election_title: "Margins matter — the textbook profile at its printed size (304 ballots, LH-only)"
scenario_description: |-
  The source profile exactly as it is printed in the textbook: 102 a>b>c, 101 b>c>a, 100 c>a>b, 1 c>b>a, with a/b/c kept as the book's own placeholder letters. Reference artifact for reproducibility — LH-only, no BetterVoting election, because the twelve-ballot version (margins_ranked_robin.yaml) is the teaching copy. Everything structural matches: Copeland ties all three, the symmetric Borda scores are 0 / +2 / -2, plurality elects a, RCV-IRV elects c. Two things this size shows that the small one cannot. First, the margins are 100 / 102 / 100 — near-identical, so margin-weighting breaks a dead heat by a HAIR, where the shrunk version's 2 / 4 / 2 makes the gap look decisive. Second, the bloc counts are coprime (gcd 1), so this profile is not a scaled-up copy of anything smaller — and the lone 1-voter is load-bearing: delete it and the symmetric Borda scores become +2 / +2 / -4, a TIE between a and b, and RCV-IRV flips from c to a.
voting_method: RankedRobin
num_winners: 1
lot_numbers: [a, b, c]
ballots: |-
  102:a>b>c
  101:b>c>a
  100:c>a>b
  1:c>b>a

expected_winners:
  - b

# file: margins_paper_exact_304.yaml