Favorite Betrayal — Does Only RCV Avoid It?¶

Voting 301 · advanced · the deep-dive behind the "You Never Have to Betray Your Favorite" slide¶

This is the hardest-to-keep-straight idea in the whole STAR-vs-RCV debate, so this file goes slow and over-explains on purpose. If you only remember one thing:

"Favorite Betrayal" and "Later-No-Harm" are two DIFFERENT promises. RCV-IRV keeps the second one and breaks the first. People — including RCV advocates — mix them up constantly.

The 30-second intuition — think of two doors (the full version is §9):

🚪 Favorite Betrayal = the front door. Is it safe to walk your favorite in first? In RCV-IRV, sometimes no — your favorite can be knocked out early for having too few first-place votes. In STAR, almost always yes.
🚪 Later-No-Harm = the back door. Is it safe to let your backups in too? In RCV-IRV, yes. In Score/Approval, no. In STAR, almost always yes.

RCV advocates say "you never have to betray your favorite" — that's the front-door promise — but what they're really describing is the back door (Later-No-Harm). RCV-IRV locks the back and leaves the front open. That swap is the entire lesson; everything below is the proof.

1. The context — what was actually challenged¶

The slide. The deck Why_STAR_Voting.md has a slide titled "You Never Have to Betray Your Favorite," with bullets like "Score both high — no forced ranking, no wasted vote" and "Honesty is your best ballot." (House style references slides by title, not number — see LINKS.md.) This page is the deep-dive that backs that claim up under cross-examination.

The challenge to the slide. "Isn't this overstated? RCV-IRV proponents claim that only in IRV do you never betray your favorite, and that STAR can make you betray your favorite. Are they right?"

The short answer. The slide is true in the everyday, practical sense, and overstated only as a strict theorem. And the RCV-IRV claim is false — RCV-IRV does not protect you from favorite betrayal; it protects a different thing. Below is why, with numbers.

2. Two promises that sound identical but aren't¶

Imagine you sincerely like A (favorite) > B (compromise) > C (worst).

Promise A — Favorite Betrayal Criterion (FBC)¶

"It is always safe to support your favorite the most. You never get a better result by pretending someone else is your favorite."

A method passes FBC if you can always put A at the top (rank 1, or score 5) without that choice ever backfiring. A method fails FBC if there are situations where you'd get a better outcome by insincerely ranking/scoring B above A — i.e., betraying your favorite A.

Promise B — Later-No-Harm (LNH)¶

"Adding or raising a LOWER choice can never hurt your favorite."

A method passes LNH if marking B and C at all (after A) can never cause A to lose. It fails LNH if expressing support for your second choice can pull the win away from your first choice.

The crucial difference¶

FBC is about your FIRST mark — is it safe to put your favorite on top?
LNH is about your LATER marks — is it safe to also support your backups?

They are not the same question, and — this is the punchline — no voting method can satisfy both at once (Section 6). So every method has to give up at least one. RCV-IRV gives up FBC. Cardinal methods (Approval/Score) give up LNH.

3. Why everyone conflates them¶

When an RCV advocate says "in Ranked Choice you never have to betray your favorite," they are almost always describing the feeling of Later-No-Harm: "I can rank my favorite first AND rank my backups, and ranking the backups won't hurt my favorite." That part is true of RCV-IRV.

But "you never have to betray your favorite" is the FBC sentence, and RCV-IRV fails that one. The advocate has swapped a true LNH claim for a false FBC claim without noticing. Your job in a debate is to gently separate the two.

4. Worked example: RCV-IRV fails favorite betrayal (Alaska 2022)¶

This is the cleanest real-world proof. Three candidates, roughly on a spectrum:

Peltola (Democrat / left)
Begich (Republican / center-right)
Palin (Republican / right)

Approximate first-choice tally:

Candidate	First choices
Peltola	~40%
Palin	~31%
Begich	~29%

What RCV-IRV does: Begich has the fewest first choices, so he's eliminated first. His ballots transfer (mostly to Palin, some to Peltola, some exhausted). In the final round Peltola beats Palin ~51.5% to 48.5%. Peltola wins.

But look at the head-to-heads (who beats whom one-on-one): - Begich beats Peltola (~53–47). - Begich beats Palin (~61–39).

