The canonical capture: 0.728× Node across thirty rows

v0.0.11 and v0.0.12 land register classes for tokenizer loops and inline stores into young dense arrays; the clean PGO capture at 8229b3fc becomes the public benchmark, with 21 of 30 rows faster than Node.

v0.0.11 (9409b8d5): register classes keep tokenizer loops on the INT tier, taking parse-large-js from 1.24× to 0.88× Node, and a forced-JIT segfault on a user-overridden Math.random is fixed. v0.0.12 (13aa7e63): inline stores into pinned young dense arrays and x | 0 fused into the wrapping add. b36702a4 publishes the 8229b3fc capture: headline ten 0.878×, all 13 normal rows 0.614×, hostile 17 0.829×, all 30 equal-row 0.728× Node, 0.594× Bun, 0.460× Deno, with 15 counterbalanced repetitions and 10,000 bootstrap samples.

Nine rows still lose to Node, led by reactish-reconcile at 1.578×, allocation-survival at 1.559× and warm-router at 1.520×. The artifacts print FASTER_THAN_NODE_ON_EVERY_ROW=0.