Repository path: workshop/translations/hafez-darad/dependence-note.md · rendered 2026-09-09

Contamination measurement — T-hafez-darad-R56-v1

Run with tools/dependence_check.py on the frozen rendering of record (IP, 349 tokens over both ghazals), against every English comparator this session could reach without opening a published rendering of either poem. Raw output: runs/RS-20260829b-inversion-price/dep.json.

Declared contamination: none, and this is the basis. Zero shared 7-grams against a complete Victorian verse Hafez, and a longest common run of three tokens.

What no comparator could be found for, said plainly. Neither sh118 nor sh124 is among Leaf's 28 odes — that is a condition of the frozen selection rule — and no published English of either poem was fetched, opened or read at any point. Payne 1901 vol. 1 may contain both; its text was not in the repository and was not re-downloaded, because doing so before the rendering was frozen is exactly what the standing contamination rule forbids and doing so after adds a number that no figure in this session depends on.

Why no figure depends on it. The design that uses these arms (E-20260829b-inversion-price) contrasts the lead's four renderings against each other. Every arm has the same author and the same source; what varies is word order and lexis, both under the hand's declared control and both machine-verified (pair_check.py, 204 checks, 0 failures). An overlap with a published hand would change nothing in that contrast. The standing rule's own constraint — the lead is never the independent third translator where a design's validity turns on independence from any published rendering — does not bind here, because nothing here turns on it.

Note (bhb) is not triggered. The lead-versus-lead figure above is on different poems, so it measures house style and not re-rendering; two tokens is the lowest such reading the project has recorded, against a record of 37.