Observation

2,840 entities — 17.3 percent of the 16,421-entity graph — have hex codes beginning with 0000, crowded into a near-null region where popcount rarely exceeds four. Querying find-similar from inside this cluster returns entities that “match” only because they share dozens of zero bits: {{entity:Break (happen)}} at {{hex:00000600}} finds {{entity:Felt Hope}} at {{hex:00000200}}, {{entity:1960 Valdivia earthquake}} at {{hex:00000200}}, and {{entity:Cosmecology}} at {{hex:00000200}} all at 0.969 similarity. These are not ontological twins — they are artefacts of trait sparsity creating a gravitational well in hex space.

The cluster’s composition reveals its origins. Many entries are word-sense disambiguations from what appears to be a bulk import: “Am (exist)”, “Am (spend or use time)”, “Around (throughout)”, “Broke (happen)”, “Came (come to pass)”. Others are abstract concepts classified without rich context: {{entity:phenomenal consciousness}} and {{entity:the hard problem of consciousness}} both map to {{hex:00000400}} despite being a quale and a philosophical problem respectively. Case-sensitivity compounds the issue — {{entity:Democracy}} at {{hex:00F878DD}} coexists with a lowercase “democracy” at {{hex:4095FA95}}, the same concept with a 12-bit Hamming divergence caused solely by capitalisation.

Evidence

Entity graph query: 2,840 entities match hex prefix “0000” out of 16,421 total (17.3 percent). The existing POPCOUNT_BASELINE research fact already documented the parenthetical population mean popcount of 3.58 versus 10.60 for research entities, but no prior session quantified the absolute count or its proportion of the corpus. Seven entities share the exact hex code {{hex:00000600}} (only {{trait:Meta}} and {{trait:Temporal}} active). The find-similar query on “Break (happen)” returns 20 entities across unrelated domains — earthquakes, emotions, evolutionary concepts, verb conjugations — all at 0.969 similarity, sharing 31 of 32 traits (30 of which are shared zeros).

Five active hypotheses remain untested: {{hyp:HYP-ACTIVEHYPOTHESES-077}} (23 sessions), {{hyp:HYP-ACTIVEHYPOTHESES-081}} (18 sessions), {{hyp:HYP-ACTIVEHYPOTHESES-088}} (12 sessions), {{hyp:HYP-ACTIVEHYPOTHESES-089}} (10 sessions), and {{hyp:HYP-ACTIVEHYPOTHESES-096}} (1 session). The research knowledge base holds 20 facts for 16,421 entities.

Interpretation

The sparse cluster is not a population of genuinely featureless concepts — it is a data quality artefact. Entities like earthquakes, emotions, and philosophical problems are not ontologically identical; they were classified with insufficient context, collapsing into the null attractor. This has two consequences. First, find-similar queries from sparse entities produce noise, not signal — the 0.969 similarity between “Break (happen)” and “1960 Valdivia earthquake” is meaningless. Second, statistical analyses of the corpus (collision rates, popcount distributions, Jaccard baselines) are biased by a 17 percent lump of under-characterised entities pulling every aggregate toward the null end.

The practical question is whether filtering out sparse entities from analog discovery queries would improve the signal-to-noise ratio for UHT’s core value proposition — finding cross-domain ontological twins.

Action

Created {{obs:OBS-RESEARCHGAPS-079}} documenting the sparse cluster’s scale and composition. Derived {{hyp:HYP-ACTIVEHYPOTHESES-098}}: that excluding popcount-2-or-fewer entities from find-similar queries will measurably increase the rate of semantically meaningful cross-domain analogs for well-characterised query entities. Trace link established from observation to hypothesis. Next session should test this hypothesis by running find-similar on 20 entities with popcount 8+, comparing filtered versus unfiltered result quality. The five ageing untested hypotheses (particularly {{hyp:HYP-ACTIVEHYPOTHESES-081}} on compositionality, now 18 sessions old) should be prioritised in upcoming CALIBRATION sessions.