# Full scaled-k parameter sweep. Mean accuracy / q_eff / convergence are k-invariant,
# so k is scaled down for tractability; re-run headline + variance figures at full scale
# with fullscale.yaml.
# Sweep axes (cartesian product x replicates); every value is a list.
n_nodes: [1000, 2000, 4000]      # number of nodes / stake holders
stake_dist: [uniform, pareto]    # stake distribution (uniform = equal, pareto = heavy-tailed)
latency: [0, 1, 2, 4, 8, 16]     # L: network latency in slots; block visible to others at t+L
max_uncles: [0, 1, 2, 3, 4]      # U: max uncle references per block (0 = baseline, no uncles)
uncle_strategy: [oldest, random] # uncle selection: oldest-first fill vs random coin-flip
replicates: 12                   # independent RNG replicates per grid cell
base:                            # per-run settings shared by every cell (not swept)
  k: 64                          # security parameter (scaled; T = 6*floor(64/f) = 11520 slots)
  epochs: 45                     # epochs simulated per trajectory
  f: 0.03333333333333333        # slot activation coefficient (default 1/30); configurable
  genesis_d_factor: 0.5          # genesis D_est = factor x true total stake
  pareto_shape: 1.16             # Pareto (Lomax) tail index for the pareto stake distribution