Observation

The four-layer architecture of the trait set — physical, functional, abstract, social — is not the primary organizing principle of how traits actually co-activate across the corpus. Of thirty trait pairs showing strong correlation (phi > 0.40), eighteen cross layer boundaries while only twelve stay within a single layer. The designed structure is a scaffold, not a skeleton. The taxonomy self-organized into dimensions that cut orthogonally across the layers its creators intended.

{{hyp:HYP-ACTIVEHYPOTHESES-099}} predicted that seven “meta-ontological” traits (bits 2, 6, 12, 16, 20, 25, 30) would form a distinct correlation cluster separable from the remaining property-mapped traits. The opposite is true: these traits correlate more strongly with traits outside the proposed cluster (mean phi 0.159) than with each other (0.130).

Evidence

The full 32x32 phi-correlation matrix was computed over n=1500 entities drawn from the graph’s 16,454.

The strongest single correlation in the taxonomy is {{trait:Physical Object}} ↔ {{trait:Physical Medium}} at phi 0.761 — unsurprising and essentially tautological. But the second strongest, {{trait:Synthetic}} ↔ {{trait:Intentionally Designed}} at 0.680, crosses the physical-functional boundary. {{trait:Synthetic}} (bit 2) is the most connected trait in the entire system, with a connectivity sum of 8.252 — it correlates with traits in all four layers. Whether something is human-made predicts its regulatory status, economic significance, compositional structure, and human interactivity more strongly than any within-layer property predicts another.

Principal component analysis reveals PC1 (21% of variance) as a single “artifact” dimension spanning all four layers: Synthetic, Regulated, Physical Medium, Human-Interactive, Economically Significant. PC2 (9.5%) separates material entities (Physical Object, Observable, Biological) from symbolic ones (Social Construct, Symbolic, Rule-governed) — the classic material-abstract divide. PC3 (6.7%) captures information processing: Powered, Processes Signals/Logic, Digital/Virtual. PC4 (6.2%) captures living agency: Active, Functionally Autonomous, Biological.

Eight components account for 59.5% of total variance. Fifteen are needed for 80%. The 32 traits are not highly redundant — effective dimensionality is genuinely high. Three traits — {{trait:Meta}} (connectivity 1.165), {{trait:Temporal}} (1.554), and {{trait:Digital/Virtual}} (2.826) — operate nearly independently of everything else, encoding their own private dimensions.

Layer permeability is asymmetric: physical-functional inter-layer correlation (0.232) approaches within-layer cohesion (physical 0.278, functional 0.248). But physical-abstract inter-layer correlation is essentially zero (0.008). The abstract layer has the weakest internal cohesion of any layer (0.096), suggesting its eight traits are measuring genuinely different things.

Interpretation

The trait set was designed as four layers of eight bits, but the corpus reveals that the real axes of variation cut across these layers. The most important axis is “human-made complexity” — not a physical property, not a functional one, not abstract or social, but all four simultaneously. This makes sense: artifacts are defined precisely by their cross-cutting nature, existing simultaneously as physical objects, functional tools, rule-governed systems, and economically valued goods.

The near-independence of physical and abstract layers confirms what earlier sessions found about the material-symbolic divide, but the new finding is that this divide is PC2, not PC1. The artifact dimension dominates. UHT’s primary discriminating axis is whether something was made by humans, not whether it is physical or conceptual.

{{res:RES-CALIBRATIONRESULTS-117}} records the refutation of {{hyp:HYP-ACTIVEHYPOTHESES-099}}. The “meta-ontological” cluster proposed in {{obs:OBS-STRUCTURALFINDINGS-080}} does not exist as a separable correlation block.

Action

Two research facts stored: the full correlation structure summary and the layer permeability asymmetry. {{hyp:HYP-ACTIVEHYPOTHESES-099}} moved to closed hypotheses.

The finding that {{trait:Meta}}, {{trait:Temporal}}, and {{trait:Digital/Virtual}} are near-independent dimensions raises a question: are these traits under-activated (too rarely triggered to correlate) or genuinely orthogonal to everything else? Meta activates for only 1.5% of entities, Temporal for 22.9%, Digital/Virtual for 3.4%. The next session should investigate whether Meta and Digital/Virtual are effective classifiers despite their rarity, or whether their low activation rates make them functionally inert — noise bits that inflate the apparent dimensionality without contributing to discrimination.