What it is
Schelling-point voting is a way to get a group of anonymous, self-interested people to reveal the truth about something (was this evidence real? did this bet resolve yes or no?) without any of them trusting each other. It works by rewarding voters not for being "right" in some external sense, but for voting the same way as the majority — named after economist Thomas Schelling's idea of a "focal point" that people converge on without communicating, because it's the obvious, natural answer.
How it works
- A dispute or question is submitted (e.g., "did this smart contract's outcome match reality?" or "was this piece of content plagiarized?") along with a stake/deposit from the party raising it.
- A pool of jurors is selected, usually pseudo-randomly, weighted by how many tokens they've staked in the system — more stake means a higher chance of being drawn to vote (and more at risk).
- Jurors review the evidence and submit a vote, frequently using a commit-reveal scheme: first submitting a hidden hash of their vote, then later revealing it, so no juror can just copy others' votes as they come in.
- Once voting closes, votes are tallied. Jurors who voted with the majority get rewarded (a share of fees, and/or the losing side's forfeited stake); jurors who voted with the minority lose part or all of their staked tokens.
- Because everyone's payout literally depends on matching what most other honest people will independently conclude, the dominant strategy is to vote your honest best guess of the truth — that's the "obvious" schelling point everyone converges to.
- Larger or more subjective disputes can be escalated to bigger juror panels or appeal rounds, with stakes rising at each level to make attacking the system more expensive the higher it escalates.
Why designers use it
- Produces a trustworthy, decentralized answer to disputes without needing a single trusted judge or oracle.
- Well suited to subjective or hard-to-verify questions (contract interpretation, content moderation, real-world event outcomes) where a formula can't decide automatically.
- Game-theoretically discourages collusion: coordinating a lie is hard when you don't know who else is voting, and the honest majority answer is the safest bet regardless.
- Scales judiciously — low-stakes disputes use small, cheap juror panels; high-stakes disputes escalate to larger, more expensive, harder-to-corrupt panels.
Failure modes
- 51% capture: if one entity accumulates enough staked tokens to dominate juror selection, they can force the "correct" answer to become whatever they say, breaking the honesty guarantee.
- Voter apathy or ignorance: jurors have an incentive to guess/free-ride on others' judgment rather than genuinely investigate, especially for low-value disputes, degrading answer quality over time.
- Ambiguous questions: schelling-point voting works best when there's one obvious correct answer; genuinely ambiguous or politically contested questions can split honest jurors, causing the mechanism to converge on a wrong or arbitrary answer.
- Bribery attacks: an attacker can openly pay jurors to vote a certain way (a "P+epsilon" bribe slightly above the honest reward), which can be economically rational for jurors even if the outcome is dishonest, undermining the incentive design in edge cases.
- Low liquidity/low stakes early on: a new schelling-point system with a low-value token is cheap to attack, since the cost of buying enough voting stake to sway outcomes is low.
What to check before using it
- Is the token/stake pool large and distributed enough that acquiring a controlling share is prohibitively expensive?
- Are questions submitted to the system genuinely objective enough to have a clear focal-point answer, or are they too subjective for the mechanism to work well?
- Is there a commit-reveal step to prevent late voters from free-riding on early votes?
- What's the appeal path, and are higher-stakes rounds expensive enough to deter a well-funded attacker from just escalating repeatedly?
- Are juror rewards/penalties large enough to justify the time cost of actually reviewing evidence, rather than voting blind?