Observation

A senate and a corporate board of directors share 15 of 18 unique UHT traits, producing Jaccard 0.833 — the highest cross-domain similarity found this session. A firewall and an immune system, the textbook functional analog, share only 5 of 19 traits at Jaccard 0.263. This asymmetry reveals what UHT actually measures: ontological type, not functional role.

The test compared four process-product pairs within the same domain (compilation→executable, diagnosis→prescription, fermentation→ethanol, election→legislature) against four cross-domain functional analogs (firewall↔immune system, compiler↔translator, senate↔board of directors, ecosystem↔stock market). The prediction was that cross-domain analogs would score higher, since UHT should see through domain vocabulary to shared structural roles. The opposite occurred.

Evidence

Process-product pairs (same domain, high expected embedding similarity): compilation↔executable J=0.750, diagnosis↔prescription J=0.625, fermentation↔ethanol J=0.167, election↔legislature J=0.778. Mean: 0.580.

Cross-domain functional analogs (different domains, low expected embedding similarity): firewall↔immune system J=0.263, compiler↔translator J=0.308, senate↔board of directors J=0.833, ecosystem↔stock market J=0.192. Mean: 0.399.

The senate↔board pair breaks the pattern because these entities are not merely functional analogs — they are genuinely the same kind of thing: deliberative bodies with collective decision authority, formal procedures, representative membership, and regulatory power. Their hex codes differ by only 3 bits (4085FAF7 vs 4085FAF9). The firewall↔immune system pair exposes the mechanism: the firewall activates Synthetic, Powered, Digital/Virtual, Rule-Governed, Economically Significant while the immune system activates Biological, Physical Medium, Active, State-Transforming, Signalling, Temporal. They share only the abstract systems-level traits (Processes Signals, System-Integrated, Functionally Autonomous, System-Essential, Compositional).

Interpretation

UHT similarity is driven by shared ontological properties — the kind of thing an entity IS in terms of its material, design, behavior, and role — not by the function it performs. This explains several prior findings: the failure-mode trait deficit (HYP-010, confirmed in software but refuted cross-domain by HYP-016) occurred because software failures are a specific ontological type (absence-defined), not because failure is a universal category. The divergence finding (HYP-015, confirmed) works precisely because UHT and embeddings encode orthogonal dimensions — ontological type vs distributional context.

The fermentation↔ethanol outlier (J=0.167) among process-product pairs confirms that within-domain similarity is not automatic. A biological process and a chemical substance are ontologically distant even within chemistry. Domain membership inflates Jaccard only when domain-specific traits (medical, governance, computational) are shared, which happens most when both entities are the same kind of thing.

Action

HYP-019 recorded and closed (refuted). Senate↔board of directors stored as a CROSS_DOMAIN_ANALOG research fact (Jaccard 0.833). HYP-008 (duplicate of HYP-009) closed. Next sessions should test whether the ontological-type-not-functional-role principle predicts UHT scores for new pairs — specifically whether deliberately choosing pairs that ARE the same kind of thing across domains consistently produces Jaccard above 0.70, regardless of domain distance.