# Results (28 October 2026)

Pre-specified (plan hash in PLAN-LOCKED.sha256), run by `scripts/rain_test.py`; added checks by `scripts/gauge_and_robustness.py` and `scripts/drought_checks.py` (listed in DEVIATIONS.md).

- Grid diagnostic: the ANM daily grid copies the gauges at their cells (median ratio 1.000, 79% within 2%), anomaly correlation 0.989 at 10 km, 0.849 at 50 km. Design A (inside vs ring) and the 2025 reversal therefore measure the smoothing, not rain inside circles; neither is shown as a finding.
- Design A, as run: −0.24%, placebo p 0.77 (not shown).
- Gauge-only: 6 stations within 10 km of dated circles, −0.3% (95% −10.2…+10.0), 125 controls; R15: 11 stations −0.7%; R20: 12 stations −0.7%.
- Circles: 9 of 80 dated circles hold a gauge with a public daily record; median distance to nearest gauge 17.7 km.
- Design B (downwind − upwind): −4.97%, placebo p 0.031 (0.035 with the added 2,500 draws); matched on baseline p 0.027. Units alone: Oltenia −11.4, Prahova −8.7, Mureș +0.6, Timiș +3.2, Vrancea +3.6, Iași +7.1. Post ≤2022: −2.5%. Without the 2018 cohort: −1.2%.
- Authority table: 134/216 equal ANM daily sums to 0.05 mm (next-day label), 74 (same-day label); 16 differ by more than 5 mm; Roman July 2021: 84.2 printed vs 136.1 / 167.4.