# S3: uncle-window sufficiency at scale (N = 1 000 vs 10 000) and the W-as-buffer question: # does a wider window absorb block-production fluctuations near the load boundary (rho ~ 1) # and soften the near-integer U undershoot? Report §3.4 / §8. n_nodes: [1000, 10000] stake_dist: [pareto] topology: [blend] degree: [6] link_latency_mean: [0.5] link_latency_dist: [geo] blend_hops: [3] blend_delay_max: [8.0, 16.0, 32.0] max_uncles: [1, 2] uncle_window: [50, 100, 200, 300, 450, 600] uncle_strategy: [oldest] init_dest: [common] replicates: 3 base: k: 256 epochs: 40 f: 0.03333333333333333 genesis_d_factor: 0.5 early_stop: true