So Begich beats both others head-to-head — he's the candidate a majority actually preferred (the Condorcet winner). RCV-IRV eliminated him in the first round because few people ranked him first. That's the center squeeze.

Now the favorite-betrayal part — the key move, read slowly:

Take the Palin voters. They ranked Palin 1st (honest), Begich 2nd, Peltola last. By being honest: - Begich (their 2nd choice) got eliminated, - Palin (their 1st) then lost the final, - so they ended up with Peltola — their LAST choice.

If enough of those Palin voters had betrayed Palin and ranked Begich 1st instead, Begich would not have been eliminated, and Begich would have won (he beats Peltola head-to-head). Begich is their 2nd choice — strictly better than Peltola, their 3rd.

Ranking their true favorite (Palin) first gave them their worst outcome. Betraying her gave them a better one. That is a favorite-betrayal failure — in a real, high-profile election. RCV-IRV did exactly the thing its advocates say only it prevents.

(Burlington, Vermont 2009 is the same story with a left-leaning twist: Montroll was the head-to-head winner, eliminated early for too few first-place votes — and this repo now carries the real 8,980-ballot reproduction, runnable, with the 750-voter raise that makes the winner lose.)

5. "But you admit STAR fails FBC too!" — yes, and here's the honest difference¶

Concede it cleanly: STAR is also not formally FBC-compliant. Equal Vote's criteria chart marks STAR ❌ on Favorite Betrayal — it's a binary pass/fail, with no percentage attached. (You may have seen "~98%" quoted, including in earlier drafts of this repo. That figure has no FBC source — it was borrowed from STAR's accuracy score, Voter Satisfaction Efficiency ~91–98%, which measures something else entirely. Don't use it.)

So instead of quoting a number we can't defend, we measured FBC directly — see fbc_simulation.py, which brute-forces every voter's best favorite-betrayal across thousands of random elections. Two honest results:

On raw frequency, neither method is FBC-proof — and STAR is not better than RCV-IRV. Under a realistic spatial model STAR is FBC-compliant in ~92–96% of elections and RCV-IRV in ~95–97% — essentially a tie, IRV slightly ahead. (FBC is an existence test, and STAR's score ballot simply offers far more betrayal ballots in which to find one that helps.) So "STAR fails FBC less often than IRV" is not a claim the numbers support — drop it.
The real, measurable difference is whether betrayal ever pays. Of the favorite-betrayals that actually change who wins, only ~2% help the voter under STAR — the other ~98% backfire — versus ~7–12% under RCV-IRV. Betrayal is several times more likely to pay off in IRV, and STAR's rare wins need extreme ballots plus near-perfect knowledge of everyone else's vote. That is the only defensible "98%": not "FBC-compliant 98% of the time," but "favorite betrayal backfires ~98% of the time you'd try it."

Where STAR's FBC leak comes from — the runoff. - Pure Score voting PASSES FBC. You can always give your favorite a 5; since the winner is just the highest total, maxing your favorite can never hurt you. - But pure Score has a different weakness: exaggeration ("I'll give my favorite 5 and everyone else 0 to inflate them"). That distorts results. - STAR adds the automatic runoff to neutralize that exaggeration. And it works — but the runoff is also what introduces the sliver of FBC vulnerability, because now your scores can change which two candidates are finalists, and in rare, delicately balanced cases moving your favorite's score can change the finalist pairing in a way that helps you. - This repo has that rare case worked, live, with numbers: the favorite-betrayal pair (BV2206/BV2207) — a 57-voter electorate where the Condorcet-winning compromise misses the runoff by 3 points, the nine double-fans have already equal-topped her (so raising her is impossible), and demoting their favorite 5 → 4 flips the finalist pairing and pays. It also demonstrates the fragility claim by construction: the betrayal needs at least 7 of the 9 coordinating (6 produces a finalist tie; 5 or fewer does nothing). - The mirror image — the IRV incentive, in its simplest form: Should I rank my favorite second?. Where the pair above shows the rare STAR construction, this shows the plain RCV-IRV case the OP always asks about — 12 honest Left voters elect their worst (Right) under IRV; 2 of them ranking the compromise first flips it; STAR and Ranked Robin elect the compromise from the honest ballots. The clean contrast between "IRV punishes honesty here" and "STAR doesn't."

