---
title: "Schelling points and coordination"
url: "https://caper.network/wiki/economics/schelling-points"
updated: 2026-08-17
license: CC-BY-4.0
license_url: "https://creativecommons.org/licenses/by/4.0/"
---

# Schelling points and coordination

|  |  |
| --- | --- |
| **Topic** | Game theory; coordination; decentralized oracles |
| **Named after** | Thomas Schelling |
| **Key mechanism** | Convergence on a salient option through mutual expectation, without communication |
| **Related** | [Voting mechanisms](/wiki/governance/voting-mechanisms) · [Proposals](/wiki/governance/proposals) · [Reputation and information asymmetry](/wiki/economics/reputation-and-information) · [Commons governance and collective action](/wiki/economics/commons-governance) · [UMA (Optimistic Oracle & the DVM)](/wiki/daos/infrastructure/uma) |

A **Schelling point** (or _focal point_) is the option that people tend to choose in a coordination problem when they cannot communicate, because it stands out as the answer each participant expects the others to expect. The idea is named after the game theorist Thomas Schelling, who observed that strangers asked to meet in a city without agreeing on a time or place often succeed anyway by converging on a salient landmark at a salient hour. Nothing enforces the choice; it works purely through mutual expectation.

Focal points matter to decentralized systems because permissionless networks routinely ask dispersed, non-communicating actors to converge on a single answer — a vote outcome, a protocol decision, or a reported data value. This article reviews the experimental evidence on when focal points work, how that logic was turned into an oracle design, and why its incentive assumptions have known limits.

## Focal points in the laboratory

Experimental work shows that focal points are powerful but conditional. Crawford, Gneezy and Rottenstreich replicated Schelling's classic result under symmetric payoffs: in a "Chicago skyscrapers" matching game where both players stood to receive a hypothetical $100 for coordinating, [90 percent of subjects chose the salient landmark](https://doi.org/10.1257/aer.98.4.1443) (the Sears Tower), for an expected coordination rate of 82 percent. Introducing a minute payoff asymmetry — $100 for one player versus $101 for the other — largely collapsed the focal point: only about 60 percent still chose the landmark, and expected coordination fell to 52 percent, barely above the roughly 50 percent achievable by a payoff-blind mixed strategy. A moderate asymmetry ($100 versus $110) eroded it further, to 48 percent choosing the landmark and roughly 50 percent coordination.

Paid experiments with abstract X and Y labels confirmed the pattern: symmetric labeled games reached 64 percent expected coordination, while a slightly asymmetric version ($5 versus $5.10) fell to 38 percent, performing essentially like games with no labels at all. Subjects did not "freely dispose" of the negligible difference in order to exploit the salient label; they attended to their own payoffs, destroying the shared focal point even though ignoring the one-dollar gap would have raised everyone's expected earnings. The authors account for the observed pattern of miscoordination with a nonequilibrium level-k model whose anchoring type is biased toward payoff-salient choices.

The picture is not uniformly bleak. Isoni, Poulsen, Sugden and Tsutsui studied tacit bargaining problems — coordination games with genuine conflict of interest and no communication — in which each player's strategies were framed as proposals over which valuable objects she would take. There, [payoff-irrelevant spatial and relational cues acted as powerful focal points](https://doi.org/10.1016/j.euroecorev.2012.12.005), reducing coordination failures even in games with no efficient and equal division, and systematically producing unequal payoff distributions that players accepted rather than miscoordinate. In bargaining-framed settings with strong cues, focal points can survive payoff conflict rather than collapsing under it — a partial counterweight to the Crawford, Gneezy and Rottenstreich findings.

## SchellingCoin: coordination as an oracle

In 2014, Vitalik Buterin proposed [SchellingCoin](https://blog.ethereum.org/2014/03/28/schellingcoin-a-minimal-trust-universal-data-feed), a mechanism for bringing external data such as the ETH/USD price onto a blockchain without any trusted authority, by paying participants to converge on the same answer. The design rests on a game-theoretic wager: absent communication, "the truth is arguably the most powerful Schelling point out there," so honest reporting is the answer everyone expects everyone else to give. Participants first submit hashed values, then reveal them; the revealed values are sorted, submissions falling between the 25th and 75th percentile are rewarded, and the median becomes the consensus output. Buterin described the resulting equilibrium as an "infinitely recursive chain of logic" backed by nothing but mutual expectation — which is both its power and its fragility.

A companion post supplied a [concrete smart-contract implementation](https://blog.ethereum.org/2014/06/30/advanced-contract-programming-example-schellingcoin): 100-block epochs split into a commit phase (blocks 0–49, submitting a hash of sender and value) and a reveal phase (blocks 50–99). The commit-reveal structure is what enforces the focal-point game — during commitment no participant can see anyone else's vote, so the rational uninformed strategy is to submit the true value everyone else is expected to submit. Buterin suggested the pattern generalizes beyond price feeds to any publicly verifiable data feed, with decentralization and participant diversity as the main safeguards.

