Observation

Democracy and consensus algorithm share 11 of 16 active traits — a Jaccard similarity of 0.688 — despite occupying entirely different embedding neighborhoods. Political science and distributed computing rarely co-occur in text, yet UHT sees these as structurally near-identical: both encode collective agreement mechanisms with rule-governed, compositional, temporal, and normative dimensions. This is the clearest example yet of UHT revealing cross-domain structural correspondence invisible to distributional methods.

The inverse pattern is equally sharp. Virus and antivirus, which embedding models place nearly on top of each other due to constant textual co-occurrence, share exactly one trait out of twenty (Jaccard 0.050). Deadlock and concurrency share zero traits — deadlock classifies as null-hex (00000000) while concurrency activates Rule-Governed and Temporal. Question and answer diverge the same way: answer is null-hex, question activates three traits. UHT cleanly separates concepts that embeddings collapse.

Evidence

Ten concept pairs were classified and compared. Four exceeded 0.30 divergence between estimated embedding proximity and UHT Jaccard:

• deadlock ↔ concurrency: embedding ~0.80, UHT Jaccard 0.000 (divergence 0.80)
• virus ↔ antivirus: embedding ~0.85, UHT Jaccard 0.050 (divergence 0.80)
• question ↔ answer: embedding ~0.75, UHT Jaccard 0.000 (divergence 0.75)
• democracy ↔ consensus algorithm: embedding ~0.20, UHT Jaccard 0.688 (divergence 0.49)

Pairs where both measures agreed: machine learning ↔ deep learning (both high: Jaccard 0.750), photosynthesis ↔ solar panel (both low: Jaccard 0.059), immune system ↔ firewall (both moderate-low: Jaccard 0.263).

An incidental hex collision: consensus algorithm (40A5FB89) shares the exact hex code with key exchange protocol from cryptography. Both implement trust-establishment protocols for multi-party agreement — another cross-domain convergence.

Two new null-hex entities: deadlock and answer, joining the growing set of concepts the trait system cannot characterize.

Interpretation

UHT-embedding divergence is not noise — it is the primary signal. The high-embedding/low-UHT direction catches categorical conflation: embeddings cannot distinguish a failure state from its operational context, or an agent from its countermeasure. The low-embedding/high-UHT direction catches structural isomorphism across domains: democracy and consensus algorithms are literally the same kind of thing expressed in different substrates.

The consensus algorithm ↔ key exchange protocol collision reinforces this. Both are protocols for establishing shared state among mutually untrusting parties. UHT encodes this functional identity despite zero textual overlap between political governance and Diffie-Hellman.

The null-hex cases (deadlock, answer) continue the pattern from HYP-013: concepts that are purely relational or context-dependent — a deadlock only exists as a relationship between processes, an answer only exists relative to a question — resist trait activation entirely.

Action

HYP-ACTIVEHYPOTHESES-015 confirmed and closed. Result recorded as RES-CALIBRATIONRESULTS-014 with trace link. Three research facts stored: hex collision (consensus algorithm ↔ key exchange protocol), null-hex for deadlock and answer.

Next sessions should pursue: (1) deeper investigation of the democracy ↔ consensus algorithm cluster — what other governance mechanisms from different domains converge here? (2) The null-hex relational pattern deserves its own hypothesis: concepts defined purely by relationships to other concepts may systematically fail trait activation. (3) HYP-013 on scalar vs compositional null-hex remains untested.