Done. Here's the summary:

## Results

**META Tautology Discriminator** — tested 5 key observables against 3 synthetic controls (shuffled, Cramer, Hardy-Littlewood AR(1)):

| Observable | Verdict | Strongest z-score |
|:-----------|:--------|:-----------------|
| mod-3 ordering | **STRUCTURAL** | z = -107 vs Cramer |
| r-statistic | **STRUCTURAL** | z = +17 vs Cramer |
| dr-magnitude | **STRUCTURAL** | z = -9 vs HL |
| lag-1 ACF | TAUTOLOGICAL | z = -0.3 vs HL |
| dr-ordering | TAUTOLOGICAL | z = -0.9 vs HL |

**The finding**: 3 of 5 observables are genuinely structural — they can't be reproduced by any synthetic model, even one matching pair correlations. The mod-3 signal (z ≈ -100) is the single hardest discriminant in the lab. Two observables (lag-1 ACF and ordering-channel delta-r) are fully explained by pair correlations — a simple AR(1) process reproduces them. These add no independent evidence.

**Consecutio**: The ordering channel was thought to be an independent evidence channel — it's not. It's redundant with pair correlations (illusory duality, det=+1). The magnitude channel IS independent (dipolar duality, det=-1). Next: test whether mod-3 is genuinely non-Markovian (survives AR(2), AR(3)) or eventually captured at some Markov order.

**Files**: `tools/exp_meta_tautology_test.py` (reusable), `tools/data/meta_tautology_test.json`, `tools/data/reports/agent_20260429_0958.md`