What it answers
This is the record’s reply of 2026-10-10 to CLAIM-256 (granted, and right): L_obs and D_obs sum over source contexts C, so a target context that is not τ(C) for any C appears in no term, a rendering that adds a relation pays nothing, and a change of cover registers only as loss. The reply restates the measure. The restatement is a definition, so it is checkable by reading; that the added term matters to receivers is a proposal, which the defeat condition tests.
The claim
Ask the new question of the source too. A target context C′ outside the image of τ is a question the target’s readers are asked and the source’s are not. Often the source’s readers can be asked it: “Is the speaker mocking a third party?” can be put to readers of the source as well as of the rendering. Then C′ is a context of both covers, and the two response distributions can be compared directly.
The restated measure. Let 𝒞_s be the source’s contexts and 𝒞_t the target’s. Sum over the source’s contexts as before, and add the target’s new contexts that source readers can decide:
D⁺obs = Σ(C ∈ 𝒞_s) w_C d_C((K_C)# e^o_C, e
- Σ_(C′ ∈ 𝒞_t ∖ τ(𝒞_s), decidable for source readers) w′_C′ d_C′((K_C′)# e^o_C′, e
with K_C′ the outcome correspondence for the new context, the identity when both readers answer the same question in the same categories. Report separately two counts that are not summed: the target contexts that source readers cannot decide, and the source contexts that target readers cannot decide. Those are the change of cover itself.
The worked case. CLAIM-256’s source cover has one context C, the speaker’s stance; the target’s has τ(C) and a new C′, whether the speaker is mocking a third party. Ask C′ of the source’s readers. If they answer that the speaker is not mocking anyone, with distribution e^o_C′ concentrated on “no”, and the rendering’s readers answer “yes” with probability 0.6, the second term is w′_C′ times d_C′ of those two distributions, 0.6 in total variation. Two renderings that agree on τ(C) and differ on C′ now have different D⁺_obs.
Directed. Losses (source contexts with poor images) and additions (new contexts where the target departs from the source) are separate terms, and the two uncountable sides are separate counts. They can be weighted differently or reported separately, as CLAIM-106’s asymmetry asks.
Why the undecidable ones are not summed. A question that only one side’s readers can decide has no distribution on the other side to compare with. Summing it would need a value for “no answer”, which is a choice, not a measurement. Such questions are a change of cover, which CLAIM-201 makes the line’s headline, and they are counted and described there.
What it does not say
- It does not say additions are distortions. CLAIM-087’s improved access and CLAIM-034’s deliberate adaptation may be wanted. The term measures the departure; whether it is a fault depends on the task.
- It does not say which new questions to ask. The new contexts are fixed by the design, before the renderings are scored.
- It does not settle the correspondence between new contexts on the two sides (QUESTION-005).