SchellingCoin remained a design sketch, but its commit-reveal focal-point game is now in production for _subjective_ disputes rather than price feeds: [Kleros](/wiki/daos/infrastructure/kleros) pays randomly-drawn, stake-weighted jurors to converge on the answer they expect other honest jurors to give, rewarding the coherent majority with tokens taken from the incoherent minority — a standing court built on exactly the "truth is the strongest Schelling point" wager.

## Attack vectors and the limits of rationality

Buterin acknowledged several attack vectors from the outset. Any entity controlling more than roughly half of the submissions can unilaterally set the median, so the mechanism inherits a majority-collusion assumption. Participants with a financial stake in the reported value can engage in "micro-cheating," subtly skewing their answers, and even small coalitions offering side payments can create competing focal points. The implementation post added structural concerns: if a single exchange dominates trading, all votes cluster around its price, so the focal point silently becomes a single point of failure when that exchange is disrupted; gas costs at epoch boundaries make direct participation expensive for small players, pushing toward pooling and thus re-centralization; and downstream uses such as stable-value assets built on the feed remain exposed to strategic large trades that move the focal value itself.

Ford and Böhme generalized the critique. Permissionless systems, they argue, cannot bound the incentives participants face: out-of-band payments, bribery, and side bets — such as shorting a token while attacking it — can make protocol-deviant behavior the profit-maximizing choice while remaining invisible to in-protocol incentive analysis. They describe bribery and discouragement attacks in which an adversary makes honest participation economically unattractive, so that rational players exit or defect and incentive-based security arguments break down. Their conclusion is metacircular: [assuming all participants are rational profit-maximizers creates exploitable blind spots](https://arxiv.org/abs/1910.08820), so that in open systems "it is rational to be irrational" — attackers who look irrational inside the model may be perfectly rational outside it. They recommend designing for resilience rather than incentive-compatibility alone: Byzantine fault tolerance against arbitrary behavior, behavioral realism, and social and institutional safeguards alongside technical ones. The point bears directly on Schelling-point mechanisms: an equilibrium sustained only by mutual expectation fails once a coalition can be paid to converge on a false focal point.

## A focal point measured on-chain

Both the laboratory finding and the theoretical critique have a counterpart that can be read straight off a public ledger. [UMA](/wiki/daos/infrastructure/uma)'s Data Verification Mechanism is, in [UMA's own description](https://docs.uma.xyz/protocol-overview/how-does-umas-oracle-work), "a Schelling point mechanism where UMA stakers resolve disputes in return for staking rewards": votes are committed encrypted for 24 hours, revealed over the following 24, and the outcome the majority lands on is paid while everyone who voted otherwise, or failed to vote at all, is slashed. Because every reveal is an Ethereum event, how far a given question actually converged is a public number rather than a survey answer.

Round 10085 of the DVM, which settled on 25 March 2025, is a useful natural experiment: it carried eight disputed questions at once, all relayed from Polygon by a prediction-market adapter. The contract froze that round's thresholds at 5,000,000 UMA of minimum participation and 11,422,874.44 UMA of minimum agreement, the latter being exactly half of the 22,845,748.88 UMA staked at the time (`rounds(10085)` on [VotingV2](https://eth.blockscout.com/address/0x004395edb43EFca9885CEdad51EC9fAf93Bd34ac)). Six of the eight converged the way a working focal point should. A college basketball game, two NCAA bracket questions, a cabinet-appointment question, a sanctions question and an audit question drew between **98.85 and 99.99 percent** agreement across roughly 21 million UMA of revealed stake.

The other two were the same question asked twice: _"Ukraine agrees to give Trump rare earth metals before April?"_, written to resolve Yes if the two governments announced "any deal … that explicitly involves Ukrainian rare earth elements", with an announcement qualifying "regardless of if/when the deal is enacted". Decoding every `VoteRevealed` event of the round gives the split. On the first request, 514 voters revealed 21,000,063 UMA: **12,292,521 (58.54 percent) voted Yes** against 8,706,371 (41.46 percent) for the reserved "too early to resolve" value. On the second, 510 voters revealed 20,999,660 UMA: **11,465,793 Yes (54.60 percent)** against 9,532,668 too-early (45.39 percent). [Both resolved to Yes in a single transaction](https://eth.blockscout.com/tx/0x9de6a2dd7cb1192a15f73915ae91cdff23596fe8e497887e615f38ab566e954b) at block 22,120,202. The second cleared the round's agreement requirement by **42,918 UMA**, a margin of 0.19 percent of everything staked.

