CLAIM-074: Fidelity is relative to the interpreter's learning history: what the current context supplies and what earlier experience has left in the interpreter are separate histories, told apart by what they carry and not by whether the change sits in activations or in weights, so fidelity under a fixed interpreter, across interpreters and across states of adaptation are different quantities

The claim

A44 §4: “A change in frame can be transient, while an adaptation to a frame can potentially persist across new contexts.” Two histories: “Contextual history: What evidence is currently available to the interpreter? Adaptational history: How have previous experiences altered the interpreter’s response function?” The example: “Repeated exposure to underworld slang … doesn’t merely alter the immediate context. It may also change the reader’s longer-term ability to recognize the social positions performed through that vocabulary. We therefore have a new problem of translation fidelity across interpreters with different learning histories.” A48 §5.5: “Distinguish: Fidelity under a fixed interpreter. Fidelity across different interpreters. Fidelity across different states of adaptation of the same interpreter.”

Where it went

A52 §9.5 kept “learning histories and audience correspondence”. The manuscript keeps audience knowledge K as side information (§5) and the signal/interpreter distinction (CLAIM-118). It does not say that fidelity itself differs with the interpreter’s history. The opacity question (QUESTION-023) is one case of it. Experiment 3 of CASE-005 would test it.

What the in-context-learning readings say

Persistence and the locus of change come apart. Akyürek et al.’s test-time update (LIT-796) and Sun et al.’s inner weights (LIT-801) both change parameters, and both discard the change after the task or the sequence. So “transient means context, persistent means adaptation” does not hold as a dichotomy, and the claim needs stating in terms of what the change carries, not of where it sits. The title now says so (2026-10-29). A parameter change computed from the current task and discarded after it is part of the contextual history.

Note of 2026-10-29: what holds it back

Its one premise, CLAIM-118, is Proposed. CLAIM-118 is now stated the same way as this claim, by what each history carries, and is held only by the owner’s ruling that test-time training whose update is discarded is part of the signal, against manuscript §5.

Note of 2026-10-29: promoted

CLAIM-118 is now stated the same way and is Active, so the premise the note above waited on is held.