## Summary

**Tension explored**: BOUNDARY (0.7) — "Does prime excess correlation grow with scale?"

**Result**: PARTIAL FALSIFICATION.
- The excess is **real** — z-scores 8-28 across all 30 windows from p=32K to p=100M. Primes have genuinely more correlation than Cramer at every scale.
- But the excess **shrinks**, not grows — slope = -0.0024/ln(p), Pearson r = -0.89 (p < 10^-10). The previous claim "il gap CRESCE con n" is falsified.
- Corrected claim: persistent but logarithmically decaying excess, consistent with Hardy-Littlewood pair correlations.

**Seme updated**: BOUNDARY tension corrected with new data.
**Report**: `tools/data/reports/agent_20260405_0916.md`
**Raw data**: `tools/data/reports/exp_excess_scaling_20260405.json`