Proposition 14 opened the primary to every voter. But it shipped with a legacy constraint: each voter selects exactly one candidate. When similar candidates split a constituency, the wrong ones advance. We have the patch.
The entire change, from the voter's perspective:
Same ballot. Same machines. Same election night. The only change is the instruction at the top.
California's top-two primary was a major upgrade. It eliminated partisan gatekeeping, opened the primary to all voters, and guaranteed a competitive general election. But the upgrade shipped with a constraint inherited from the legacy system it replaced.
Each voter may mark exactly one candidate. This is the plurality constraint. When three or more candidates compete for the same constituency, their supporters split their votes and can collectively lose to a candidate with narrower support. This isn't a theory — it's a formally characterized failure mode with a name: vote-splitting.
The bug produces second-order effects. Strong candidates decline to enter races to avoid splitting votes with ideological neighbors. Donors make premature viability bets, creating a shadow primary driven by funding rather than preference. Incumbents benefit from fractured opposition. These are symptoms of the constraint, not of the candidates.
Approval voting modifies exactly one parameter: the number of candidates a voter may select. Under the current system, that value is hardcoded to 1. Under approval voting, it becomes [1, n]. Voters may approve of as many candidates as they wish. Each approval counts as one vote. Top two advance. Everything else stays the same.
The top-two architecture remains. The nonpartisan primary remains. All-voter participation remains. The only change is removing the artificial constraint on voter expression.
Top-two general election structure. Nonpartisan primary ballot. All-voter participation. Candidate party-preference labeling. Precinct-level tabulation.
Voters can express support for every candidate they find acceptable, eliminating the forced tradeoff that causes vote-splitting.
No rankings. No runoffs. No new rounds. No algorithm more complex than addition. No new ballot layout. No new equipment.
This is not a rhetorical flourish. It is an engineering fact. Every component in the current election infrastructure already supports this change.
Every optical-scan system certified for use in California already reads multi-marked ballots. These systems count votes per candidate, not votes per voter. A ballot with three candidates marked is counted as one vote for each. No new machines. No new certification.
Election management systems aggregate votes per candidate. Approval voting's tabulation requirement is identical: sum and rank. The only change is removing the overvote flag that treats multi-marked ballots as errors rather than valid expressions of preference.
"Vote for One" becomes "Vote for One or More." This is a print-run change that occurs with every election cycle regardless. Marginal cost: effectively zero.
Poll workers learn one new instruction: multi-mark ballots in voter-nominated primary races are valid. This is simpler than any alternative reform.
The voting reform space has multiple proposals. Here's the engineering comparison.
| Property | Approval | RCV / IRV | STAR | Top-4 + RCV |
|---|---|---|---|---|
| New hardware required | No | Often | Yes | Often |
| Precinct-summable | Yes | No | Partial | No |
| Ballot complexity | Same | Rankings | Scores | Rankings |
| Tabulation | Addition | Multi-round | Two-phase | Multi-round |
| Ballot exhaustion | None | Common | None | Common |
| Adopted in U.S. | Fargo, St. Louis | NYC, AK* | None | AK (under repeal threat)* |
| Centrist/consensus bias | Documented | Weak | Moderate | Weak |
| Voter comprehension | Trivial | Moderate | Difficult | Difficult |
* Alaska adopted Top-4 + RCV in 2020. A repeal measure failed by just 743 votes in 2024; a second repeal has already qualified for the 2026 ballot.
Computational social choice research — including Monte Carlo simulations measuring Voter Satisfaction Efficiency — consistently demonstrates that approval voting elects candidates closest to the center of the voter distribution. It maximizes aggregate voter utility. This isn't a partisan claim; it's a mathematical property of the mechanism.
The visualization below simulates score voting — which lets voters rate candidates on a scale. Approval voting is just score voting on a {0, 1} binary scale, so it inherits the same centrist-finding dynamics. Watch how score voting consistently finds the candidate nearest the center of the electorate, while plurality drifts toward polarized winners:
These simulations include approval voting directly alongside score, plurality, IRV, and others across thousands of elections. Approval voting consistently selects candidates with the broadest support, near the center of the voter distribution — performing comparably to its continuous cousin.
Vote-splitting rewards polarization. Approval voting rewards breadth. A candidate who is the genuine second choice of 60% of voters — but the first choice of only 15% — gets destroyed under plurality. Under approval voting, that candidate wins, because the mechanism correctly aggregates preferences.
Approval voting isn't theoretical. It has been adopted by voters, implemented by election administrators, and sustained across election cycles.
Think of California's election system as an OS. The Constitution is the kernel. Statutes are userspace. Proposition 14 was a major kernel upgrade — it replaced the old partisan-primary architecture with a unified top-two system. But the upgrade shipped with a legacy constraint inherited from the old codebase: the single-selection rule.
This measure patches that constraint at the only level where it exists: the Constitution. It doesn't modify the general election. It doesn't create new institutions. It doesn't alter the relationship between parties and the state. It changes one line in the kernel.
In software engineering terms: this is a one-line fix to a known, characterized, reproducible bug. The fix has been tested in production. It has no known regressions. It requires no dependency changes. It is the kind of patch that, once shipped, makes everyone ask why it wasn't in the original release.
The current system has a known, characterized, reproducible failure mode. We have a tested, zero-cost patch. The filing package is ready.