Anvil and Rifle, which uniform sampling reached zero times, are 5,284 and 781 of 20,000 - so the corpus can produce the cases the formation work needs. Order is the one row of thirty-three it cannot: it wants four of the same model, and the generator fills each slot independently. Two findings for the precedence order: Battle and Berserker/Close have identical satisfies counts at every seed, so Battle is never what a force is, and Ranger and Heavy Battle qualify in thousands and win in dozens. Row stability is measured against a second seed rather than computed, because lances are clustered by faction and year and a binomial reads too tight. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Change-Id: Ie4bdb13a06ab54c2689e1154cc785625508b2cef