STAR Voting FAQ — the mechanics, answered with worked examples¶

Level: 201 · for voters

The official STAR Voting FAQ answers dozens of questions in prose. This page mirrors the mechanics questions — the ones about how the count works — and answers each with a runnable example: a small YAML election you can tabulate yourself, with the engine's own output shown inline. The aim is an impartial walk-through of the mechanics: where STAR fails a criterion or carries an honest tradeoff, this page says so and links a worked case, rather than arguing the method. (For the case for STAR, see Why STAR Voting?; for its limits, STAR's honest limits.)

For the political, legal, and historical questions (cost, constitutionality, security, the League of Women Voters, presidential use, Oregon history…), the official FAQ is the right source — those aren't things a tabulation engine can demonstrate.

New to the terms? Start with Ballot & Terminology Basics and the Glossary.

The one election behind half these answers¶

Most of the runoff questions below are really the same question. Here's a 10-voter STAR election that shows all of them at once — Berry is broadly liked and wins the Scoring Round, but more voters strictly prefer Almond, who wins the Automatic Runoff. One voter scores both finalists 5 (Equal Support).

▶ Live on BetterVoting: vote · results ↗ (BV2182, election tg4779). Reader page: Almond vs Berry — a runoff reversal · run it locally: bv2182_tg4779_faq_runoff_reversal.yaml.

Abridged for the lesson — not verbatim engine output

Count × Almond,Berry,Cocoa
    4 ×      5,    4,    1
    3 ×      0,    5,    2
    2 ×      5,    4,    0
    1 ×      5,    5,    0        ← scores both finalists 5 = Equal Support

Scoring Round
   Berry         -- 44 -- First place      ← highest total score
   Almond        -- 35 -- Second place
   Cocoa         -- 10
 Berry and Almond advance.

Automatic Runoff Round
   Almond        -- 6 -- First place       ← preferred head-to-head
   Berry         -- 3
   Equal Support -- 1
 Almond wins.
   Voters with a preference: 9 of 10 (1 Equal Support).
   Almond 6 (67%) vs Berry 3 (33%); majority = 5.

¶

Q: Why bother with the automatic runoff? Why not just elect the highest scorer?¶

The two rounds measure different things: total score rewards breadth (how many stars a candidate collects), the runoff rewards pairwise preference (which of the two front-runners more voters scored higher). In the example, Berry collects the most stars, but asked "Almond or Berry?" six of the nine decided voters pick Almond — so Almond wins. Note the scope of that claim: the runoff decides only between the two top-scored finalists. It does not guarantee the overall majority or Condorcet winner — a third candidate can beat both finalists head-to-head yet never reach the runoff, which is a genuine STAR failure mode (the Condorcet winner who scores third).

This "score leader ≠ runoff winner" outcome is a Runoff Reversal, and whether it's the right call is genuinely contested: sometimes the runoff rescues a majority-preferred candidate from vote-splitting, sometimes it overrides broad average support on a thin pairwise margin. Both readings are worked out — not just the favourable one — in STAR's second round — an FAQ, with real elections in runoff reversal cases. Deeper: The Automatic Runoff.

Q: What if I give both finalists the same score?¶

Then your ballot expresses no preference between them — that's Equal Support. It's counted, not discarded: it did its job in the Scoring Round (helping both candidates' totals) and in the runoff it goes to neither finalist's column. In the example, the 5,5,0 voter is the Equal Support -- 1 line, and the runoff is decided among the 9 voters who did express a preference (majority = 5, not 6). The tradeoff to be clear-eyed about: scoring the two finalists equally means you give up your say in the runoff between them — so if you do have a preference there, express it. And while equal scores don't split your vote in the Scoring Round, STAR is not entirely immune to vote-splitting — see the rare residual vote-splitting case. Full treatment: Abstention vs. a zero vs. NOTA.

Q: "Wasted votes" — what's the difference between an exhausted ballot in RCV and an Equal Support vote in STAR?¶

Both often get filed under the same loaded label — "wasted votes" — but they're two different things. The outcome can look similar — neither ballot affects the final pairwise decision — but the mechanism differs, and that's the honest distinction. In RCV-IRV, an exhausted ballot (all its ranked candidates eliminated) is set aside in the deciding round; the voter didn't choose that — elimination did. In STAR, an Equal Support ballot reaches the runoff, but the voter rated the two finalists equally, so it contributes to neither — a choice the voter made, and one where the ballot's scores still counted in the Scoring Round. So the difference is agency and stage (involuntary drop late vs. voluntary no-preference, ballot still counted earlier), not that one voter matters and the other doesn't — a STAR voter can equally decline to help decide the runoff by scoring both finalists the same. Worked example: Exhausted ballots (IRV).

