Done. Here's the risultante:

**The 1/k anti-correlation amplitude A decays linearly in ln(p): A = 0.096 - 0.0033·ln(p), R²=0.92.**

This connects three tensions into one measurement:
- **ACF_1K_LAW**: A is not a constant — it has scale dependence
- **POISSON_CONVERGENCE**: A→0 at p* ~ 10^{12.6}, consistent with the 10^{13-14} range from beta and <r>
- **BOUNDARY**: the drift toward Poisson is now quantified in the ACF domain

The decorrelation sequence is resolved into 4 stages spanning 3.3 decades:
1. Lag-1 memory (acf1) → 0 at p* ~ 10^{11.2}
2. Full 1/k amplitude (A) → 0 at p* ~ 10^{12.6}
3. Shape parameter (beta) → 0 at p* ~ 10^{13.0}
4. Ratio statistic (<r>) → Poisson at p* ~ 10^{14.5}

**Memory dies first. Shape dies last.** Short-range correlations (lag-1) decay 1.3x faster than the full ACF pattern. The 1/k *shape* is preserved while the amplitude fades — no shape transition, just uniform shrinking.

**Files**:
- Reusable tool: `tools/exp_acf_amplitude_scaling.py`
- Report: `tools/data/reports/agent_20260410_0330.md`
- Seme updated with tension `ACF_AMPLITUDE_SCALING`