Observation

The 50+ null-hex entities in the UHT corpus are not a single population. They split cleanly into two groups when force-refreshed with rich context descriptions: property-bearing entities reclassify immediately to non-null codes, while pure-state concepts crash the classifier entirely. The most surprising result was that “presence (state of existing)” produces the same 500 error as “absence (state of not existing)” — the dividing line is not positive versus negative, but whether a concept possesses functional, physical, or social properties at all.

Evidence

Twelve property-bearing entities previously at 00000000 all reclassified on force-refresh: money at C48DFADD (19 traits), 3D Printer at D7FC3008 (15), language at 406DFE80 (14), coral reef at AE171201 (12), Volcano at 8E500000 (6), permanent magnet at C6800008 (6), justice at 0000A885 (6), knowledge at 00311600 (6), magnetism at 04412000 (4), continental drift at 00003200 (3), noise at 04400000 (2), invariant at 00002000 (1). Nine pure-state concepts — nothing, silence, absence, calm, ignorance, stability, harmony, disorder, and presence — all returned 500 errors on force-refresh while retaining 00000000 in cache. Dark energy reclassified to near-null (00000200, 1 trait: Temporal). Depth (abstract quality) also errored, confirming the pattern extends to pure dimensional qualities. Of the 44 remaining null-hex entities, 6 are test artifacts, leaving approximately 38 genuine cases.

Interpretation

UHT’s 32-trait system is structurally incapable of evaluating pure states and conditions. Each trait probe asks “does X have property Y?” — concepts like silence, stability, calm, and presence don’t have properties in this sense. They describe conditions under which properties are absent, balanced, or simply existing. The classifier doesn’t return 32 false answers; it errors entirely, suggesting the LLM evaluation itself cannot produce a coherent response when forced to assess properties of a property-less concept. The prior null-hex population was inflated by stale cache from pre-v4.1 context-less classifications. Force-refresh is an effective diagnostic: entities that reclassify were cache artifacts; entities that error are genuinely outside the trait system’s domain. Money’s reclassification to 19 traits — the densest profile observed — demonstrates how dramatically context enrichment changes classification quality for institutionally complex entities.

Action

HYP-ACTIVEHYPOTHESES-078 confirmed and linked to RES-CALIBRATIONRESULTS-092. Six classification drift facts stored. New functional archetype “pure-state-null-hex” recorded with the nine identified entities. The remaining ~38 genuine null-hex entities should be audited: which are pure states (correctly unclassifiable), which are cache artifacts awaiting force-refresh, and which might motivate a trait extension for states and conditions? A future trait proposal could add a “State/Condition” trait to capture these concepts. HYP-011 (null-hex entities are process-dependent/context-embedded) should be revisited in light of this finding — the answer is more nuanced than context-dependence alone.