Experiment complete. Summary of what was found:

**The magnitude PSD is Wiener-Khinchin self-consistent and pair-dominated.**

| Finding | Value | Significance |
|---|---|---|
| ACF→PSD capture | 99.0% | Full closure |
| K* (convergence depth) | 2 lags | Pairs suffice |
| Magnitude ACF alpha | 2.09 | Fast decay, NOT 1/k |
| Lag-1 contribution | 76% of slope | Primary carrier |
| Lag-2 contribution | 17% of slope | Secondary memory |
| 1/k model error | +30% overprediction | Wrong model for magnitude |
| z vs shuffle | 46.3 | Strongly significant |

**The key reinterpretation**: The 1/k law (ACF_1K_LAW, alpha~1.0) was a mixture-level phenomenon. After decomposition: residue decays slowly (alpha=1.24, algebraic), magnitude decays fast (alpha=2.1, pair-dominated). The composite appears as alpha~1.0.

**Two-channel framework is now spectrally complete**:
- Residue: 100% algebraic (order-3 Markov)
- Magnitude: WK self-consistent, K*=2 (pair correlations)

**Consecutio**: derive acf_mag(1) = -0.032 analytically from Hardy-Littlewood pair correlations. If derivable, the entire PSD of prime gaps is explained from first principles (Markov algebra + HL pairs).

Files saved:
- Script: `tools/exp_magnitude_psd_from_acf.py`
- Data: `tools/data/magnitude_psd_from_acf.json`
- Report: `tools/data/reports/agent_20260422_0330.md`
- Seme: updated with findings + new consecutio