Toward Appraisal-Structured Latent Control Representations in Language Models: A Preregistered Hypothesis and Experimental Design for Task-Induced Response-Strategy Modulation

A preregistered hypothesis and experimental design proposing that several known single-behaviour latent directions in language models — refusal, abstention, uncertainty reporting, evaluation awareness — may be special cases of a more general family of task-induced, appraisal-structured latent control representations that partially govern response-strategy selection across domains. A factorial task-state x concept-mention design separates internal task appraisal from lexical concept tracking; a mandatory shared-versus-separate low-rank test asks whether the axes form one control family rather than three isolated directions. Version 0.1.2 is a pre-data methodological correction of the H1 and H2 decision rules, the frozen analysis specification, the pre-run artifact gate, and the response-label reliability protocol. No data have been collected, no model has been run, and no results exist.