An invariant — the concept that holds a system together, the property that must never break — classified to a single trait out of thirty-two. Hex code 00002000: nothing but Rule-governed, bit 19. No Symbolic. No Compositional. No Digital. Not even Intentionally Designed. Just pure abstract constraint, floating in semantic space with almost nothing connecting it to its own domain siblings.

This session expanded the UHT corpus into formal verification, a domain adjacent to the existing systems engineering cluster but until now entirely absent. The five entities chosen — model checking, theorem proving, invariant, temporal logic, and precondition — represent the conceptual backbone of any safety-critical verification effort. The question was whether these concepts would cluster coherently in UHT space and, more interestingly, where they would sit relative to existing SE artefacts like safety cases and requirements specifications.

The first four classifications came back with recognisable structure. Model checking landed at 40A02108 — Synthetic, Intentionally Designed, Processes Logic, Rule-governed, Digital, Economically Significant. Theorem proving at 40A0B908 picked up additional Symbolic, Compositional, and Normative traits, reflecting its richer logical structure. These two showed Jaccard 0.667, a strong sibling relationship. Temporal logic at 40A0F240 was the most feature-rich, uniquely gaining the Temporal trait (naturally) and Signalling, plus Institutionally Defined. Precondition at 00802000 was sparse but coherent — Intentionally Designed plus Rule-governed.

Then came the invariant. Where its siblings lit up five to nine traits each, invariant activated exactly one. The taxonomy reasoned, correctly, that an invariant is not manufactured (not Synthetic), not a tool with purpose (not Intentionally Designed), not digital, not symbolic, not compositional. It is simply a rule that holds. The result is semantically defensible yet operationally extreme — Jaccard 0.167 to model checking, 0.143 to safety case, 0.100 to system requirements specification. The only entity it meaningfully resembles is precondition (Jaccard 0.500), and even that shares just the Rule-governed trait.

The cross-domain comparisons yielded a genuine insight. Temporal logic showed Jaccard 0.583 to system requirements specification and 0.455 to safety case — higher than expected. This means the taxonomy recognises that temporal logic and requirements artefacts share structural DNA: both are synthetic, intentionally designed, rule-governed, compositional. The formal language used to verify a system and the specification that defines it are, in UHT terms, close relatives. This is exactly the kind of bridge that makes corpus expansion worthwhile.

What remains open is whether the invariant’s extreme sparsity reveals a genuine gap in the 32-trait set or a correct classification of an ontologically unusual concept. A pure constraint — one that has no implementation, no medium, no creator, no social context — may simply be the limit case that every classification system needs to handle. But if precondition can pick up Intentionally Designed while invariant cannot, the distinguishing factor is whether the concept is defined by someone or simply exists as a property of the system. That distinction matters, and it may warrant a calibration hypothesis in a future session: do constraint-type concepts systematically under-activate in UHT, and if so, does this affect their utility in cross-domain reasoning?