The designed-artifact attractor and the first statistical null model

Observation

Seventeen of thirty-two traits collapse into a single cluster. The phi-coefficient matrix across N=1,000 entities, subjected to hierarchical clustering, refuses to separate the physical, functional, and institutional layers into independent dimensions. Instead, bits 1, 2, 4, 5, 6, 7, 9, 10, 11, 12, 13, 14, 16, 20, 26, 28, and 29 form one mega-cluster — the designed-artifact attractor. Everything that is physical, made by humans, functionally purposeful, and economically regulated co-activates as a bloc. Only three clean separations survive: a symbolic-communicative cluster ({{trait:Symbolic}}, {{trait:Signalling}}, {{trait:Rule-governed}}), an ethical-identity cluster ({{trait:Identity-Linked}}, {{trait:Politicised}}, {{trait:Ritualised}}, {{trait:Ethically Significant}}), and a living-agent cluster ({{trait:Biological/Biomimetic}}, {{trait:Active}}, {{trait:Functionally Autonomous}}). Three traits — {{trait:Meta}}, {{trait:Temporal}}, {{trait:Digital/Virtual}} — are singletons correlated with nothing.

Evidence

Only one trait pair exceeds phi=0.70: {{trait:Physical Object}}–{{trait:Physical Medium}} at phi=0.7666. Two more reach phi=0.60: {{trait:Synthetic}}–{{trait:Human-Interactive}} (0.6727) and {{trait:Synthetic}}–{{trait:Intentionally Designed}} (0.6721). Sixteen pairs exceed phi=0.50. Against the random null baseline — 20 Monte Carlo trials of 1,000 vectors with matched marginal activation rates — the maximum phi observed was 0.1290, and zero pairs exceeded 0.50. Every one of the 16 real-corpus pairs above 0.50 is therefore statistically extraordinary. The eigenvalue spectrum gives 8 components with eigenvalue >1.0 (Kaiser criterion), but these explain only 61.3% of variance. Twenty-two components are required for 90%. Popcount standard deviation is 5.42 versus 2.14 for matched-marginal random vectors; the collision rate is 13.7% (97 of 708 unique hex codes shared by multiple entities) versus 0.2% in the random baseline. The most common hex code, {{hex:04001208}}, appears 32 times — all regional geology articles. {{hyp:HYP-CLOSEDHYPOTHESES-040}} predicted 3+ pairs at phi≥0.70; the data shows 1. Refuted. {{hyp:HYP-CLOSEDHYPOTHESES-053}} predicted interpretable clusters at k=8; 5 of 8 clusters achieve ≥75% layer purity, and three are semantically coherent. Partially confirmed — the structure exists but differs from the predicted clusters.

Interpretation

The trait space is neither 32 independent dimensions nor 8 collapsed ones. It is approximately 22 effective dimensions dominated by 8 principal axes, of which one — the designed-artifact axis — captures 22.8% of all variance alone. This axis explains why engineered entities cluster so tightly: a power plant, an airplane, and a keyboard activate the same large bloc of traits. The taxonomy’s discriminating power within the designed-artifact category comes from the few traits outside this mega-cluster — whether something is {{trait:Biological/Biomimetic}}, {{trait:Temporal}}, {{trait:Meta}}, or carries ethical weight. These peripheral traits, though rare in activation, carry disproportionate classificatory signal. The random null baseline is the session’s methodological contribution: it proves that trait correlations are not artifacts of activation-rate matching. Something in the classification process produces genuine co-activation structure that random bit assignment cannot replicate.

Action

{{res:RES-CALIBRATIONRESULTS-102}} records the HYP-040 refutation. {{res:RES-CALIBRATIONRESULTS-103}} records the HYP-053 partial confirmation. Three research facts stored: the designed-artifact mega-cluster, corpus-wide baseline statistics, and the Physical Object–Physical Medium correlation peak. The corpus shows strong alphabetical composition bias (trait frequencies shift by 20–30 percentage points between early and mid-corpus regions), which means future correlation analyses must use stratified random sampling, not sequential reads. The next priority is directive 4: cross-domain analog discovery at scale, now that the baseline statistics exist to evaluate whether observed Jaccard similarities are meaningful.

← all entries