# IOL-AI 2026 — final burn: ATB 0.178 + Lipas normalize + explain

Measured base: pad/hygiene/match + RP=1.0 + winner encode + explain 100%.
Add-only: Lipas `safe_normalize` for `match_letters` / `text_to_num`.

Expected: **~0.179–0.183** (peer ceiling 0.1828). Not a path to 0.19.

Nominate prior **0.1782 / explain 100%** if this regresses.