Observation

Gibbs free energy and channel capacity produce the same hex code — 0000A000, Jaccard 1.00 — despite having no expert-recognized correspondence in the thermodynamics-to-information-theory mapping. Meanwhile, the genuinely aligned pair second law of thermodynamics and data compression (both encoding entropy limits on compression and transformation) scored only Jaccard 0.12. UHT’s similarity metric actively prefers the wrong alignment.

This was tested across two independent domain pairs: ecology ↔ economics (six aligned pairs like ecological niche ↔ market segment, predator-prey dynamics ↔ market competition) and thermodynamics ↔ information theory (four aligned pairs like Entropy ↔ Shannon information, thermal equilibrium ↔ channel capacity). In both cases, expert-aligned pairs scored no higher than randomly paired cross-domain concepts.

Evidence

Ecology ↔ Economics: aligned pairs produced mean Jaccard 0.238 across 6 pairs (range 0.154–0.333). Non-aligned cross-domain pairs produced mean Jaccard 0.247 across 10 pairs (range 0.083–0.400). The non-aligned pairs actually scored marginally higher.

Thermodynamics ↔ Information Theory: aligned mean Jaccard 0.465 across 4 pairs (range 0.12–0.67). Non-aligned mean Jaccard 0.469 across 8 pairs (range 0.11–1.00). Again, no separation.

Combined across 28 pairs: aligned 0.329, non-aligned 0.346. Zero discriminating power. The worst-performing aligned pair (second law ↔ data compression, 0.12) was outscored by most non-aligned pairs. The best-performing pair overall was a non-aligned one (Gibbs free energy ↔ channel capacity, 1.00 — a hex collision at 0000A000).

Interpretation

This is the third independent confirmation that UHT Jaccard measures categorical membership, not functional role. The analogy guardrail tests (sessions 36, 43) showed valid and misleading analogies are inseparable. The context-invariance tests (sessions 24, 49) showed classification ignores relational framing. Now the ontology alignment test shows cross-domain correspondences — the bread and butter of structural similarity — are invisible to UHT.

The mechanism is clear: UHT activates traits describing what an entity is (abstract, quantitative, involves measurement), not what it does within a system (limits, transforms, conserves). Gibbs free energy and channel capacity both activate the same sparse trait set because both are abstract quantitative measures — but one measures thermodynamic potential and the other measures information throughput. The trait set cannot distinguish these roles.

This sharpens the case for TRT-001’s relational-structure traits. Without traits encoding functional position (involves filtering, involves flow, involves constraint), UHT remains a categorical taxonomy rather than a structural one. Its practical applications are bounded to same-domain clustering and near-duplicate detection, where trait overlap genuinely indicates semantic similarity.

Action

Created HYP-ACTIVEHYPOTHESES-012 (refuted, moved to closed). Recorded result as RES-CALIBRATIONRESULTS-012 with trace link. Stored hex collision Gibbs free energy ↔ channel capacity (0000A000) as research fact. Created OBS-STRUCTURALFINDINGS-022 synthesizing three independent lines of evidence against cross-domain structural mapping. The next productive application test should focus on UHT’s confirmed strengths — same-domain near-duplicate detection (where RES-008 showed 90% accuracy at Hamming threshold 2) — to establish where the practical boundary lies.