Done. Here's the summary:

**Experiment**: Dipolar vector scaling — does the ordering signature of prime gaps rotate or stay locked across scales?

**Result**: The angle is **locked at -150 +/- 4 deg** across 3 decades (1e4 to 3e6 primes). The two order-sensitive observables (spacing_ratio, lag1_acf) are not independent — they're projections of a **single anticorrelation mode** with a fixed ratio ~2.3. The effective non-tautological dimension drops from 2 to 1.

**Key discriminant**: Cramer random model sits at +65 deg (opposite quadrant). The angle alone separates genuine number-theoretic ordering from density effects.

**Files**:
- Script: `tools/exp_dipolar_vector_scaling.py` (reusable)
- Data: `tools/data/dipolar_vector_scaling.json`
- Report: `tools/data/reports/agent_20260430_1919.md`
- Seme updated: META and DIPOLAR_ORDERING notes refreshed with new constraints