The experiment is complete. Here's the summary:

## Results

**The dipolar phase transition is nearly universal (5/7 ordered sequences), but the locked direction is diagnostic.**

Key findings:
1. **Phase transition = method property** — any ordered sequence under partial shuffle shows direction lock + linear magnitude decay + zero-crossing flip. This means the previous report's "phase transition exists" claim is partially tautological.

2. **Direction = ordering fingerprint** — each sequence type locks at a distinct angle: GUE (-97.8°), AR1 (-94.7°), Primes (-104.2°), Periodic 2,4,2,4 (-104.0°), Logistic (+110.4°). Two classes: "repulsion" (~-97°) vs "confinement" (~-104°).

3. **Primes = Z/6Z confinement** �� the prime direction matches the periodic 2,4,2,4 pattern to 0.2° (0.17 sigma). The ordering CHARACTER of primes in the (SR, L1) plane IS the mod-6 constraint (F2). This geometrically confirms Lemke Oliver-Soundararajan: pair statistics (= Z/6Z structure) explain the direction.

4. **The Markov-3 residual (z=6203) is invisible in (SR, L1)** — since primes and periodic-24 are indistinguishable in this projection, the higher-order correlations must live in a different observable dimension.

**Verdict**: CONSTRAINT on BOUNDARY (phase transition is method, not discovery) + CONFIRMED on DIPOLAR_ORDERING (direction is diagnostic, primes = confinement class).