Skip to content

Every voting criterion STAR fails — with a worked example for each

The honest, one-stop index: each formal criterion STAR does not satisfy, what STAR does instead, and a runnable example — the reader page, the tabulatable .yaml, the live BetterVoting election (BV id), and the frozen _bv_export.json where one exists. The prose companion (why each tradeoff is deliberate, and the defenders' replies) is STAR's honest limits; this page is the checklist with receipts.

Read this fairly. No method satisfies every criterion — Arrow and Gibbard–Satterthwaite prove it. STAR passes the ones that bite most in real elections — no center squeeze, monotone (mono-raise), no spoiler/vote-splitting in practice, precinct-summable, no Condorcet loser, favorite-betrayal-resistant — and the failures below are mostly rare, constructed, or deliberate tradeoffs. This is the "safety datasheet" (Voting 401), not a verdict.


The failures, each with a receipt

Criterion What STAR does instead Worked example Run it (.yaml) Live BV · export
Condorcet — elect the head-to-head winner scores pick the two finalists, so a Condorcet winner who's only 3rd by score never reaches the runoff STAR misses the Condorcet winner · three notions of winner bv2156_…yaml BV 3grpbb · json
IIA (independence of irrelevant alternatives) in a Condorcet cycle, a candidate who can't win can flip the result by changing which two reach the runoff the cycle spoiler bv2212_…yaml BV g3f7r2 · json
Later-No-Harm deliberately given up — scoring a 2nd choice can help it beat your favorite (the price of favorite-betrayal resistance; they're provably incompatible) Favorite Betrayal 301 (LNH vs FBC, worked)
Favorite-Betrayal Criterion (FBC) rare lab constructions where ranking your true favorite top can be worse than betraying them the favorite-betrayal pair (honest vs betrayal-pays) bv2207_…yaml BV b6xrdr (honest: 7mckyg)
Majority-favorite a polarizing majority-favorite can lose to a candidate almost everyone likes (STAR treats that as the right call) Black Curtain — a polarizing winner / hidden consensus · Majority Criterion hub (see the set) (BV-backed set)
Mutual majority a majority-favored set isn't guaranteed one of its own — the residual of vote-splitting residual vote-splitting (worked in the page)
Mono-raise-delete (Woodall) passes ordinary monotonicity, but raising X and deleting the loser below X can cost X the win monotonicity worked set § the 301 nuance page · mono_raise_delete_after.yaml LH-only (synthetic)
Participation (no-show) in rare constructions a voter can get a worse result by showing up participation / no-show set bv2175_…yaml BV 9dhv8y (baseline yyhr66)
Consistency (reinforcement) winning every district separately doesn't guarantee winning the whole — "who advances to the runoff" isn't additive two districts, one mayor ex01_district_combined.yaml BV 923q3d · json

Notes

  • Condorcet, majority-favorite, and mutual-majority are the same mechanism three ways: STAR's finalists come from the score round, so a broadly-preferred-but-low-intensity candidate — or a divisive majority-favorite — can miss the top two. That's the preference-vs-support tradeoff, and it's what separates STAR from Ranked Robin (which elects the Condorcet winner) — see the 30 STAR-vs-RR divergence samples.
  • Later-No-Harm and FBC are the deliberate half. STAR chooses to fail LNH to keep favorite-betrayal resistance; RCV-IRV makes the opposite choice. The "failure" is a value pick, and Shentrup argues LNH is a bug, not a feature.
  • The mean, one-sided version of this list (curated for sport) is the Mudroom's STAR reel; the fair, cite-in-a-debate version is this page + honest limits.

Sources: each linked case is LH-verified (and BV-verified where a BV id is shown). What STAR passes — and why these tradeoffs are the right ones — is argued in STAR's honest limits and Voting 401.