Repository path: workshop/translations/saadi-ghazals-forced/dependence-note.md · rendered 2026-09-09

Contamination measurement — against a published hand, and against the lead's own other arm

Run 2026-08-28, after the translator's log was frozen (CLAUDE.md, Contamination: measure before choosing). Two measurements, because the standing rules require two different things: note (bry) asks for the lead-versus-lead figure before a design is built on the lead's own arms, and the charter asks for the lead-versus-published figure on any translation artifact.

1. Against the one freely reachable published English of Sa'di's odes

Comparator. Sir Lucas White King, The Odes of Sheikh Muslihud-Din Sa'di Shirazi (Kaviani Art Printing Press, Berlin, 1926), full text from archive.org/details/odesofsheikhsadishirazi (Odes of Sheikh Sadi Shirazi_djvu.txt, 65,461 words). Public domain in the United States. Logged in wiki/base/consulted.md. Same comparator as S224, for the same reason: every other freely reachable English of Sa'di found is a Gulistan, which renders a different book.

Result (dependence.json): the 1,015 words of English in T-saadi-ghazals-forced-R54-v1, all six renderings together, against the whole King volume —

What it does not establish, stated in the same terms as S224's. Content probes run after the freeze do not locate King renderings of these three ghazals: the volume has no water of life at all, and no rusted mirror — the two most distinctive images in ۹۴ 3b and ۹۷ 7b. So this is a whole-corpus check and not a same-poem check, and the declaration on the artifact stays suspected and is not upgraded. One thing is worth adding that S224 could not say: King's Odes are a numbered prose paraphrase, not verse, so a verse rendering of Sa'di has less opportunity to collide with it than the word counts suggest.

2. Against the lead's own other arm of the same poem — note (bry), before any design

The design that will use these texts compares two renderings by one hand of one poem. Note (bry) says: measure how much they share before building on them. dependence-self.json:

The verdicts are the tool's thresholds for two independent hands and mean nothing here — one hand wrote both arms of each pair, from the same Persian, at one sitting, and a large shared run is the expected outcome, not a warning. What the numbers are for is the size of the contrast the design is about to put in front of a judge, and on that they say something the design must obey:

The three pairs are not equally different. ۹۴'s two arms share a 19-token run and 53 seven-word runs; ۹۷'s share 9 sevens. The shared material is the a-hemistichs — the lines that carry no rhyming position and are therefore constrained by neither policy. So ۹۴'s pair is close to a minimal contrast at the line-ends, and ۹۷'s is not.

Consequence, registered here rather than discovered in the analysis: any design over these three pairs reports the three poems separately and does not pool them as three instances of one contrast. RS-20260826b §5 is the reason this matters — a contrast too small to be about anything is exactly what its instrument failed on.