Q: Does STAR give everyone exactly one equal vote?¶

In the runoff, yes. However many stars you gave out, the Automatic Runoff counts one vote per voter — for whichever finalist you scored higher (or none, if you scored them equally). In the example all 10 voters cast one runoff vote each (6 Almond, 3 Berry, 1 Equal Support); scoring 5-vs-4 carries the same runoff weight as 5-vs-0. But be precise about scope: the Scoring Round is different by design — a score ballot lets intensity and breadth count, so a voter who spreads 5/3/1 shapes which candidates become finalists differently than one who bullet-votes 5/0/0. STAR's one-person-one-vote claim is specifically about the runoff step, not a claim that every voter has identical influence in round one. See The Automatic Runoff and the equally-weighted vote.

Q: Why is a blank — or a skipped candidate — counted as a zero?¶

Because a blank means "I did not support this candidate," which is exactly what a 0 encodes — it can't help that candidate advance. But STAR's engine keeps why a zero is a zero: this library records a small marker vocabulary — - blank · ~ race abstention · & candidate abstention · ? spoiled · % spoiled+reissued — all of which tabulate as 0 but are reported distinctly, so a deliberate abstention isn't confused with an active low score. A whole ballot left blank is set aside as an abstention — but set aside is not thrown out: it still counts as turnout (and toward quorum), it just carries no preference to fold into any candidate's total. An all-equal scored ballot (0,0,0, 5,5,5) is a bookkeeping fork between the two tabulators: BetterVoting sets it aside as an abstention too, while the LH engine keeps it cast — its stars tally in the scoring round and it files under Equal Support in the runoff. Neither choice can change the winner: an all-equal ballot shifts every candidate by the identical amount, so it tips no contest either way. Full detail: Abstention vs. a zero vs. NOTA and "Preference" — the word that causes half the confusion.

Q: What is a preference matrix?¶

It's the head-to-head grid: for every pair of candidates, how many voters prefer each, and how many express Equal Support. It's what the runoff reads, and it lets you check any pairwise contest at a glance. The engine prints one for the example above:

--- Runoff (Preference) Matrix ---
Legend: For - Equal Support - Against   (* = Top-2 Finalist)
                 |  * Almond   |  * Berry    |    Cocoa    |
-------------------------------------------------------------
      * Almond > |     ---     | 6 -  1 -  3 | 7 -  0 -  3 |
       * Berry > |  3 - 1 -  6 |    ---      |10 -  0 -  0 |
         Cocoa > |  3 - 0 -  7 | 0 -  0 - 10 |    ---      |

Read the top row: Almond beats Berry 6–3 (with 1 Equal Support) and beats Cocoa 7–3 — here Almond happens to be the Condorcet winner (beats everyone head-to-head), so the runoff lands on the same candidate. The matrix is also how you audit STAR against itself: in other elections it exposes when the score-and-runoff winner is not the Condorcet winner (a real STAR failure — the Condorcet winner who scores third) or when there's no Condorcet winner at all (a cycle). It's the same pairwise grid a Condorcet method reads.

Q: What happens if there's a tie?¶

It doesn't go to a coin flip first. An exact tie falls through a fixed ladder of deterministic tests, and only if every rung is still level does a pre-drawn lot order settle it. STAR has two rounds, so it has two ladders: a Scoring Round tie (who takes the last finalist slot) breaks on pairwise → five-star → lot, and an Automatic Runoff tie breaks on score → five-star → lot. The swap is the elegant part — the measure that tied can't be the one that separates, so each round borrows the other round's yardstick. A scoring tie means the totals are equal, so STAR asks the runoff's question instead: whom do more voters prefer head-to-head (read straight off the matrix above)? A runoff tie means the preference is equal, so it falls back to total score. Here is the first rung doing the whole job, in a 5-voter election where Ben and Cora tie at 14 for the second finalist slot:

Abridged — excerpted from the full report

[STAR Voting: Scoring Round]
 The two highest-scoring candidates advance to the next round.
   Ana           -- 15 -- First place
   Ben           -- 14 -- Tied for second place
   Cora          -- 14 -- Tied for second place
   Dev           -- 11
 Ana advances, but there's a two-way tie for second.

