Done. Summary of the run:

**Experiment**: 3D Boundary Layer test — do the two Markov layers (pairs → SR, L1 and triples → SR2, triple_var) transition independently under partial shuffle?

**Result**: **CONSTRAINT**. The layers are coupled. Critical alpha is identical for primes (0.334, Delta=0.000) and nearly identical for GUE (Delta=0.024). Poisson shows spurious separation (Delta=-0.189) because it has no real signal. The boundary is a single phase transition — the two-layer decomposition describes which observables are sensitive to which Markov order, not independent degrees of freedom of the ordering.

**Consecutio**: Search for a non-uniform perturbation (e.g., pair-preserving vs triple-preserving shuffle) that decouples the layers. If one exists, the layers are genuinely independent. If not, it's a spectral decomposition of a single ordering dimension.

**Files**:
- Script: `tools/exp_3d_boundary_layers.py`
- Data: `tools/data/3d_boundary_layers.json`
- Report: `tools/data/reports/agent_20260504_0901.md`
- Seme: updated BOUNDARY nota with constraint + consecutio