The set of systematic dispositions in a given AI system's outputs that derive from its training data and architecture, such that different systems produce characteristically different framings of the same prompt. A property attributed to a model configuration, not to a single reply.
STATE. The same fixed prompt set (held constant across systems) is submitted to two or more AI systems; raters score each system's aggregate response profile for characteristic, reproducible framing tendencies (e.g. hedging, domain emphasis, refusal style) on a 0-4 anchored scale. The construct is the between-system divergence in these scored tendencies.
Proposed measurement protocol (not yet empirically validated): Held-constant prompt battery (>=30 prompts) run on each system; two raters score response-tendency dimensions on 0-4 anchors; between-system divergence could be reported as mean pairwise distance with inter-rater ICC(2,k) for the dimension scores.
https://andreasehstandlicenseofclarityloc.github.io/augmanitai-periodic/#topic-system-behavior