STAR's Residual Vote-Splitting¶
One line: STAR removes the forced vote-splitting that wrecks Choose-One — you can score two similar candidates both high, so running an ally doesn't bleed your support. What survives is a narrow, self-inflicted residual in the top-two runoff: a faction can still split itself if it refuses to use the score ballot, or tries to game which ally claims the second finalist slot.
→ Glossary: spoiler effect · related: Center Squeeze · debate version: Favorite Betrayal — Does Only RCV Avoid It?
Why STAR kills the ordinary spoiler¶
Choose-One forces every voter to name one candidate, so two similar candidates mechanically divide their shared support and a minority can slip through. STAR's scoring round removes that: a coalition voter gives Ada a 5 and ally Ben a 4, so Ben's presence costs Ada nothing. Add a clone, lose nothing. That is the whole spoiler-resistance pitch, and it is real.
Where the residual lives — the top-two runoff¶
Only the two highest score totals advance to the automatic runoff. That single fact is the only place a split can sneak back, and it takes voter behavior to trigger it. Two flavors:
1. Self-inflicted split (bullet-voting). If a faction tribally bullet-votes — Ada-fans give ally Ben a 0, Ben-fans give Ada a 0 — they divide their own score totals, and the opponent can grab a finalist slot on the split field. Nothing about the electorate's true preferences caused this; the voters declined to express the support they actually feel. The cure is on the ballot: score your ally even a 3 and the split vanishes.
2. The "chicken dilemma" (strategic). When two allies both could make the runoff, each wing has a tiny incentive to under-score the other ally so that their favorite, not the ally, claims the second seat. If only one wing does it, it may work; if both do it, they knock their shared side down and hand the seat to the opponent. It needs accurate polling and coordination, and it backfires when you guess wrong.
Minimal test case — run the matched pair¶
Same three candidates and the same true preferences; only the expressiveness changes (60 voters on the Ada/Ben side, 40 for opponent Cara):
→ 05a — bullet-voting · 05b — expressive fix (run: 05a · 05b .yaml)
| File | Ada/Ben behavior | Scores (Ada / Ben / Cara) | STAR winner |
|---|---|---|---|
| 05a | bullet-vote (ally = 0) | 175 / 125 / 200 | Cara (minority) — split not rescued |
| 05b | score ally a 3 | 250 / 230 / 200 | Ada (majority side) — split fixed |
In 05a the engine's [Vote-splitting check] is honest: it flags the majority coalition, then reports "STAR elected Cara" — STAR does not paper over a split the voters inflicted on themselves. In 05b, [Divergence from STAR] shows Choose-One still electing Cara while STAR diverges and keeps the seat on the majority side.
Why it's a narrow edge case, not the spoiler all over again¶
- The ordinary spoiler is automatic — no one has to coordinate or even intend it: under Choose-One the split happens on its own whenever similar candidates run (whether it actually flips the winner then depends on the margins). Splitting is the default, not a strategy someone has to execute.
- STAR's residual is the exception — it needs a near-tie for the second runoff slot, accurate prediction, coordinated mis-scoring, and it backfires if you're wrong. The brute-force simulation finds that a favorite-betrayal-style play in STAR helps only ~2% of the time it changes the result — it backfires ~98% of the time. So honest, expressive scoring stays your safest ballot.
Same root cause as STAR's favorite-betrayal and center-squeeze edge cases: all three come from "only the top two advance."
How to say it honestly¶
Not "STAR eliminates vote-splitting" (full stop) but "STAR ends forced vote-splitting and shrinks the rest to a narrow, hard-to-coordinate, backfire-prone edge case — dramatically reduces, doesn't claim to eliminate." The honest version is still the winning version.
This is the same discipline the rest of the library applies to every claim: concede the real limit, then make the (still strong) accurate case. See STAR's honest limits for the full set, and — for the mirror image, our own side's overclaims about the other methods — the "claims our own side oversells" section. "Dramatically reduces, doesn't eliminate" is exactly the honest register that keeps the whole case credible.