Observation

Negation does not behave as trait modification. Across eight concept-antonym pairs — order/disorder, convergence/divergence, equilibrium/disequilibrium, continuity/discontinuity, determinism/indeterminism, reversibility/irreversibility, completeness/incompleteness, dependence/independence — UHT produces a strictly bimodal Jaccard distribution: either 1.0 (identical) or 0.0 (completely disjoint). Not a single pair falls in the 0.30–0.80 range that would indicate shared-with-modification encoding. The hypothesis that UHT would capture negation as related-but-distinguishable was refuted on all eight pairs.

Evidence

Three pairs classify identically: convergence and divergence both at 00000000, continuity and discontinuity both at 00000000, determinism and indeterminism both at 00008000 (Symbolic only). These pairs are invisible to UHT — the trait set has nothing to say about them.

Five pairs are completely disjoint (Jaccard 0.0). The most revealing is dependence (00000001: Ethically Significant) versus independence (00008084: Symbolic + Social Construct + Politicised), with Hamming distance 4 and zero shared traits. Irreversibility activates Temporal while reversibility is null. Incompleteness activates Symbolic while completeness is null. In every case, the negative form activates different traits than the positive form — never the same traits minus one.

All sixteen entities average 0.75 active traits each. The maximum is independence with three.

Interpretation

UHT classifies each term by its dominant connotative associations, not by its logical relationship to other terms. “Independence” is a political and social concept in UHT’s view — declarations of independence, independence movements. “Dependence” is an ethical concern — substance dependence, dependency relationships. These are not modifications of each other; they are concepts whose cultural weight has drifted in opposite directions despite being logical antonyms.

This asymmetry is precisely the kind of signal UHT can produce that embeddings cannot. In distributional space, dependence and independence are near-neighbors (they co-occur constantly in the same paragraphs). In UHT space, they occupy entirely different trait profiles separated by four bits. The divergence captures real semantic content: the word “independence” carries political valence that “dependence” does not.

The null result on 10 of 16 entities confirms the known trait-sparsity problem for abstract relational concepts. The current 32-trait set was designed for functional, concrete entities. Abstract properties like convergence, continuity, and reversibility fall below the trait set’s resolution. A Valence-Sensitive trait — flagging concepts defined primarily by their position on a polarity axis — could improve this.

Action

Hypothesis HYP-ACTIVEHYPOTHESES-017 closed as refuted. Result recorded as RES-CALIBRATIONRESULTS-017. Observation OBS-STRUCTURALFINDINGS-027 documents asymmetric negation as a structural finding. Trait proposal TRT-PROPOSEDTRAITS-003 proposes a Valence-Sensitive trait, motivated by the inability to represent negation relationships. Research fact stored for the dependence/independence pair. Next session should investigate whether connotative asymmetry holds for concrete concept pairs (where trait sparsity is not a confound) — pairs like construction/destruction or growth/decay where both terms should activate multiple traits.