Reading these fairly — the test for an honest "whoops"¶
Edge cases are powerful and dangerous. With a contrived enough construction you can make any voting method look ridiculous — so a gallery like this can quietly become propaganda. This page is the guardrail. Apply it before adding a case, and apply it to your own favorite method first.
Why the danger is real¶
- Every method fails something. Arrow's theorem (no ranked method satisfies a short list of reasonable fairness criteria at once) and Gibbard–Satterthwaite (every method is manipulable by strategy) guarantee it. So "method X fails criterion Y" is always true for some Y. The honest question is never whether it fails, but which failures matter and how often they bite.
- The strategic angle muds everyone. Because every method is manipulable, "you can game it" is the cheapest attack — it applies to STAR, Approval, Condorcet, and IRV alike. Lead with it and you're just throwing mud.
The four-part test¶
A case earns a place here when it passes all four:
- Structural, not measure-zero. Does the failure occupy a region of realistic configurations, or does it need a knife-edge with absurd weights and exact ties? Center squeeze is a whole zone of normal spatial electorates (fair); a result that needs 1000-to-1 weights and three-decimal scores is not.
- Sincere, not strategic. Failures under honest voting are fair currency. If a critique needs coordinated strategy, say so, loudly — and remember the same trick usually works against your favorite method too.
- Realistic electorate. 1-D / 2-D spatial models, natural preference distributions — not alien voter behavior invented to trip one method.
- Bonus: it really happened. A real election (Burlington 2009, Alaska 2022) is evidence, not a construction. Strongest footing of all.
Symmetry is the rule (and fairness ≠ false balance)¶
Hold every method to the same standard, with equal prominence:
| Method | Its honest "whoops" | Roughly how often / how bad |
|---|---|---|
| Plurality | spoilers / elects a candidate a majority opposes | common; the reason this whole field exists |
| RCV-IRV | center squeeze; non-monotonicity; exhausted ballots | structural; real cases (Burlington, Alaska) |
| STAR | can miss the Condorcet winner; reversal surprises people | rare (~98% Condorcet-efficient in spatial models) but structural |
| Approval | threshold strategy; bland lowest-common-denominator winner | strategy-sensitive by design |
| Condorcet / Ranked Robin | cycles (no winner without a tiebreak); ignores intensity | cycles rare with many voters; intensity-blindness is inherent |
| Borda | teaming / clone vulnerability; easy to manipulate | strategy-sensitive |
But same standard ≠ pretending all methods are equally broken. That's its own distortion. Honesty means stating frequency and severity, with sources — "here is IRV's failure and how often it shows up; here is STAR's analogous failure and its (lower) rate." Let the proportions show; don't flatten them, and don't inflate them.
The fairness box (paste into every case)¶
Every lesson in this folder ends with this block, filled in honestly:
> ### Reading this fairly
> - **How common:** structural region · rare-but-real · knife-edge construction
> - **Sincere or strategic:** does it need anyone to vote dishonestly?
> - **What this method does well:** the other side of the ledger
> - **The symmetric whoops:** the analogous failure in STAR / Approval / RR / IRV
A gut-check before you publish¶
If this example embarrassed your favorite method instead of your least favorite, would you still call it fair? If not, don't use it. If yes, you've found an honest one — and you should go build the one that embarrasses your favorite, too.
→ Standing house policy on terminology and not-being-a-purist: Tips — Terminology: RCV vs IRV vs RCV-IRV (and friends). The even-handedness duty is also why STAR's own limits are documented as peers, not footnotes: STAR's limits · three winner notions.