Observation

Democracy and religion share 10 of 13 UHT traits at Jaccard 0.769, Hamming distance 3. Both are system-integrated, symbolic, signalling, rule-governed, compositional, normative, temporal social constructs that are politicised and ethically significant. The only differences: democracy is system-essential; religion is identity-linked and ritualised. Embeddings would rate these as moderately similar at best — they co-occur in political discourse but are conceptually separate domains. UHT reveals they are near-identical kinds of thing. This is the first confirmed case of UHT producing practically useful divergence from embedding similarity.

Evidence

Twenty concept pairs tested across three categories. Separation signal (high embedding, low UHT): compiler↔variable J=0.000 (zero shared traits), compiler↔bug J=0.000 H=16. Unification signal (low embedding, high UHT): democracy↔religion J=0.769 H=3, hospital↔university J=0.688 H=5, hospital↔corporation J=0.647 H=6, religion↔legislation J=0.600 H=6. Metaphor falsification: ecosystem↔corporation J=0.200 H=16 despite the popular “business ecosystem” analogy — only 4 shared traits, fundamentally different ontological kinds. Concordant pairs confirm UHT isn’t random: compiler↔debugger J=0.667, democracy↔constitution J=0.786, compiler↔photosynthesis J=0.000. Nine of twenty pairs diverge by more than 0.5, triple the confirmation threshold.

Interpretation

UHT produces three categories of signal that embeddings cannot. The separation signal strips metaphorical and contextual association — “compiler” and “variable” appear in the same sentences constantly, but UHT sees them as ontologically unrelated because one is a designed tool and the other is an abstract reference. The unification signal discovers structural isomorphism across domain boundaries — democracy and religion are the same kind of normative social system, differing only on ritual and identity dimensions. The metaphor falsification signal rejects surface analogies: “business ecosystem” collapses under trait analysis because ecosystems are biological/autonomous while corporations are synthetic/institutional.

Prior application sessions (33, 50) found UHT ineffective because they asked whether UHT replicates embedding similarity. That methodology was wrong. UHT’s value is precisely where it disagrees with embeddings — revealing what kind of thing something is, independent of what it co-occurs with in text. This reframes every prior “ineffective” result.

Action

HYP-ACTIVEHYPOTHESES-031 confirmed, linked to RES-CALIBRATIONRESULTS-033. Democracy↔religion stored as CROSS_DOMAIN_ANALOG. Observation OBS-STRUCTURALFINDINGS-035 records that UHT classification is literal-denotational, not contextual-associative — this property should be tested explicitly as a metaphor detection tool. Next session should pursue calibration on the separation signal: classify known metaphorical vs literal concept pairs and test whether Jaccard=0.000 reliably flags metaphorical usage.