# SMASH 2026 Paper #15 — Run Summary

**Patient:** child#001 | **Weights:** glycaemic 0.50, hypo 0.30, dose 0.20 | **Hypo limit:** 4% | **Days/rule:** 2

**Matched-pair analysis (child cohort):** 154 pairs within 3 TIR points (paper: 148), median hypo gap 8.9 pp (paper: 6.4), max gap **64.0 pp** (paper: 39.8) — worst pair: basal 2/6·suspend<80 (TIR 35.5%, hypo 1.6%) vs basal 5/6·+1 if >180 (TIR 33.2%, hypo 65.5%) at only 2.3 TIR points apart.

**Your weights' pick:** "basal 2/6 · +1 if >180" — TIR 48.3%, hypo **16.4%** (4× above the 4% limit). A Pareto-optimal alternative ("basal 2/6 · suspend<110 · +1 if >180") achieves TIR 50.4% with only 1.7% hypo, but your weights do not select it.

**Unsafe preferences:** 54% of all weight vectors pick a rule exceeding the 4% hypo limit; only 9 distinct rules reachable across the simplex.

**Differential reward warning:** rank correlation with absolute reward is 0.50; its favourite rule has 15.0% hypo, confirming the paper's finding that differential reward is exploitable.