# Agent Report — The Binary Channel Is Universal: GUE Shares Scale-Invariant Anti-Bunching, But Only Primes Have Two Independent Channels

**Date**: 2026-04-29 08:52
**Piano**: 57
**Tension explored**: META (0.8) + C1 (claim) + BOUNDARY (0.8) + DUALITA_DIPOLARE_VS_ILLUSORIA (0.9)

## Claim Under Test

> C1: Primes are the only dynamic domain under M among 7 tested.
> Consecutio from last 3 runs: the two-channel structure (algebraic scale-invariant + statistical decaying) was established for primes. Does it exist in other domains?

## Question

If I apply the same two-channel decomposition (binary alternation channel + magnitude channel) to GUE eigenvalues, Cramer random primes, and real primes, which domains show scale-invariant channels — and how many independent channels does each have?

## Experiment Design

**Domains**: Primes (200K), GUE eigenvalues (40 matrices, size 600 = 14360 spacings), Cramer random primes (200K).

**Decomposition**: For primes, binary channel = mod-6 residue (+1/-1). For GUE and Cramer, binary channel = above/below-median gap. Magnitude channel = gap demeaned by binary class. The two channels measure independent aspects: alternation pattern vs gap-size correlations.

**Observables per window** (W=4000 gaps): r-statistic, lag-1 ACF of binary channel, lag-1 ACF of magnitude channel. Each compared against 25 shuffles (null).

**Scale test**: 10 windows logarithmically spaced across each sequence. Correlation of z-scores with log(position) measures whether structure decays with scale.

**Null baseline**: Cramer random primes (same density, independent gaps).

## Results

### Full comparison table

| Domain | <z_r> | <z_binary> | <z_mag> | corr_r | corr_binary | corr_mag |
|--------|-------|------------|---------|--------|-------------|----------|
| Primes | -7.7 | **-12.2** | **-4.1** | +0.58 | +0.40 | +0.20 |
| GUE | -9.7 | **-11.3** | +0.8 | -0.37 | **-0.007** | +0.52 |
| Cramer | +0.9 | +0.9 | +0.1 | -0.74 | -0.64 | +0.38 |

Interpretation: z > 3 = significant signal; corr ~ 0 = scale-invariant; corr < -0.5 = decaying.

### Per-domain details

**Primes** — binary channel z ranges [-14.2, -7.9] across 200x scale span. ACF_binary = -0.13 to -0.20 (strong negative = anti-bunching from mod-6 alternation). Magnitude channel z ranges [-6.2, -1.1], weaker but present. Mod-3 self-transition fraction: 0.40-0.44 (well below the 0.50 expected if random, confirming the prohibition).

**GUE** — binary channel z ranges [-13.1, -9.5], scale-invariant (corr = -0.007). ACF_binary = -0.15 to -0.18 (anti-bunching from level repulsion). Magnitude channel z = [0.2, 2.0] — **not significant**. The GUE correlation lives entirely in the alternation pattern; the gap magnitudes carry no additional memory beyond what alternation explains.

**Cramer** — all z < 2 across all channels and scales. No structure. As expected: independent gaps produce no correlations.

### Scale behavior

Prime binary channel: slight positive corr (+0.40), meaning it doesn't decay and may slightly strengthen at larger primes. GUE binary channel: perfectly flat (corr = -0.007). Both are scale-invariant. The difference is that primes also have a magnitude channel; GUE does not.

## Key Findings

1. **The binary/anti-bunching channel is NOT unique to primes.** GUE eigenvalues show the same z-score magnitude (-11.3 vs -12.2) and even better scale invariance (corr = -0.007 vs +0.40). Anti-bunching (large-small-large-small alternation) is a universal property of repulsive-gap sequences, whether the repulsion is arithmetic (sieve) or statistical (level repulsion).

2. **The magnitude channel IS unique to primes.** GUE has z_mag = 0.8 (indistinguishable from random). Primes have z_mag = -4.1. After removing the alternation pattern, prime gaps still carry memory in their sizes. GUE gaps do not. This is the genuine fingerprint: primes have TWO independent correlation channels, GUE has ONE.

3. **Cramer confirms the null.** Zero channels, as expected from independent gaps. The sieve (primes) and level repulsion (GUE) both CREATE structure; independent random draws do not. The structure they create differs in dimensionality (number of channels), not in strength.

4. **C1 is refined, not falsified.** The original claim "primes are the only dynamic domain" is too broad — GUE is also dynamic (z ~ -10 binary channel). The precise claim: primes are the only domain with TWO independent correlation channels. GUE has one (alternation). Cramer has zero. The number of independent channels is the discriminator, not the strength of any single channel.

## Verdict

**CONSTRAINT on C1 + NEW on BOUNDARY + DUALITA_DIPOLARE_VS_ILLUSORIA**

- **C1**: Primes are not unique in having scale-invariant correlations — GUE shares the binary channel. Primes ARE unique in having a second, independent magnitude channel. Reformulate C1: "Primes are the only domain with two independent scale-invariant correlation channels under the binary/magnitude decomposition."

- **BOUNDARY**: The boundary between "structured" and "unstructured" is not binary. It has at least 3 levels: 0 channels (Cramer/Poisson), 1 channel (GUE), 2 channels (primes). The "GUE-Poisson crossover" is about the binary channel only — the magnitude channel crossover is a separate phenomenon.

- **DUALITA_DIPOLARE_VS_ILLUSORIA**: GUE's single channel is purely statistical (level repulsion) — it's generative duality (det = -1). Primes' two channels combine algebraic (mod-3, sieve) with statistical (magnitude) — two KINDS of duality operating simultaneously. The dual duality is the prime fingerprint.

## Bicono della scoperta

- **Due radici** (dipolo primario): one-channel structure (GUE: anti-bunching only) / two-channel structure (primes: anti-bunching + magnitude). The GUE pole is "correlation without algebra" — the primes pole is "correlation with algebra." Inverted: in one, all memory lives in alternation; in the other, alternation is the surface and magnitude carries the depth.

- **Singolare**: the raw gap sequence before decomposition. It does not distinguish one-channel from two-channel. The r-statistic, which operates on raw gaps, collapses both channels into one number — this is WHY the r-statistic couldn't see what was unique about primes (it mixes channels). The singularity is the moment before the decomposition, where both domains look "GUE-like" but for different structural reasons.

- **Invariante di passaggio**: the binary/anti-bunching channel survives the crossing between primes and GUE. It is the universal signature of repulsive-gap sequences, regardless of whether the repulsion is arithmetic or statistical. What does NOT survive: the magnitude channel. This is the asymmetry that discriminates.

- **Campo di possibilita**: diventa possibile — classify domains by their NUMBER of independent correlation channels (0, 1, 2, ...) as a structural fingerprint. Diventa non-possibile — use any single observable (r, ACF, Brody beta) to discriminate primes from GUE. The single observable collapses channels; the discrimination lives in the decomposition.

## Files

- Script: `tools/exp_two_channel_cross_domain.py` (reusable, parametric)
- Data: `tools/data/two_channel_cross_domain.json`
- Report: `tools/data/reports/agent_20260429_0852.md`