nx_bench_dist_emit_test.nx

nx_bench_dist_emit_test.nx -- 10-run distribution measurement. Runs the v2 LLM 10-iter scenario TEN INDEPENDENT TIMES, captures wall-clock for each, computes min/max/mean/p50/p95/stddev across the 10 samples, and emits the distribution summary to stdout. This is the substrate-honest version of "single-point benchmark": every serious comparison reports distributions because variance matters. A p50 + p95 + stddev tells the operator far more than a single wall-clock number. Output (stdout): nx_bench_dist_v2_10iter N=10 min=... max=... mean=... p50=... p95=... stddev=... plus raw 10 wall-clock samples line so operators can re-analyze HONEST SCOPE: qemu-riscv64 emulated time. Variance under qemu is dominated by interpreter scheduling artifacts; native hardware will show different absolute numbers + variance shape. The SHAPE of the distribution (tight p50-p95 spread vs long-tail outliers) is the substrate-honest signal.