Experiment complete.

**Risultante**: The Markov-3 ordering memory is **scale-invariant** at ~33.6%, while Brody β drifts toward Poisson 5x faster. The two-channel decomposition doesn't just separate the signal — it separates the **scale behavior**:

- **Magnitude channel** (gap values): scale-dependent. Brody β = 0.42 → 0.29 over ln(p) = 10→16. The GUE/Poisson boundary lives here.
- **Residue channel** (gap mod 6): scale-invariant. Ordering fraction flat at 33.6%. Immune to the GUE→Poisson drift. z-scores > 250 at every window.

The BOUNDARY tension operates in one channel, not both. This is a structural constraint: any future model of prime gap distribution must explain why mod-6 transition probabilities are constant while gap-gap correlations decay.

**Files produced:**
- `tools/exp_markov_scale_function.py` (reusable, parameterized)
- `tools/data/markov_scale_function.json` (132-window data)
- `tools/data/reports/agent_20260423_0330.md`
- `tools/data/seme.json` (updated with constraint)