01_STAR/04_Real_Elections — real BetterVoting races, read end to end¶
Constructed examples make one idea clear; real elections prove the method survives contact with real ballots. Every set here is a live BetterVoting race imported into this repo and reconciled line by line against the LH engine — so you can see both how a result looks in the BV interface and how it tallies in an independent implementation.
That reconciliation is also how this repo finds BetterVoting bugs: where the two disagree, the disagreement is the lesson.
Levels follow the curriculum: 🟢 101 · 🟡 201 · 🔴 301.
| Set | Level | What it shows |
|---|---|---|
| "What Makes the Best Pet?" | 🟡 201 | The whole pipeline on one race: 7 candidates, 461 ballots, raw ballots → winner, read section by section. The worked example behind the runoff percentages lesson. |
| Runoff reversals on BV | 🟡 201 | Real elections where the Scoring-Round leader loses the Automatic Runoff to the finalist more voters prefer — STAR's headline behavior, not a constructed one. Two-view: BV screenshots beside the engine report. |
| Abstain, blank & zero handling | 🟡 201 | Where BetterVoting and the engine diverge: BV counts an all-equal ballot (0,0 and 5,5) as an abstention, the engine counts an explicit score as a real vote. The visible symptom is BV's "0 tallied votes yet a winner" (#884). |
| Goodberry's Best Flavor 2026 | 🟢 101 | Open now — a live poll, not an import: ten frozen-custard flavors, paper ballots at the Cary NC shop plus a QR code, no seed ballots. The frozen export is the mint-time snapshot (zero votes), so this set is the before picture; re-freeze it when the poll closes and the reconciliation joins the rows above. |
| Florida 2026 — four STAR races | 🟡 201 | A public poll runs the 2026 Florida statewide ballot as four STAR races on one open, all-candidate field — the contest the closed partisan primary cannot hold. Small (25 ballots) and self-selected, so it settles nothing about Florida; what it shows is what the ballot records: an unaffiliated candidate placing 4th on secondary support, and a Senate runoff decided by 7 of 25 voters because 18 scored both finalists the same. Also a live instance of the 0-vs-blank abstention divergence in three races at once. |
| Does scale usage matter? (pres24) | 🔴 301 | 2,772 real ballots, 8 candidates — the public 2024 presidential poll, imported to answer a live question: 14% of ballots never used both ends of the 0–5 scale. Reproduces the figure, then shows the winner is invariant (the scoring leader is also the Condorcet winner) while the second finalist slot is a coin flip from sampling noise. Also finds the flagged ballots carry more runoff weight than average. |
Related: the constructed teaching progression → 02_Examples · the tie behaviors these races run into → 03_Criteria · every BV-backed case in the library → the BV registry.