; the tool downgrades a pair whose consonant before the rime is shared. This is the same hand-versus-tool disagreement RS-20260826b §Arms recorded at L3 and L8, and it is recorded here, before any design exists, so that a design using these texts can declare in advance which reading it treats as confirmatory.
3. In ۹۴ REP the tool backed up past his to of and stopped, because both are stopwords and its rule permits backing up once. So the tool's of is not the bearer §1 identifies (the possessed noun, one token further back). The rule was not altered to chase this; it is reported.

7. What the pair cost, stated as craft rather than as measurement

Four facts the limb produced, none of which needed an API call, and none of which is a claim about what a reader prefers:

1. The REP policy forces an unstressed line-ending at every position in two of the three poems. …of the friend, …of yours — the English radif of a genitive construction ends on a noun preceded by a preposition, and the last stressed syllable of the line falls one or two syllables before the end. CHI ends every line on a stressed rhyme. This is not a fact about these poems; it follows from §1. Any ghazal whose radif stands in a genitive to its bearer will have a REP rendering with weak endings and a CHI rendering with strong ones — which means a reader comparing the two policies is never comparing repetition against chime alone.
2. CHI cannot keep the poet's own repeated bearer, and REP can. Sa'di commits ایطا in two of these three poems. REP reproduces it exactly (beauty…beauty, equal…equal). CHI's rule forbids it. So the policy that keeps repetition keeps a repetition Sa'di's prosodists count a fault, and the policy that keeps the chime silently corrects him.
3. CHI was reachable at 8 of 8 in all three poems, and cost three logged strains — ۹۷ 2b, ۹۷ 6b, ۹۴ 7b — each a metonym or a named emotion the Persian leaves implicit. REP cost none, and could not: with the tail fixed and no sound to satisfy, every line was free to be literal. The chime is bought with sense; the repetition is bought with rhythm.
4. The rhyme family in CHI was chosen after the eight senses were listed, not before, and in all three cases the family that worked was the one containing a word for the hardest position — ‑AY for مجال (go away), ‑ED for پیام (a word he has said), ‑AIR for نظیر (compare). A hand that picks the family first will fail; the constraint is the least tractable position, not the most frequent sound.

Translator's log

Frozen 2026-08-28, before any evaluation of these texts was designed (R54 rule 7, charter §3 A4, CLAUDE.md page conventions rule 8). Decisions and alternatives; no quality claims. Any evaluative sentence here carries internal-judgment-only.

(a) The enumeration took longer than any rendering, and it produced the session's craft finding. Deciding whether each poem was a forced case meant writing out the English frames for each radif before any line existed. That is where the two mechanisms in §1 separated. I had expected all three to fail the way ۱۶ and ۳۱ failed at step 1 — on the radif's part of speech — and two of them fail somewhere else entirely, in the direction the genitive is ordered. ۹۴ then failed for a third reason again. The part-of-speech rule survives as a necessary condition and stops being the whole story.

(b) Which policy is harder to write is not the same question as which is better, and I can only answer the first. CHI took roughly three times as long per poem. All the difficulty is in the last position to be solved, and in all three poems it was a position whose sense had no member in any convenient family. REP was mechanical once the tail was fixed: the tail carries no information, so it constrains nothing but the word order of the last phrase.

(c) Where I refused, and where I did not. Three positions in CHI are logged in §3–§4 as strains rather than refusals, and the distinction I applied is whether the added word changes what is asserted or only how it is pictured. A fortunate ray for an auspicious ascendant pictures the same fact by a part of it. That is my dread names an emotion that the image of a heart turning to blood implies but does not state — this is the nearest of the three to a refusal, and another hand would be within its rights to call it one. No position was graded REFUSED in the end, which surprised me; I had expected CHI to break at one position in at least one poem.

(d) A rhythmic cost I did not anticipate and could not avoid. §7.1. I noticed it after ۹۷ REP was written and before ۹۴ was begun, and I deliberately did not repair it — there is no repair that does not either move the radif off the line-end or promote the bearer into the pre-radif slot, which is the very thing §1 says English will not do. It is a property of the policy, so it belongs on the page. It is also the largest confound in any comparison of these two renderings, and the design that uses them has to be built so that it cancels rather than being estimated.

(e) What I did not do. I did not attempt an AB rendering of any of the three. §1's enumeration says all three are unreachable, and R51 step 1 showed that the honest way to test that prediction is to attempt it. The prediction here is therefore untested by construction, and I say so rather than calling §1 confirmed: the three verdicts in §1 are enumerations, not outcomes, and the only outcome-grade evidence for them is that six of six such predictions came out right at step 1 on different poems. A later session that wants §1's two mechanisms as a result rather than as a reading will have to try to build the thing and fail.

(f) The second arm of a pair reuses the first arm's non-rhyming lines, and I did not notice while writing. Measured after the freeze, per note (bry) — ../dependence-self.json, and reported in ../dependence-note.md §3. The two arms of ۹۴ share 53 seven-word runs and a 19-token run; ۱۲۶ shares 19 sevens; ۹۷, whose arms were written in the other order, shares 9. The reused material is the a-hemistichs, which carry no rhyming position and are therefore unconstrained by either policy — "helpless I have stayed, a whole day long, in his snare, and now I am fallen into an exile" is nineteen tokens of ۹۴ that both arms share. This is a property of the pair, not a defect in it: the a-lines have no reason to differ, and forcing them apart would mean writing one of them worse on purpose. But it means the three pairs are not equally different from each other, and any design over them must report the three poems separately rather than pooling them as if they were three instances of one contrast.

(g) On ۹۴ REP's last bayt. «تا خود غلام کیست که سعدی غلام اوست» is a question whose two halves share a word — whose slave is he, whose slave Sa'di is. REP must end the line on of his, which forces the second half into a relative clause and loses the mirroring of غلام…غلام. CHI keeps the mirroring (whose slave is he, whose slave Sa'di is) and pays elsewhere. Neither policy is free at this bayt and the losses are in different places, which is the arm's whole subject in one line.