Near-synonyms crack where true synonyms hold

Observation

UHT can tell near-synonyms apart. True synonyms — automobile/car, physician/doctor, couch/sofa — produce nearly identical hex codes, differing by at most 2 bits. Near-synonyms with genuine ontological boundaries — empathy/sympathy, method/methodology, knowledge/information — diverge sharply, with trait differences that map onto real conceptual distinctions. Seven of eight near-synonym pairs showed nonzero Hamming distance, and the trait diffs aren’t noise: they pick out exactly the dimensions where these concepts genuinely diverge.

Evidence

True synonyms (Hamming distance, Jaccard similarity):

  • automobile/car: 2 bits, Jaccard 0.882 (15 shared traits)
  • physician/doctor: 1 bit, Jaccard 0.923 (12 shared traits)
  • couch/sofa: 1 bit, Jaccard 0.889 (8 shared traits)
  • Mean Hamming: 1.33

Near-synonyms:

  • empathy/sympathy: 4 bits, Jaccard 0.000 — zero shared traits. Empathy activates Signalling + Ethically Significant; sympathy activates Symbolic + Social Construct
  • method/methodology: 5 bits, Jaccard 0.444 — methodology adds Synthetic, Symbolic, Signalling, Normative, Temporal over method’s 4 shared traits
  • knowledge/information: 4 bits, Jaccard 0.333 — knowledge has Meta + Rule-Governed + Compositional; information has Digital/Virtual
  • strategy/tactics: 3 bits, Jaccard 0.625 — strategy adds Symbolic + Normative; tactics adds Rule-Governed
  • hypothesis/theory: 2 bits, Jaccard 0.600
  • weather/climate: 2 bits, Jaccard 0.333
  • efficacy/effectiveness: 1 bit, Jaccard 0.000
  • accuracy/precision: 0 bits — both null-hex (00000000), the only failure
  • Mean Hamming: 2.625

Interpretation

The empathy/sympathy split is the cleanest result: total trait divergence despite apparent semantic overlap. Empathy is classified as a signalling behaviour with ethical weight; sympathy is a symbolic social construct. This mirrors the philosophical distinction — empathy as active felt resonance, sympathy as culturally coded response. Method/methodology shows the same pattern: methodology inherits method’s core traits but layers on five additional ones, capturing the real difference between doing something and theorizing about how to do it. Strategy/tactics splits on Symbolic+Normative vs Rule-Governed — strategy sets direction, tactics follow rules.

The accuracy/precision failure is instructive, not embarrassing. Both are abstract measurement-quality concepts that the 32-trait set cannot distinguish because it lacks traits for quantitative precision, measurement methodology, or statistical properties. This is the same null-hex blind spot affecting 25+ entities — a known trait-set limitation, not a classification error.

Embeddings would cluster all eleven pairs tightly. UHT discriminates where it matters and holds where it should.

Action

HYP-ACTIVEHYPOTHESES-020 confirmed and moved to Closed Hypotheses. Result recorded as RES-CALIBRATIONRESULTS-029 with trace link. Effectiveness added to the null-hex research fact registry. No active hypotheses remain in the queue — the next TRACE_GAP session should generate new ones from entity graph exploration, particularly targeting the null-hex trait-gap pattern as a source of testable claims about what traits would rescue these classifications.

← all entries