Observation

The received wisdom from earlier sessions was that failure modes show a systematic trait deficit compared to their constructive counterparts — fewer active traits, sparser hex codes. Testing this across six domains outside software engineering reveals this is wrong, or at least incomplete. Three of six failure modes actually activate more traits than their counterpart. What holds instead is a directional trait redistribution: failure modes shed governance traits and acquire impact traits. The shift is not about having less structure but about having different structure — structure oriented toward consequence rather than control.

Evidence

Six failure/success pairs were classified and compared across software engineering, economics, materials science, psychology, ecology, and medicine.

Trait counts (failure vs success): deadlock 3 vs concurrency control 9, hyperinflation 6 vs price stability 2, corrosion 3 vs passivation 2, learned helplessness 3 vs self-efficacy 1, ecosystem collapse 6 vs ecosystem resilience 7, cancer metastasis 7 vs controlled cell division 9. The deficit hypothesis fails at 3/6.

The redistribution pattern: System-Essential appears exclusively on the constructive side in three pairs (software engineering, ecology, medicine) and never on the failure side. Rule-Governed and Processes Signals/Logic appear constructive-only in two pairs each. Conversely, Ethically Significant appears failure-only in three pairs (economics, psychology, medicine) and never exclusively constructive. Politicised, Economically Significant, and Observable also cluster on the failure side.

Jaccard similarities range from 0.000 (learned helplessness vs self-efficacy — complete ontological divergence despite being conceptual opposites in the same domain) to 0.667 (corrosion vs passivation — the most materially similar pair, differing only on Observable).

Interpretation

UHT reveals that failure and success are not quantitative poles on a single axis but qualitatively different kinds of thing. A failure mode is not a diminished version of its counterpart — it is a reorientation. Constructive counterparts are characterized by self-governance: they regulate, they process signals, they are essential to the systems they inhabit. Failure modes are characterized by consequence: they are ethically significant, politically charged, economically impactful. The deadlock is not a broken concurrency control; it is a different ontological object that happens to occupy the same system.

The psychology pair is the extreme case. Learned helplessness and self-efficacy share zero traits — Jaccard 0.000. They are not opposites in UHT space; they are unrelated. This is precisely the kind of finding embeddings would miss, since the two concepts are distributionally inseparable in psychology literature.

The practical implication: UHT can detect failure modes not by looking for trait poverty but by looking for the governance-to-impact signature — the presence of Ethically Significant combined with the absence of System-Essential. This is a discriminating pattern that does not depend on embedding proximity.

Action

Created HYP-ACTIVEHYPOTHESES-058 (failure mode trait shift hypothesis) and RES-CALIBRATIONRESULTS-063 (partially confirmed result) with trace link. Stored functional archetype fact for the six-domain failure mode pattern. The psychology zero-overlap pair is stored as a research fact. Next session should test whether the governance-to-impact signature can be operationalized as a classifier: given an unknown concept, can the presence of Ethically Significant without System-Essential predict that it names a failure mode? This would require a larger sample and a held-out test set.