Skip to content

The Black Curtain — one electorate, four "identical" landslides

A worked case study from the video "The Black Curtain" by Common Sense for Uniting America (Ted Getschman / MaxVoting.org). BetterVoting template: https://bettervoting.com/p9gwc3/vote. Adam's notes doc: The Black Curtain (Google Doc). This folder on GitHub: https://github.com/masiarek/YAML/tree/master/method_comparisons/black_curtain. This folder re-creates all four elections as runnable YAML cases for this repo's engine.

The thought experiment

Five voters score candidates on a 0–9 ballot ("least" → "most"). You are shown four different elections, one at a time, and asked the same question each time: "Which candidate should win? Write down your answer."

Then the curtain drops. If all you count is each voter's first choice — which is all Choose-One (plurality) uses, and all RCV-IRV needs here (C has 3 of 5 first choices, an instant >50% majority, so the count stops immediately) — the four elections are indistinguishable: four identical landslide victories for candidate C, 3 votes to 2. Everything that made them different is hidden behind the black curtain.

The score ballots reveal four very different electorates:

# What the scores actually show
1 A hidden consensus. Every voter scored B an 8 — but B is nobody's first choice.
2 Two near-clones. Everyone loves both A and C (8s and 9s); B is universally hated.
3 A polarizing "winner." C is loved by 3, zeroed by 2; A is loved by all five.
4 Four candidates, same curtain. A compromise candidate (B) leads on average score by a hair.

The four elections (original 0–9 video ballots)

Election 1 — first choices: C 3, A 2

voter A B C
1–3 0 8 9
4–5 9 8 0

Choose-One → C · RCV-IRV → C (instant >50%) · STAR → C (runoff 3–2 over B) · Approval (approve = 5+) → B (5 vs 3 vs 2) · Score 0–9 → B (avg 8.0 vs C 5.4, A 3.6)

Election 2 — first choices: C 3, A 2

voter A B C
1–3 8 0 9
4–5 9 0 8

Choose-One / RCV-IRV / STAR / Score 0–9 → C (avg 8.6 vs A 8.4) · Approval → coin toss A/C (5 approvals each)

Election 3 — first choices: C 3, A 2

voter A B C
1–3 8 0 9
4–5 9 1 0

Choose-One / RCV-IRV / STAR → C · Approval → A (5 vs 3) · Score 0–9 → A (avg 8.4 vs C 5.4, B 0.4)

Election 4 — first choices: C 3, A 2

voter A B C D
1–3 0 4 9 5
4–5 9 8 0 1

Choose-One / RCV-IRV / STAR → C · Approval → coin toss C/D (3 each) · Score 0–9 → B (avg 5.6 vs C 5.4, A 3.6, D 3.4)

Results at a glance

Election Choose-One RCV-IRV STAR Approval (5+) Score 0–9 All methods
1 C C C B B compare →
2 C C C tie A/C C compare →
3 C C C A A compare →
4 C C C tie C/D B ranked & approval agree; only Score differs

The All methods page for each election is the generated cross-method write-up — STAR · RCV-IRV · Ranked Robin · Approval · Range / Score · Condorcet — with every method's tabulation and links back here. (Election 4 agrees across the ranked and approval methods, so the divergence ledger doesn't page it; its Score result is above.)

Lessons learned

  1. There is no "right" answer — but the method decides. The video's own framing: "There was no right answer, only your answer. However, your answer will impact how united or divided this society is." Identical voters, different counting rules, different winners. Choosing a voting method is choosing which question you ask: who is the majority's favorite? (Choose-One, RCV-IRV, STAR) or who has the broadest support? (Approval, Score).

  2. The ballot decides what you can even know. A choose-one ballot records only the top of each voter's mind. A ranked ballot records order but not intensity — 9-vs-8 and 9-vs-0 both flatten to "1st > 2nd" (see Scores vs. Ranks — Don't Confuse Ranks and Ratings). Only the score ballot distinguishes these four electorates. And because C has an outright first-choice majority, RCV-IRV's count ends in round one — the published result is identical in all four elections. Election 1's universally-liked B and Election 3's universally-liked A are invisible: the black curtain.

  3. STAR reads score ballots but answers the majoritarian question. In all four elections STAR sees the full score data, puts the consensus candidate in the runoff where relevant, and still elects C — because the automatic runoff hands the decision to the 3-voter majority. That is by design: the runoff is exactly what separates STAR from pure Score (compare the Runoff-Reversal BV cases, especially Runoff_03_enthusiasts_vs_majority). Whether that's a feature or a bug is the values question this video is about.

  4. Approval is Score at 1-bit resolution. The approve/don't-approve threshold (here: video score 5+) collapses the near-clones of Election 2 into a coin toss and manufactures a C/D tie in Election 4. More resolution, more information, finer distinctions — the fidelity ladder (The Fidelity Ladder — converting between scores and ranks).

  5. Scale granularity matters (the 0–9 → 0–5 problem). The video uses six distinct scores — 0, 1, 4, 5, 8, 9 — which map one-to-one onto this repo's 0–5 scale (0→0, 1→1, 4→2, 5→3, 8→4, 9→5). Order inside every ballot is preserved, so every Choose-One, RCV-IRV, STAR, Approval, and Condorcet result above reproduces exactly. The single casualty: Election 4's pure- Score winner, decided by a 0.2-average margin (B 5.6 vs C 5.4), can't be expressed at 0–5 resolution (integer totals become C 15, B 14). No monotone integer mapping into 0–5 can save it — six distinct values are forced onto exactly 0,1,2,3,4,5. A near-tie that flips under rescaling is itself a teaching point: the finer the scale, the more of the electorate's mind survives the count. For a case where rescaling flips not just a pure-Score near-tie but the STAR winner itself, see Scale granularity can flip the winner (a 301 case) (Curriculum 301.9).

  6. Verify, don't trust — even friendly sources. Two on-screen averages in the video don't reproduce from its own ballots: Election 1 shows A's average as 1.6 (the ballots as shown compute to 3.6), and Election 3 shows C's as 5.6 (computes to 5.4). Neither changes any winner — but catching them is exactly why this repo embeds expected_winners: in every file and re-runs everything through the engine.

Files in this folder

The Results page (left) is the friendly, readable write-up — start there; the raw .yaml (right) is the tabulatable source.

Results page (read this) Method Winner Shows Source
01 — hidden consensus STAR Cal Majority favorite beats the everyone's-second consensus .yaml
01a — approval variant Approval Bob Same 5 voters, different method, different winner .yaml
01b — the compressed ballot STAR (0/5) Bob The approval marks read pairwise: the Condorcet winner moves from Cal to Bob — see the write-up .yaml
02 — near-clones STAR Cal Two loved near-clones; approval can't tell them apart .yaml
03 — polarized on Cal STAR Cal The "landslide" winner is zeroed by 40% of voters .yaml
04 — four candidates STAR Cal Four candidates, same curtain; score near-tie lost at 0–5 .yaml

Each Results page carries the full tabulation (scoring round, runoff, matrix, winner). Candidate names follow the BetterVoting template cast: Ann, Bob, Cal (+ Dee). Run any file:

python STARVote_LH_tabulation_engine/starvote_larry_hastings.py method_comparisons/black_curtain/cases/Black_Curtain_01_c3_b5_hidden-consensus.yaml

file: README.md