What a Test Needs

Rigidity and Measurement in the Language of Stress

What each prediction needs in order to be tested, and the distinction among default rigidity, engagement and allocation.

Pace, J. C. (2026). What a Test Needs. FigShare. DOI: 10.6084/m9.figshare.34059579

Overview


What a Test Needs sorts the discriminating predictions of the Language of Stress by what testing them requires. The corpus has repeatedly stated that its predictions wait on a measure of archetype rigidity that does not exist. This document shows that statement is too broad. Of the ten discriminating predictions in the empirical predictions register, three can be run now with existing paradigms, two more need no reading of rigidity but wait on special populations, four need only an ordinal reading, and only the pathology prediction, with the strong form of one other, needs an instrument that compares people.

The document names three things the corpus has used without distinguishing: default rigidity, the tension an archetype engages at, set by substantiation history; engagement, whether a situation brings an archetype into play; and allocated rigidity, the departure from the default once an archetype is engaged. Because allocated rigidity overlaps with what precision-weighting accounts predict, the framework's discriminating tests lie where rigidity tracks stakes rather than confidence, and in default rigidity measured apart from attention.

It proposes a stakes-manipulation design for the dissociation of stakes from confidence, adapted from post-decision commitment paradigms; sets out ordinal signs of rigidity and the guard that keeps them distinct from the outcomes they order; lists the requirements an instrument that compares people would have to meet; and proposes a calibration route toward one. It separates identifying candidates for consciousness, a judgment about architecture, from measuring rigidity, and gives one canonical statement of the measurement limit for other documents in the corpus to cite.

Nothing in the document is evidence, and no design in it has been run. It adds no term to the distortion relation and changes no prediction.