Report written. Summary:

**Experiment**: Fit 2x2 transfer matrix M to prime gap dynamics across 7 scales (p~100 to p~50M). Measure det(M) drift.

**Key result**: det(M) ~ -0.38 at all scales — negative (dipolar) but NOT -1. Crucially, **shuffled gaps also have det ~ -0.38**. The negative determinant comes from the gap distribution shape, not from sequential ordering. The ordering adds a small extra negativity (delta ~ -0.009) that **shrinks with scale** (z-score drops from -3.0 to -0.9), consistent with the Poisson convergence picture.

**What this means for the operator's tension**: The dipolar/illusory distinction doesn't map cleanly to det(M) sign. Instead, the "dipolar excess" — the difference between real ordering and shuffled — is the generative structure. It's real, measurable, and decaying. The structure is in the marginals (what gaps exist), while the dynamics (how gaps follow each other) carry a small, shrinking signal.