Experimental work: TRITON/MODEL CO-DESIGN BODY (x7): architecture family detected in measured rows; exact flags are preserved in the row details. 11L GQA U-NET BODY (x5): 11-layer dim-512 GQA U-Net transformer base with XSA, BigramHash, value embedding, and SmearGate. Specs: num_layers=11; model_dim=512; vocab_size=1024; num_heads=8; num_kv_heads=4; head_dim=64. CRAWLER-SAFE TRAIN/EVAL SURFACE (x5): Crawler-safe training/eval surface around a widened shared-crawler stack with final int6 and int8 checks. Specs: model_dim=512; vocab_size=1024; num_heads=8; num_kv_heads=4; BigramHash; XSA.
Neural work: 11L GQA U-NET BODY (x1): 11-layer dim-512 GQA U-Net transformer base with XSA, BigramHash, value embedding, and SmearGate. Specs: num_layers=11; model_dim=512; vocab_size=1024; num_heads=8; num_kv_heads=4; head_dim=64.
Day 013 measured 22 architecture/test rows against VAL_BPB, with 21 valid rows and 0 TTT leakage readings excluded. Techniques to copy/reference: SmearGate; BigramHash; XSA; RoPE; crawler loop rope scales; instruction bottleneck; int6 roundtrip; int6 sliding-window eval.
No new personal best. Best valid of the day = 1.1091 (recurrent green-branch prototype.log); best-before-today = 1.0672 still stands. 1 crash-class outlier (4.10) excluded.