Observation

Thirty-four percent of UHT’s trait space is completely invisible to natural language property descriptions. When asked to list the essential properties of a candle, a person says “wax, wick, light” — not “{{trait:Intentionally Designed}}, {{trait:Observable}}, {{trait:Synthetic}}, {{trait:Compositional}}, {{trait:Economically Significant}}.” The map-properties and semantic-triangle diagnostic tools — never previously tested — expose this gap precisely. They form a pipeline: semantic-triangle decomposes a concept into its Ogden-Richards semiotic components and extracts essential properties; map-properties maps those properties to candidate UHT trait bits and suggests a hex code. The predicted hex codes diverge dramatically from actual classifications.

Evidence

Across 10 entities from 10 domains ({{entity:Candle}}, {{entity:Democracy}}, {{entity:Photosynthesis}}, {{entity:wedding}}, {{entity:carbon credit}}, {{entity:3D Printer}}, {{entity:Symphony Orchestra}}, {{entity:bankruptcy (economic cessation)}}, {{entity:vaccine}}, {{entity:Cathedral}}), the map-properties pipeline predicted hex codes with a mean Hamming distance of 11.3 bits from actual classifications — 64.7% bit agreement against a 50% random baseline. The mechanism is asymmetric: precision was 0.79 (when map-properties predicts a trait, it is correct four times out of five) but recall was 0.21 (it misses 79% of actually activated traits). Mean predicted popcount was 3.5 versus actual 12.8.

Eleven traits were missed 100% of the time: {{trait:Processes Signals/Logic}}, {{trait:State-Transforming}}, {{trait:System-integrated}}, {{trait:Functionally Autonomous}}, {{trait:System-Essential}}, {{trait:Symbolic}}, {{trait:Signalling}}, {{trait:Compositional}}, {{trait:Meta}}, {{trait:Temporal}}, {{trait:Politicised}}. These span the functional layer (bits 11-16) and abstract layer (bits 17-23) almost completely. The social layer was partially visible — {{trait:Ritualised}} was missed only 33% of the time, suggesting ritual is the most lexically salient social property.

A secondary test showed that semantic-triangle enriched context produces classifications within 0-5 bits of original classifications (mean 2.7), confirming the classifier is robust to rephrased context. The bottleneck is the property-to-trait mapping itself, not context quality.

Interpretation

UHT traits encode second-order ontological properties — properties about the kind of existence something has — not the first-order properties humans naturally enumerate when describing things. Asking “what are the essential properties of a vaccine?” yields “stimulates immune response, contains antigens, prevents disease.” None of these map to any of the 13 traits the classifier actually activates for {{entity:vaccine}} at {{hex:E688025D}}. The classification prompts perform a fundamentally different cognitive operation: not “what does this thing do?” but “what kind of thing is this?” — is it designed, does it process signals, is it system-integrated, is it compositional?

This explains why classify with rich --context descriptions consistently outperforms bare-name classification. The context gives the classifier material for its second-order analysis. But the analysis itself is irreducible to keyword matching — it requires the 32 individual trait evaluation prompts that each make a binary judgment from a different ontological angle.

The high precision of map-properties means it remains useful as a partial explainability tool: when it identifies a trait, that identification is reliable. The recall gap is the diagnostic signal — traits that map-properties misses but classify activates are precisely the implicit ontological judgments that make UHT non-trivial.

Action

Created {{hyp:HYP-ACTIVEHYPOTHESES-093}} (refuted, moved to closed) with result {{res:RES-CALIBRATIONRESULTS-105}} and trace link. Recorded structural observation {{obs:OBS-STRUCTURALFINDINGS-071}} documenting the first-order/second-order property gap. Stored research fact on the 11 invisible traits. A natural follow-up: test whether the 11 invisible traits form a coherent cluster in the trait correlation matrix — if they do, they may represent a single latent dimension of “systemic embeddedness” that natural language systematically fails to articulate.