election_title: "Downward monotonicity (San Francisco D7 2020) — AFTER: rank the loser lower, he wins"

scenario_description: |-
  The BEFORE file (sf_d7_downward_before) exactly, with ONE change: 800 ballots of
  the form Engardio>Nguyen>Melgar are shifted DOWN to Nguyen>Engardio>Melgar. Nothing
  else moves. This gives the previous LOSER Engardio LESS first-place support (14119
  -> 13319) — yet he now WINS. Why: the shifted votes lift Nguyen's first-round count
  (10855 -> 11655) just past Melgar's 11652, so MELGAR is eliminated first instead of
  Nguyen; Melgar's ballots transfer, and Engardio beats Nguyen in the final,
  15819-15155. Ranking the loser lower made the loser win — the downward monotonicity
  paradox (Graham-Squire & McCune, arXiv:2301.12075). As in the Alaska upward case,
  the paradox works by changing the ORDER of elimination. The Melgar/Nguyen Round-1
  gap is just 3 votes — this is a genuine knife-edge, which is why the real, full-scale
  counts (not a reduced model) are used. Companion page: downward_monotonicity_sf.md.

video_script: |-
  Same election as BEFORE — we only moved 800 Engardio-first voters to rank Nguyen
  first and Engardio second. Round 1 now: Engardio 13319, Nguyen 11655, Melgar 11652.
  Melgar is out this time, by three votes. Melgar's ballots transfer and Engardio
  wins 15819-15155. Engardio got LESS support and won. That's the mirror of the
  Alaska paradox: a method should never reward a candidate for being ranked lower.
  STAR and Ranked Robin can't do this.

paradoxes: [non-monotonicity]
voting_method: RCV_IRV
num_winners: 1

ballots: |-
  13319:Engardio>Melgar>Nguyen
  2500:Melgar>Engardio>Nguyen
  3500:Melgar>Nguyen>Engardio
  5652:Melgar
  6909:Nguyen>Melgar>Engardio
  3051:Nguyen>Engardio>Melgar
  1695:Nguyen

expected_winners:
  - Engardio

# file: sf_d7_downward_after.yaml