Observation

A database and a compiler are Jaccard 0.800 — nearly identical in UHT space — despite being functionally opposite: one stores, the other transforms. Meanwhile, a database and a warehouse share the container archetype yet score only 0.333. This asymmetry holds across every combination tested. UHT organizes entities by what kind of thing they are, not by what functional role they play.

Evidence

Three structural-archetype groups (container, filter, transformer) each contained five entities drawn from five different domains. Their mean pairwise Hamming distances were 11.40, 11.60, and 12.80 — overall 11.93. Three domain groups (software, biology, physical/built) each contained five entities with different functional roles. Their mean pairwise Hamming distances were 6.20, 6.40, and 6.80 — overall 6.47. Domain clustering was tighter than archetype clustering in all nine comparisons.

Jaccard verification told the same story. Database versus compiler (same domain, different roles): 0.800, sharing Synthetic, Intentionally Designed, Processes Signals/Logic, System-Integrated, Rule-Governed, Compositional, Digital/Virtual, and Economically Significant. The only two bits that differ are State-Transforming and System-Essential. Database versus Storage Warehouse (same archetype, different domains): 0.333, sharing only Synthetic, Intentionally Designed, System-Integrated, Compositional, and Economically Significant.

Biology entities showed the same pattern: Human Stomach, Kidney, Digestive Enzyme, and Hemoglobin cluster within Hamming 2–5, despite serving roles from container to filter to transformer to carrier.

Interpretation

Of the 32 UHT traits, roughly eight to ten encode ontological category — Physical Object, Synthetic, Biological, Digital/Virtual, Observable, and their downstream correlates. These traits fire together for everything in a domain, producing a domain-level signature that dominates the hex code. Only two to three traits (State-Transforming, Processes Signals/Logic, System-Essential) distinguish functional roles within a domain, and even these are shared by most software entities regardless of role. The trait vocabulary is a categorical taxonomy, not a functional one. Cross-domain functional archetypes — containers, filters, transformers — cannot be detected because no trait in the current set captures containment, selective passage, or input-output transformation as distinct properties.

This explains why prior sessions found high cross-domain Jaccard scores for pairs that happened to share domain-like properties (senate and board of directors at 0.833 both being institutional governance) but not for pairs sharing only functional role.

Action

HYP-023 is closed as refuted. RES-CALIBRATIONRESULTS-023 records the data. A trait proposal (TRT-PROPOSEDTRAITS-004) recommends three functional-role traits: Containment, Selective Passage, and Input-Output Transformation. If adopted, these would make cross-domain archetype clustering possible — but would require reclassifying all 1,707 entities. The next session should test HYP-022 (null-hex structural atomicity), which remains the oldest untested active hypothesis.