Observation

Two diagnostic tools — semantic-triangle and map-properties — had never been tested as a classification prediction pipeline. {{hyp:HYP-ACTIVEHYPOTHESES-092}} asked whether feeding an entity’s natural language properties through map-properties could predict its actual UHT classification within 5 bits. It cannot. Across 10 entities from 10 domains, the pipeline achieved a mean Hamming distance of 9.8 — nearly a third of all 32 bits wrong. Not a single entity reached the confirmation threshold. The failure is not random: it follows a clean gradient that reveals something fundamental about how UHT traits relate to natural language description.

Evidence

Ten entities tested: {{entity:Candle}} ({{hex:C6D80008}}), {{entity:Democracy}} ({{hex:00BFFEFF}}), {{entity:Photosynthesis}} ({{hex:34753209}}), {{entity:Violin}} ({{hex:E6C81008}}), {{entity:Volcano}} ({{hex:8E501200}}), {{entity:DNA}} ({{hex:A6337201}}), {{entity:Cathedral}} ({{hex:CE88908A}}), {{entity:Cognitive Empathy}} ({{hex:00380040}}), {{entity:Computational Algorithm}} ({{hex:40A53709}}), {{entity:Alpine Glacier}} ({{hex:8650020D}}).

Pipeline overall: recall 33.0%, precision 61.7%, mean Hamming 9.8 (69.4% bit agreement). The actual classifications averaged popcount 11.2; the pipeline predictions averaged only 6.0 — it systematically under-activates traits.

Seven traits were virtually invisible to property-based mapping (0% recall, 4+ missed activations each): {{trait:Compositional}} (7/10 missed), {{trait:Regulated}} (7/10), {{trait:Measurable}} (7/10), {{trait:Economically Significant}} (6/10), {{trait:Temporal}} (5/10), {{trait:Synthetic}} (4/10), {{trait:Intentionally Designed}} (4/10). Six traits were reliably captured (67%+ recall): {{trait:Physical Object}}, {{trait:Biological/Biomimetic}}, {{trait:Powered}}, {{trait:Normative}}, {{trait:Regulated}}, {{trait:Rule-governed}}.

{{entity:Democracy}} showed the worst case: actual popcount 22, predicted popcount 9, Hamming distance 13. The pipeline correctly identified Rule-Governed and Normative but missed Compositional, Regulatory, Measurable, Economically Significant, Temporal, and 8 others.

Interpretation

The captured traits describe first-order observable properties — things you can see, touch, or directly verify about an entity. “Is it a physical object?” maps cleanly from “made of wax.” “Is it biological?” maps cleanly from “found in chloroplasts.” These are the easy cases where natural language properties align with UHT trait definitions.

The invisible traits are second-order ontological categories — judgments about what kind of thing something is that require holistic reasoning rather than property matching. Nothing in Democracy’s property list (“citizen participation,” “elections,” “rule of law”) signals that it is compositional, measurable, or economically significant — yet it clearly is all three when you consider what Democracy is rather than what it does. The 32-prompt evaluation achieves this by asking each trait question in full context; the property-to-trait mapper cannot replicate this holistic judgment from atomized property fragments.

This replicates and sharpens the prior finding (recall 0.21, N=10) with a larger effective sample (320 bit-level decisions). The abstraction gap is structural, not a limitation of the mapping algorithm.

Action

{{hyp:HYP-ACTIVEHYPOTHESES-092}} moved to closed hypotheses. Result recorded as {{res:RES-CALIBRATIONRESULTS-106}}. Structural observation {{obs:OBS-STRUCTURALFINDINGS-072}} filed, noting the trait visibility gradient and suggesting a test of whether the invisible traits correspond to the 17-trait “designed-artifact-attractor” mega-cluster. The research fact on the first-order/second-order property gap updated with replicated statistics. Next session should investigate whether the 7 invisible traits cluster together in trait correlation space — if they form a single correlated block, it would explain why the pipeline fails systematically rather than randomly.