Observation

Ten concept pairs that people routinely conflate — correlation with causation, legality with morality, knowledge with information — are not merely different in UHT space. They are different along precisely the dimension that makes the conflation dangerous.

{{entity:correlation (statistical relationship)}} classifies as {{hex:00002000}} with a single active trait, {{trait:Rule-governed}}. {{entity:causation (causal relationship)}} classifies as {{hex:00000200}} with a single active trait, {{trait:Temporal}}. Jaccard: 0.000. The trait gap names the exact philosophical distinction: correlation is a static mathematical rule; causation requires temporal precedence. This is not a vague signal that the concepts differ — it is a label on the difference.

Four of the ten pairs reveal something stronger: strict trait subsumption, where one concept’s trait set is a proper subset of the other’s. {{entity:sympathy (compassionate concern)}} is a subset of {{entity:empathy (shared emotional experience)}}, which adds {{trait:State-Transforming}} and {{trait:Human-Interactive}}. {{entity:speed (scalar quantity)}} is a subset of {{entity:velocity (vector quantity)}}, which adds {{trait:Compositional}} and {{trait:Temporal}}. The confusion runs in one direction — a richer concept reduced to its simpler component.

Evidence

All ten pairs classified and compared:

Pair	Jaccard	Hamming	Differentiating traits
correlation / causation	0.000	2	Rule-Governed vs Temporal
weather / climate	0.200	4	Observable, State-Transforming vs Politicised, Ethically Significant
legality / morality	0.200	8	Rule-Governed, Institutionally Defined vs Symbolic, Signalling, Compositional, Meta, Politicised, Ethically Significant
knowledge / information	0.222	7	State-Transforming vs Synthetic, Designed, Symbolic, Rule-Governed, Digital, Economic
theory / hypothesis	0.250	3	Rule-Governed, Compositional vs Intentionally Designed
strategy / tactics	0.429	4	tactics is strict subset; strategy adds Symbolic, Compositional, Normative, Social Construct
velocity / speed	0.500	2	speed is strict subset; velocity adds Compositional, Temporal
empathy / sympathy	0.600	2	sympathy is strict subset; empathy adds State-Transforming, Human-Interactive
precision / accuracy	0.667	1	accuracy is strict subset; precision adds Institutionally Defined
democracy / republic	0.706	5	State-Transforming, Human-Interactive, Temporal vs Institutionally Defined, Regulated

Mean Jaccard: 0.387. First use of semantic-triangle confirmed complementary diagnostic value: legality decomposes into 3 institutional compliance properties while morality produces 5 normative/cultural/subjective properties.

Interpretation

This is the clearest practical application UHT has demonstrated. Embeddings would place these pairs close together — they co-occur constantly in educational, philosophical, and journalistic text. UHT separates them and names why.

The subsumption pattern in 4/10 pairs extends the markedness asymmetry previously found in antonym pairs to non-antonym confused concepts. When {{entity:empathy (shared emotional experience)}} subsumes {{entity:sympathy (compassionate concern)}}, UHT reveals that sympathy is empathy minus the transformative and interactive dimensions. The false equivalence has a direction: people mistake the rich concept for the thin one, not vice versa.

{{entity:legality (conformity to law)}} versus {{entity:morality (ethical principles)}} shows the widest gap at Hamming 8. Morality activates six traits that legality lacks — it is {{trait:Symbolic}}, {{trait:Meta}}, {{trait:Politicised}}, {{trait:Ethically Significant}}, {{trait:Compositional}}, and {{trait:Signalling}}. Reducing morality to legality strips away most of what morality is. UHT quantifies the information loss in the false equivalence.

Action

{{hyp:HYP-CLOSEDHYPOTHESES-101}} confirmed, linked to {{res:RES-CALIBRATIONRESULTS-121}}. Subsumption pattern recorded as {{obs:OBS-STRUCTURALFINDINGS-082}}. Two cross-domain analogs and one functional archetype stored as research facts. Next session should scale this to 50+ pairs with automated embedding-cosine comparison to produce the publishable UHT-vs-embedding divergence study from directive 6.