All tasks complete. Here's the summary:

---

**Experiment: ACF Stationarity — Dipolar vs Illusory Duality in Prime Gap Correlations**

Starting from a dissonance in today's nightly data (only 6/50 ACF lags negative, contradicting ACF_1K_LAW's "all 20 negative"), I tested whether PNT-normalization (dividing gaps by ln(p)) recovers the all-negative pattern.

**Results:**
- **PNT-normalization recovers 45/50 negative ACF lags** (vs 6/50 raw). Alpha = 0.990, confirming the 1/k law on stationary data.
- **Crossover lag 6** is where the PNT trend (+0.008/lag) overtakes the structural anti-correlation (-0.007/lag). This is the quantitative boundary between dipolar duality (det=-1, genuine) and illusory duality (det=+1, trend artifact).
- **70% of the anti-correlation is hidden** in raw gap analysis. Total ACF sum flips from +0.17 (raw) to -0.22 (normalized).
- **Alpha drifts from 0.9 to 0.6** at larger primes — the decorrelation shape changes, not just the amplitude.

**Files created:**
- `tools/exp_acf_stationarity.py` — reusable tool
- `tools/data/exp_acf_stationarity.json` — full results
- `tools/data/reports/agent_20260417_0803.md` — report
- `tools/data/seme.json` — updated with DIPOLAR_ILLUSORY_BOUNDARY tension