Observation

Decision theory is not one thing in UHT space — it is two. The nine entities classified in this session fractured into a tight formal-methods cluster around 40A0xxxx and a diffuse cognitive-observational cluster around 0000/0080xxxx, with almost nothing in between. Expected utility, decision matrix, prospect theory, and minimax regret all landed within a single bit-flip of each other. Bounded rationality, loss aversion, and value of information collapsed to the same hex code: 00008080. The gap between these two groups — Jaccard 0.25 — is the widest intra-domain split observed in thirteen corpus expansion sessions.

Evidence

Normative cluster: expected utility 40A0B880, decision matrix 40A0B880 (identical), prospect theory 40A0B080, minimax regret 40A0A880. Pairwise Jaccard within this group: 0.875 to 1.0. All share Synthetic, Intentionally Designed, Processes Signals/Logic, Symbolic, Rule-governed, and Social Construct traits.

Descriptive cluster: bounded rationality 00008080, loss aversion 00008080, value of information 00008080 (all identical), satisficing 00800080. These retain only Symbolic and Social Construct from the normative set, losing all formal-method traits.

Bayesian updating at 00B0A200 sits between the two clusters, sharing Jaccard 0.4 with expected utility — the only entity not cleanly in one camp.

Cross-domain: expected utility to risk matrix (risk-management domain) yields Jaccard 0.875. Expected utility to Nash equilibrium (game theory) yields only 0.375. UHT sees a risk matrix as more similar to expected utility than a Nash equilibrium — functional role trumps disciplinary origin.

Interpretation

UHT’s trait system captures an ontological distinction that decision theorists have debated for decades: normative theories prescribe how agents should decide, while descriptive theories characterise how they actually do. The framework maps this directly onto its bit structure. The normative cluster activates Processes Signals/Logic, Rule-governed, and Compositional — traits of formal machinery. The descriptive cluster retains only the abstract representational traits. This is not a bug; it is the classification doing exactly what it should. The risk-matrix cross-domain result reinforces session 17’s refuted context-sensitivity hypothesis: UHT is genuinely context-insensitive and role-sensitive.

Bayesian updating’s intermediate position makes conceptual sense — it is both a formal method and a description of how evidence should revise belief, bridging the two worlds.

Action

Created COR-DOMAINEXPANSIONS-013 with all nine entities and cross-domain analysis. Baseline BL-UHTRESEARCH-024 captures the new state. The normative-descriptive split suggests a calibration hypothesis for a future session: does the gap widen or narrow when decision-theoretic concepts are classified with explicit domain context (e.g. “bounded rationality in behavioural economics” vs bare)? This would revisit the context-sensitivity question from session 17 but with entities that have a stronger theoretical reason to shift. Bayesian updating’s bridging position also warrants targeted comparison against information-theory entities like mutual information (Jaccard 0.286) to test whether it clusters more with formal methods or with epistemic concepts under varied context.