`model_0` from `qwen3.8-flash-next`, 105/105 on the seam's own corpus, admitted on the first batch. The three `layout*` entry points now default to it instead of `containing_block::resolve_baseline`, and every other seam's runner holds it at the admitted candidate the way they already hold the others. The batch: 77 of 82 candidates compiled across four gateway models (flash-next 16/16, 27b-fast 30/32, 27b-fast-2 30/32, deepseek-v4.1-flash 1/2), and every one that compiled and scored reached 105/105 but a single candidate that stayed at the baseline's 30/105. The loop stopped after one batch because no failing case was left to repair. Recorded in docs/model-first-cost.md, docs/seams.md and roadmap.md. `cargo test --workspace` passes, `fragment_check --random` is 227/227 with every category unchanged, and bench/seam-floor.sh's four floors hold at floats 109/109, stacking-order 105/105, relative-positioning 104/104, block-height 105/105.