The claim
This is the record’s reply of 2026-10-10 to CLAIM-244. It reconciles the objection’s dilemma rather than refuting it, and it is a proposal: that holistic judgements behave as the claim says is something a study can find false.
CLAIM-244 says the line must choose. Either the criterion in the defeat conditions (one holistic judgement of fidelity) is well defined, and the relativity theses (CLAIM-128, CLAIM-074, CLAIM-106) are in trouble; or they are right, and the criterion is not well defined. The dilemma is not forced.
CLAIM-128 says that “no single ranking of translations holds across tasks”. It does not say that a population of judges, asked “how faithful?”, brings no task to the question. A population can have a default task, the one its members assume when none is stated, and that task fixes a weighting of the profile’s components (TERM-043). Holistic fidelity is then fidelity under that weighting. It is one task among many, which is what CLAIM-128 allows.
The model. For a rendering r with profile components D_1(r), …, D_k(r), the holistic judgement J(r) of population P is modelled as
J(r) ≈ g(w_1 D_1(r) + … + w_k D_k(r)),
with g a decreasing link (more distortion, less fidelity) and the weights w estimated on one set of renderings A, judged by one sample of judges from P, and then tested on another set B judged by other judges from P.
A worked case, O04’s own. Two components: D_f, the loss of footing, and D_s, the loss of stance, each on [0, 1]; take g(x) = 1 − x. Rendering A keeps footing and loses stance: D_f(A) = 0, D_s(A) = 1. Rendering B keeps stance and loses footing: D_f(B) = 1, D_s(B) = 0. Suppose the held-in renderings give w_f = 0.7, w_s = 0.3. Then
- J(A) = 1 − (0.7 · 0 + 0.3 · 1) = 1 − 0.3 = 0.7,
- J(B) = 1 − (0.7 · 1 + 0.3 · 0) = 1 − 0.7 = 0.3,
so the weighting predicts that P’s judges rank A above B, before they see either. If held-out judges from P rank B above A, the prediction fails.
The two horns.
- Circular. The objection’s first horn is that the weights are fitted to the criterion. They are, on A. They are not on B, where they are fixed before the judgements are seen, so predicting B is a finding and not a fit.
- Unfalsifiable. The second horn is that a mismatch can be blamed on the judges having weighed for a different task. A defeat condition that names P and the preregistered w leaves no room for that: if P’s judges on B depart from the fixed w, the thesis that used them failed, and changing w after the data is not allowed.
So the criterion is well defined (as fidelity under P’s default task), and the relativity theses stand (another task, another weighting).
What would show it wrong. If no weighting fitted on A predicts the holistic judgements on B better than the best single component does, then holistic judgements are not a weighting of the profile, and the reconciliation fails. That is the defeat condition.
What it answers
CLAIM-244’s dilemma, and its second suggested answer (“weights estimated from one set of renderings and judges and tested on another”), which this entry adopts. What the objection is right about, that the stated conditions name no task and no receivers, is met by restatement in CLAIM-220 and CLAIM-204, and held open in QUESTION-075. CLAIM-115’s and CLAIM-139’s own conditions are unchanged.
What it does not say
- It does not say holistic judgements are the right criterion. The revised thesis CLAIM-220 uses a task-indexed behavioural one, and keeps holistic ratings as a secondary outcome.
- It does not say P’s default weighting is privileged. It is one task’s.
- It does not say the link is linear or the components fixed. The model’s form is part of what the preregistration fixes.
Note of 2026-10-10: a documented case
CASE-067, found by a search the owner asked for, illustrates this claim with a documented reception split over one translation, Deborah Smith’s The Vegetarian. Each side’s default task is stated in print: the Booker judges evaluate “the finished English-language work on their own terms”, and assessment in Korea “does usually involve comparison”. One critic, Charse Yun, gives two verdicts on one set of perceived features by changing task. It does not test the defeat condition: it is qualitative, and no weights can be fitted.
It also shows a limit of the model. Part of the split is access, not weights. Some divergences, such as the confusion of “arm” (pal) and “foot” (bal), are invisible to readers without Korean (“Most English readers will simply glide over them unaware”). For those components the Anglophone weight is zero by blindness, not by a default task. A test of this claim across populations with unequal access to the source has to separate the components both populations can perceive from those only one can, and fit weights on the first kind only.
Note of 2026-10-27: the limit filed as an objection
The limit the note of 2026-10-10 recorded, that part of a reception split between populations with unequal access is access and not weighting, is filed as CLAIM-458, a counter, Proposed. A test of this claim across such populations meets it by fitting weights only on the components both can perceive. The claim is unchanged.
Note of 2026-10-30: what holds it back
The result its own text says decides it is missing: no weighting of the profile’s components has been fitted on one set of renderings and judges and tested on another set from the same population against the best single component. CASE-067 is qualitative.