Observation

The 32 UHT traits encode only 7.3 bits of effective information. Hierarchical clustering of the phi-coefficient matrix across 400 entities recovers eight semantically coherent dimensions — and the predicted agency cluster splits in two, revealing a distinction the trait designers never named.

Session 103 hypothesised that the ~8 effective dimensions (estimated from joint entropy) would correspond to interpretable trait clusters, specifically predicting a physical-material cluster (bits 1, 6, 7), an agency cluster (bits 4, 10, 12, 15), and an institutional cluster (bits 26, 28, 29). The physical-material and institutional predictions held cleanly. The agency prediction fractured into something more interesting.

Evidence

From 400 entities with 265 unique hex patterns, the phi-coefficient matrix yields these clusters under Ward linkage at k=8:

Physical-material (bits 1, 5, 6, 7, 16, 17, 25): phi between bits 1-7 reaches 0.878, the strongest pairwise correlation in the entire matrix. Observable (bit 6) and Physical Medium (bit 7) are nearly synonymous at phi=0.721.

Designed-institutional (bits 2, 9, 13, 14, 26, 28, 29): Synthetic and Intentionally Designed correlate at 0.681. Regulated and Economically Significant at 0.669. The engineering and governance traits form one dimension — things that are both built and regulated.

Active-processing (bits 4, 10, 11, 12, 20): Powered things that transform state and process signals. Outputs Effect and State-Transforming at 0.561. This is the machine cluster — smartphones, nuclear reactors, quantum computers.

Autonomous-living (bits 3, 8, 15): Biological, Active, and Functionally Autonomous. Active and Functionally Autonomous correlate at 0.707. This is the organism cluster — octopi, wolves, bacteria. The predicted “agency” dimension is actually two: one for engineered systems that process, another for living systems that act.

Socio-normative (bits 21, 27, 31): Normative, Identity-Linked, Ritualised. Temporal-ethical (bits 18, 23, 30, 32): Signalling, Temporal, Politicised, Ethically Significant. Rule-digital (bits 19, 24). Meta (bit 22, singleton — activated in only 2.2% of entities).

Sum of individual trait entropies is 19.7 bits; actual joint entropy is 7.3 bits. The traits are 63% redundant. Only 265 of 4.3 billion possible patterns appear.

Interpretation

The trait set implicitly encodes a nature/artifact boundary that no single trait captures. Bits 3+8+15 co-activate for living things; bits 4+10+11+12 co-activate for engineered things. Both clusters involve “agency” in the colloquial sense, but they activate through entirely different trait pathways. A dolphin and a smartphone are both active, state-changing systems — but UHT separates them along orthogonal dimensions.

The 7.3-bit effective space means trait proposals should target gaps between these eight clusters rather than adding more traits within them. A 33rd trait correlated with existing physical-material traits adds almost no information. A trait that discriminates within the active-processing cluster — say, distinguishing computational from mechanical transformation — would expand the effective dimensionality.

Action

HYP-053 confirmed with caveats and moved to Closed Hypotheses. RES-CALIBRATIONRESULTS-060 records the full result with trace link. New hypothesis HYP-ACTIVEHYPOTHESES-055 asks whether the 8-dimensional cluster profiles create a useful cross-domain similarity space: do entities sharing identical profiles come from different domains at least 30% of the time? Research fact stored for the dimensional structure. Next session should test HYP-055 or pursue APPLICATION if the timer is ready.