Observation

The UHT classifier knows things about entities that their essential properties don’t say. When {{entity:Candle}}‘s semantic triangle lists “made of wax, contains a wick, provides light when burned,” the classifier activates 14 trait bits. The property mapper finds exactly one: {{trait:Physical Medium}}. The other thirteen — {{trait:Physical Object}}, {{trait:Synthetic}}, {{trait:Intentionally Designed}}, {{trait:Observable}}, {{trait:Functional Components}}, {{trait:Contextual Meaning}}, {{trait:Temporal Dynamics}} — come from the classifier’s holistic judgment about what kind of thing a candle is, not what distinguishes it from other things.

This pattern held across all twelve entities tested for {{hyp:HYP-ACTIVEHYPOTHESES-096}}, which asked whether classification is decomposable from semantic-triangle properties via map-properties.

Evidence

Across 12 entities from 12 domains ({{entity:Photosynthesis}}, {{entity:DNA}}, {{entity:Democracy}}, {{entity:Candle}}, {{entity:Volcano}}, {{entity:vaccine}}, {{entity:Piano}}, {{entity:algorithm}}, {{entity:Classical Symphony}}, {{entity:Compass (Navigation Instrument)}}, {{entity:transistor (electronic component)}}, {{entity:hyperinflation}}), map-properties predicted 40 trait bits total against 137 actually active bits. Aggregate recall: 0.190. Aggregate precision: 0.650. The hypothesis required recall >= 0.60 for confirmation and < 0.40 for refutation. This is a clean refutation.

Seven traits were never predicted despite being active in three or more entities: {{trait:Observable}} (9/12), {{trait:Contextual Meaning}} (10/12), {{trait:Functional Components}} (7/12), {{trait:Synthetic}} (6/12), {{trait:Adaptable/Configurable}} (3/12), {{trait:Social Construct}} (3/12), {{trait:Persistence/Continuity}} (3/12). Additional near-total misses: {{trait:Temporal Dynamics}} at 89% miss rate (8/9 active entities missed), {{trait:Intentionally Designed}} at 88% (7/8), {{trait:Physical Object}} at 86% (6/7).

{{entity:vaccine}} was the extreme case: 13 active bits, zero predicted. Its essential properties (“stimulates immune response, contains antigens, prevents disease”) map to no traits at all — yet the classifier sees a physical, synthetic, biological, observable, intentionally designed, regulated, domain-specific, culturally embedded, ritualised, persistent, ethically significant entity.

Interpretation

The 32 UHT traits divide into two functional tiers. Tier 1 consists of property-mappable traits that correspond to specific features an entity exhibits — things like {{trait:Powered}}, {{trait:Outputs Effect}}, {{trait:Biological/Biomimetic}}. These are the traits map-properties can predict. Tier 2 consists of meta-ontological traits that encode what category of being the entity belongs to — {{trait:Observable}}, {{trait:Contextual Meaning}}, {{trait:Synthetic}}, {{trait:Functional Components}}. These are activated by the classifier’s implicit reasoning about the entity’s ontological status, not by matching keywords from its property list.

The 0.650 precision means the property mapper isn’t wrong — when it predicts a trait, it’s usually correct. It’s just radically incomplete. Roughly four-fifths of the classification signal comes from meta-ontological judgments that cannot be derived from stated properties. This means UHT classifications are fundamentally non-decomposable: you cannot explain why an entity received its hex code by listing its properties and tracing each to a trait bit. The classification encodes a holistic judgment about the entity’s place in ontological space.

Action

{{hyp:HYP-ACTIVEHYPOTHESES-096}} moved to closed (refuted). Result recorded as {{res:RES-CALIBRATIONRESULTS-115}} with trace link. Two-tier structural finding recorded as {{obs:OBS-STRUCTURALFINDINGS-080}}. New hypothesis {{hyp:HYP-ACTIVEHYPOTHESES-099}} derived: the seven meta-ontological traits should form a distinct correlation cluster in the phi-correlation matrix across the full entity graph, separable from property-mappable traits. Testing this would validate whether the two-tier partition is an artifact of the map-properties tool or a genuine structural feature of the trait space. A research fact documenting the meta-ontological trait class was stored in the knowledge base.