So STAR's FBC leak is a deliberate trade: it gives up a sliver of FBC to buy a genuine majority finish and immunity to score-exaggeration.

Why the two failures differ in kind — not merely in frequency (this is the point): - STAR's failures are fragile and unactionable — extreme ballots, perfect information, and a betrayal that backfires ~98% of the time you'd try it. They essentially don't occur in real elections. - RCV-IRV's failure is systematic and predictable — the center squeeze — and it bites in exactly the competitive 3-viable-candidate races reform is meant to fix, where a squeezed wing can often see it coming (Alaska, Burlington).

One-liner: "Neither of us is favorite-betrayal-proof — on paper we fail about equally often. The difference: in STAR the betrayal backfires almost every time you'd try it; your center-squeeze failure is predictable enough that a whole wing has a reason to betray."

6. The theorem that ends the argument: you can't have both¶

Favorite-Betrayal and Later-No-Harm are provably incompatible. No deterministic ranked or rated method can satisfy both at the same time. (Woodall-style impossibility; the synced deck states it as "Many criteria are mutually exclusive, including 'Favorite Betrayal' and 'Later No Harm.'")

So every method must give up at least one:

Method	Favorite Betrayal (safe to top your favorite?)	Later-No-Harm (safe to mark backups?)
Choose-One Plurality	❌	n/a (only one mark)
RCV-IRV	❌ (center squeeze)	✅
Approval	✅	❌
Score	✅	❌
STAR	❌ rare — and betrayal backfires ~98% of the time it's tried	❌ rare

Read across the STAR row: it commits to neither criterion fully, accepting rare failures of each in exchange for a majority-backed, exaggeration-resistant result. The payoff isn't that STAR passes FBC more often than RCV-IRV (it doesn't — see 06_Other/simulations/) — it's that in STAR a favorite-betrayal almost never pays off, so honesty stays your safest ballot.

The reframe to leave them with: Later-No-Harm sounds nice, but it's the very property that forces center squeeze — guaranteeing your later choices never help your favorite is mathematically the same as guaranteeing a broadly-liked compromise can't be rescued by being everyone's strong second. STAR gives that up on purpose, because "your honest support for a compromise should be allowed to help elect them" is the better value.

The sharp-end version of this point comes from STAR's own co-inventor: Clay Shentrup, Later-no-harm is a bug, not a feature. His argument is exactly the reframe above — a criterion that guarantees your backup support can never help your favorite is precisely the mechanism that lets a factional candidate beat a broadly-preferred one, so passing LNH is a cost, not a virtue. (Advocacy source — Shentrup co-founded the Center for Election Science and helped originate STAR — but the logical point stands on its own, and it's the honest counter to "but RCV keeps Later-No-Harm.")

7. The debate — an RCV advocate vs a STAR advocate¶

The claim in the wild is rarely a calm question; it's an argument. So here's the same lesson as the confrontation you'll actually have — the RCV advocate making the strongest version of the case, the STAR advocate separating the two criteria and conceding what's true.

RCV Advocate: In Ranked Choice you never have to betray your favorite. Rank the person you love first, add your backups below, done — and supporting those backups can't hurt your favorite. STAR can't promise that.

STAR Advocate: You've just folded two different promises into one, and that's where this whole argument goes wrong. "Is it safe to rank my favorite first?" is one criterion — Favorite Betrayal. "Is it safe to also mark my backups?" is a completely different one — Later-No-Harm. RCV-IRV keeps the second promise. It does not keep the first.

RCV Advocate: That's a distinction without a difference. If marking my backups never hurts my favorite, my favorite is safe. What more is there?

STAR Advocate: The two come apart in real elections. "My backups don't hurt my favorite" says nothing about whether ranking my favorite first was safe in the first place — because in RCV-IRV your favorite can be eliminated for having too few first-place votes before your backups ever get counted.

RCV Advocate: When has that actually happened?

STAR Advocate: Alaska, 2022. Begich beat both other candidates head-to-head — a majority preferred him to Peltola, and a majority preferred him to Palin. But he had the fewest first choices, so IRV eliminated him in the very first round. Honest Palin voters — Palin first, Begich second, Peltola last — watched their second choice get cut, then their first lose the final, and ended up with Peltola: their last choice.

