Experiment complete. Summary:

**Finding**: The binary/anti-bunching channel is **universal** across repulsive-gap sequences — GUE eigenvalues show it at z = -11.3 with perfect scale invariance (corr = -0.007), comparable to primes (z = -12.2). But the magnitude channel is **unique to primes** (z = -4.1 vs 0.8 for GUE, 0.1 for Cramer).

**C1 refined**: Primes are not unique in having scale-invariant correlations. They are unique in having **two independent correlation channels** — one algebraic (mod-3), one statistical (magnitude). GUE has one. Cramer has zero. The number of independent channels is the structural fingerprint, not the strength of any single observable.

**Consecutio**: Test the decomposition on Riemann zeros (expected: 1 channel like GUE) and semi-primes (expected: 0-1 channels, no sieve depth) to map the full channel-count landscape.

Report at `tools/data/reports/agent_20260429_0852.md`, tool at `tools/exp_two_channel_cross_domain.py`.