## What the split shows

The contrast inside one round is the finding. Six questions that nobody held a position in converged almost perfectly; the single question a large amount of money turned on split roughly 55/45 and settled on the narrowest of margins. That is the Crawford, Gneezy and Rottenstreich result at production scale: the focal point holds while payoffs are close to symmetric and comes apart once they are not. Nothing about the truth of the rare-earth question was harder to observe than the result of a basketball game; what differed was that voters had reason to care which way it went.

The reveal data also shows where the deciding stake sat. The three largest Yes voters held 1,829,777, 1,811,522 and 1,442,193 UMA, together **5,083,492 UMA**: 22.25 percent of all staked supply, and 44.3 percent of the Yes bloc on the narrower request. Because that request cleared its requirement by only 42,918 UMA, any one of the three switching or simply abstaining would have left it unresolved. Reporting at the time by [The Defiant](https://thedefiant.io/news/defi/polymarket-s-usd7m-ukraine-mineral-deal-debacle-traced-to-oracle-whale) and [CoinDesk](https://www.coindesk.com/markets/2025/03/27/polymarket-uma-communities-lock-horns-after-usd7m-ukraine-bet-resolves) attributed the bloc to one holder voting from several wallets; the ledger itself shows only the addresses and the arithmetic.

Four weeks later, the threshold moved. UMA's agreement requirement had stood at 50 percent of staked supply since VotingV2 was configured in February 2023; on [20 April 2025](https://eth.blockscout.com/tx/0xb4435c2aff6e61f73f722d342e417ff2eaff5b2fcba25b5bcd570aed6cd77f35) a governance proposal executed on-chain raised it to 65 percent, leaving the participation floor at 5,000,000 UMA. Under that setting, neither rare-earth request would have resolved in round 10085. A higher bar does not turn an unshared focal point into a shared one. What it changes is that a failure to converge surfaces as an unresolved question, instead of settling at 54.6 percent.

## Relevance to Caper

Caper governance depends on the same coordination logic. Token holders decide [proposals](/wiki/governance/proposals) through [on-chain voting](/wiki/governance/voting), and any execution path that consumes externally reported data inherits focal-point dynamics. The experimental record cautions that convergence on the "obvious" outcome is reliable mainly when participants' payoffs are close to symmetric; holders who acquired their tokens at different prices, or who differ in stakes and information (see [reputation and information asymmetry](/wiki/economics/reputation-and-information)), face exactly the kind of payoff asymmetry that eroded coordination in the laboratory. Designing [voting mechanisms](/wiki/governance/voting-mechanisms) that anticipate asymmetric incentives and out-of-band pressure is therefore a [security consideration](/wiki/dao-governance/concepts/analysis/dao-security-and-governance-attacks), not only a fairness one.

## References

1. Vincent P. Crawford, Uri Gneezy, and Yuval Rottenstreich (2008). [The Power of Focal Points Is Limited: Even Minute Payoff Asymmetry May Yield Large Coordination Failures](https://doi.org/10.1257/aer.98.4.1443). American Economic Review, 98(4): 1443–1458.
2. Andrea Isoni, Anders Poulsen, Robert Sugden, and Kei Tsutsui (2013). [Focal points in tacit bargaining problems: Experimental evidence](https://doi.org/10.1016/j.euroecorev.2012.12.005). European Economic Review, 59: 167–188.
3. Vitalik Buterin (2014). [SchellingCoin: A Minimal-Trust Universal Data Feed](https://blog.ethereum.org/2014/03/28/schellingcoin-a-minimal-trust-universal-data-feed). Ethereum Foundation Blog.
4. Vitalik Buterin (2014). [Advanced Contract Programming Example: SchellingCoin](https://blog.ethereum.org/2014/06/30/advanced-contract-programming-example-schellingcoin). Ethereum Foundation Blog.
5. Bryan Ford and Rainer Böhme (2019). [Rationality is Self-Defeating in Permissionless Systems](https://arxiv.org/abs/1910.08820). arXiv preprint arXiv:1910.08820.
6. UMA, [How does UMA's oracle work?](https://docs.uma.xyz/protocol-overview/how-does-umas-oracle-work). The DVM described by its authors as a Schelling point mechanism, with the commit-reveal timing and the agreement threshold.
7. Ethereum mainnet, [UMA VotingV2 (0x0043…34ac)](https://eth.blockscout.com/address/0x004395edb43EFca9885CEdad51EC9fAf93Bd34ac). Source for the round-10085 parameters (`rounds(10085)`), the `VoteRevealed` and `RequestResolved` logs behind the tallies above, and the `GatAndSpatChanged` events of 24 February 2023 and 20 April 2025.