RCV Advocate: …and you're going to tell me ranking Begich first would have fixed that.

STAR Advocate: It would. Begich survives the first round and wins — he beats Peltola head-to-head. Those voters get their second choice instead of their last. Ranking their true favorite first is the thing that hurt them. That is a favorite-betrayal failure, in a real, high-profile race, in the system you just told me is the only one that prevents it.

RCV Advocate: Fine — but you can't sit there and claim STAR is favorite-betrayal-proof. It isn't. Your own criteria chart marks STAR ❌ on Favorite Betrayal.

STAR Advocate: Correct, and I won't pretend otherwise: STAR is not formally FBC-compliant. On raw frequency we fail it about as often as you do — some simulations put IRV slightly ahead. And I'm not going to quote you a "98% compliant" figure, because that number doesn't exist; it was borrowed from a different metric and it's wrong.

RCV Advocate: Then what exactly is your advantage?

STAR Advocate: Whether the betrayal ever pays. We measured it directly. Of the favorite-betrayals in STAR that would actually change the winner, about 98% backfire — you'd hurt yourself trying. Only ~2% help, and those need extreme ballots plus near-perfect knowledge of how everyone else votes. In IRV, betrayal pays several times more often — and your failure has a name and a shape, the center squeeze, so an entire wing can see it coming and has a standing reason to betray. Mine are lab constructions; yours turns up in exactly the competitive three-candidate races reform is supposed to fix.

RCV Advocate: So strip away the rhetoric and you're just choosing which criterion to fail.

STAR Advocate: We both are — that's a theorem, not a dodge. No method can satisfy Favorite Betrayal and Later-No-Harm at once. You keep Later-No-Harm and give up favorite-betrayal resistance. We give up Later-No-Harm on purpose — because guaranteeing your later support can never help your favorite is the same math that lets a broadly-liked compromise get squeezed out. We think your honest support for a second choice should be allowed to help elect them. That's the real disagreement, stated honestly — not "only one of us fails."

8. How to deploy it¶

Don't open with this. It's Voting 301. Raise it only when an RCV advocate makes the "only RCV avoids favorite betrayal" claim. With a general audience, the "You Never Have to Betray Your Favorite" slide stands on its own.
Lead by separating the two criteria before you argue — half the disagreement evaporates once "first mark" vs "later marks" is on the table.
Concede STAR fails FBC too — immediately, and without a percentage. Say "betrayal almost always backfires in STAR," not "STAR is 98% compliant." Your candor is the credibility that makes the center-squeeze point land.
End on the value, not the math: Later-No-Harm causes center squeeze; STAR trades it away so a consensus candidate can actually win.

9. A mental model to hold onto¶

Favorite Betrayal = the front door. Is it safe to walk your favorite in first? (RCV-IRV: no. Score/Approval: yes. STAR: almost always.)
Later-No-Harm = the back door. Is it safe to let your backups in too? (RCV-IRV: yes. Score/Approval: no. STAR: almost always.)
You cannot lock both doors. RCV-IRV locks the back and leaves the front open (center squeeze). STAR keeps both mostly shut and refuses to fully sacrifice either — and the front-door failures it does have almost never reward the burglar.

Cross-references¶

The worked STAR favorite-betrayal pair — BV2206 (honest: the tepid Condorcet winner misses the runoff) / BV2207 (nine voters demote their favorite and it pays), live on BetterVoting and LH-verified. The numeric backing for §5's "the leak is the runoff."
The "You Never Have to Betray Your Favorite" slide in Why_STAR_Voting.md (the claim this defends) and its Part 2 talking point #12.
are_equal_score_votes_discounted.md — the runoff / Equal-Support sibling of this concede-then-reframe argument.
LINKS.md → Full Deck 2025 ("RCV Common False Claims", "Alaska '22", "Burlington 2009", the pass/fail criteria slides); Why STAR 2.
fbc_simulation.py + 06_Other/simulations/README.md — the brute-force FBC measurement behind Section 5 (FBC compliance frequency and the betrayal works:backfires ratio, STAR vs RCV-IRV).
residual_vote_splitting.md — the same "only top-two advance" root cause seen as vote-splitting: STAR's self-inflicted bullet-voting / chicken-dilemma edge case, with the split_voting/05a–05b matched demo.