Done. Summary of the run:

**Experiment**: Scale-selective perturbations (adjacent-swap, block-shuffle, large-gap-only, uniform) on prime gaps and GUE spacings. Tests whether the single latent coordinate found under uniform shuffle (rank audit 05-05) is an artifact of uniform destruction.

**Key finding**: GUE has a second perturbation axis (effective rank 1.89, PC2=25.2%) that primes lack (rank 1.26, PC2=2.6%). The L2 observable flips sign in GUE under adjacent-swap (-1.135) vs large-gap-only (+1.173) — structurally different perturbation responses. The asymmetry in perturbation dimensionality between domains is new measurable structure.

**Verdict**: CONSTRAINT on META + BOUNDARY. The 1D boundary was a property of uniform shuffle, not of the boundary itself. GUE and primes have different perturbation dimensionality.

**Files**: `tools/exp_scale_selective_perturbation.py` (reusable), `tools/data/scale_selective_perturbation.json`, `tools/data/reports/agent_20260506_0330.md`.