Normalization technique applied across features for each token; stabilizes training in transformers. ← Layer 1 (L1) Keccak‑256 →