[STAR Voting: Scoring Round: First tiebreaker]
 The candidate preferred in the most head-to-head matchups advances.
   Cora          -- 3 -- Second place
   Ben           -- 2
   Equal Support -- 0
 Ana and Cora advance.

Note what a tie is not: an Equal Support ballot isn't a tie — that's one voter declining to separate the two finalists (the equal-scores question above), and the runoff can still finish 6–3. A tie is the count landing exactly level. Two honest caveats. The five-star rung can be dead: it counts votes at the scale maximum — literally a 5 — so if none of the tied candidates earned one, it reads 0–0, separates nobody, and the tie drops straight to the lot without ever stepping down to the 4s; in low-scoring or coarse-scaled contests the lot therefore decides earlier and more often than the ladder's length suggests (the dead-rung cases). And the floor differs by engine: this library's engine ends on a published lot order, so even a lot-decided winner stays reproducible from the file, while BetterVoting ends on a seeded shuffle — recorded, but never a function of how anyone voted. Full ladder rung by rung, worked examples with ties in both rounds, and the LH-vs-BetterVoting comparison: STAR Tie-Breaking — the full chain. The election above is Tied for the second finalist (BV2276, qhjyr2). If you would rather see it as a picture, every branch of both ladders is drawn in a map of every tie case.

Q: Does STAR fail Later-No-Harm? Should I "bury" a strong second choice?¶

STAR fails Later-No-Harm (LNH): honestly scoring a second choice can, in constructions, help that candidate beat your favorite in the runoff. That's a real, admitted property — worth stating plainly, not waving away. Two things put it in context, without denying it: (1) LNH and the Favorite-Betrayal Criterion are provably incompatible — no method satisfies both, so every system fails one of the pair, and the two are routinely confused. (2) Whether deliberately burying a rival pays in STAR is a measurable question, not a rhetorical one: a brute-force simulation finds it rarely helps and backfires more often than it works, because under-scoring a strong rival risks electing someone you like even less. So it's a genuine tradeoff STAR accepts and this repo quantifies rather than dismisses — see Favorite Betrayal — Voting 301 and STAR's honest limits.

Q: Are STAR Voting ballots summable?¶

Yes — STAR is batch summable: each precinct can total its own stars and its own pairwise-preference counts, and those partial sums combine into the final result. RCV-IRV is not summable that way — its sequential eliminations depend on the full ballot set, so the deciding round can't be reconstructed from precinct subtotals alone. That's a real, method-inherent difference in how the count composes (it bears on precinct-level reporting and auditing); how much it should weigh against other properties is a judgment for the reader, not this page. Worked side-by-side: STAR summability and the summability demo.

Q: Can RCV-IRV eliminate a broadly-liked centrist? (the center squeeze)¶

It can. IRV eliminates each round by fewest first-choice votes, so a centrist who is many voters' second choice but few voters' first choice can be knocked out early — the center squeeze — leaving two more-polarized finalists. This is a property of IRV's elimination mechanic specifically; it is not a property of ranked ballots in general (a Condorcet / Ranked Robin count of the same ballots isn't squeezed). STAR's score-then-runoff doesn't eliminate the centre the same way — but to be even-handed, STAR is not immune to failure either: it can miss the Condorcet winner (above) and it fails LNH, so this isn't "STAR wins on everything." Worked examples: center squeeze and the center-squeeze cases. Method-to-method, evenhandedly: RCV-IRV vs. STAR.

Where STAR falls short (so this stays honest)¶

No method escapes the impossibility results (Arrow, Gibbard–Satterthwaite) — every system trades something away, and STAR is no exception. Its documented weaknesses:

It can miss the Condorcet winner — the score-and-runoff winner isn't always the candidate who beats everyone head-to-head. Worked case: the Condorcet winner who scores third.
It fails Later-No-Harm — see the LNH question above.
It has rare residual vote-splitting — residual vote-splitting.
The runoff can override strong average support on a thin pairwise margin — the "jarring" side of a runoff reversal.

The full, honest catalogue is STAR's honest limits and Three notions of "winner" — read those alongside the answers above, not instead of them.

For everything else¶

Cost, constitutionality, security, One-Person-One-Vote law, presidential use, endorsements, and STAR's history are covered at the official STAR Voting FAQ. This page stays in its lane: the tabulation mechanics, shown with examples you can run.