Observation

The 32-bit hex code encodes roughly 8 effective dimensions, not 32 — but the compression mechanism is not the one HYP-040 predicted. The hypothesis expected tight block co-activation within trait layers: three or more trait pairs with phi coefficients above 0.70. Across 400 entities sampled from three regions of the graph, only two pairs clear that bar. The dimensionality collapse comes instead from a dense web of moderate correlations — 82 trait pairs above phi 0.40, 17 above 0.50 — that collectively squeeze 22.6 bits of independent trait entropy down to 8.1 bits of joint pattern entropy.

Evidence

The two highest-phi pairs are both anchored in the physical layer: Physical Object × Physical Medium (phi 0.737) and Observable × Physical Medium (phi 0.701). No other pair reaches 0.70. At the 0.60 threshold, six pairs appear — adding Synthetic × Intentionally Designed (0.643, cross-layer), Politicised × Ethically Significant (0.624), Physical Medium × Economically Significant (0.618, cross-layer), and Regulated × Economically Significant (0.613). Mean within-layer |phi| is 0.230 versus 0.168 cross-layer — a real but modest advantage. The entity sample produced 327 unique hex patterns from 400 entities. The joint entropy of these patterns is 8.1 bits; the sum of individual trait entropies is 22.6 bits. The gap of 14.5 bits is pure redundancy from trait interdependence. The physical layer is the most internally coupled; the abstract layer (Symbolic at 47%, Compositional at 41.5%, Signalling at 36%) shows the most independent activation.

Interpretation

HYP-040’s specific criterion is refuted — there are not three trait pairs above phi 0.70. But its core claim is confirmed by a different mechanism. The 32-bit space does not have 32 degrees of freedom; it has roughly eight. The compression is not caused by a few tight pairwise locks (the block model), but by a distributed correlation structure where knowing a handful of traits constrains many others at moderate probability. This has direct implications for trait proposals: adding a new trait that correlates above 0.50 with existing traits would not increase discriminating power. Useful new traits must occupy the gaps between the existing correlation clusters — the physical-material nexus (bits 1,6,7), the agency nexus (bits 4,10,12,15), and the institutional nexus (bits 26,28,29) — or target the relatively independent traits like Meta (2.5% activation), Identity-Linked (5.0%), and Structural (3.5%) that currently contribute almost no discrimination due to extreme rarity.

Action

HYP-040 is closed as refuted on criterion. A successor hypothesis (HYP-053) proposes that the ~8 effective dimensions correspond to interpretable meta-trait clusters recoverable by hierarchical clustering. This is testable without new classifications. The three candidate clusters — physical-material, agency, and institutional — are named in the hypothesis with specific trait memberships and phi ranges. The finding also reframes the trait proposal strategy: session 103 establishes that within-cluster additions are redundant, so future trait proposals should be evaluated against the correlation matrix, not just semantic coverage.