A voting mechanism is the rule that turns a group's individual preferences into a single collective decision. Two mechanisms can take the same ballots and reach opposite outcomes, so the choice of mechanism is itself one of the most consequential decisions any organization makes. For a DAO it is doubly so: the rule is encoded in a contract, runs without a referee, and is exposed to adversaries who can hold tokens, borrow them, or mint fresh identities at will.

Every mechanism sits somewhere on one central axis: capital-weighted against participation- or identity-weighted. Capital weighting (one token, one vote) is naturally resistant to fake accounts — buying influence means buying the scarce token — but it concentrates power in the wealthiest holders. Identity weighting (one person, one vote) is egalitarian but collapses the moment one actor can pose as many people. Most of the designs below are attempts to move along that axis without falling off either end. Aragon frames the same landscape by the "input space" a ballot gives the voter — the richer the input (a rank, a score, a budget of vote credits), the more faithfully it captures what people actually want.

Token-weighted (coin) voting

The default in on-chain governance: voting power is proportional to governance tokens held, and a proposal passes when support clears a threshold — often alongside a quorum (a minimum of the supply that must participate) and a proposal threshold (a minimum stake required to even put something to a vote). It is popular because it is simple, legible, and Sybil-resistant by construction.

Its weaknesses are equally well known:

• Plutocracy. Whales decide. Studies of large DAOs repeatedly find voting power concentrated in well under 1% of participants, with the long tail effectively unrepresented.
• Voter apathy. A single small holder's vote rarely changes an outcome, so most don't cast one; turnout in the single-digit percentages is common, which means a determined minority can carry a proposal.
• Vote-buying and bribery. As Vitalik Buterin argues in Moving beyond coin voting governance, a voter who borrows governance tokens has "no financial exposure" to the outcome — they can back a value-destroying decision and feel none of the damage. Voters are collectively accountable but never individually so.

The sharpest illustration is the Beanstalk attack of April 2022. An attacker took roughly a billion dollars in flash loans, converted them to the protocol's governance stake within a single transaction, instantly commanded a supermajority, and passed a proposal that drained about $182M from the treasury — all before the loan had to be repaid. The lesson DAOs took from it: voting weight must be snapshotted at a block in the past, never read live, or governance can be rented by the second. (CoinDesk)

Quadratic voting

Quadratic voting (QV) lets a voter cast multiple votes on an issue they care about, but the cost of votes rises with their square: one vote costs one credit, three votes cost nine, ten votes cost a hundred. Because the marginal cost of influence climbs steadily, people end up buying votes roughly in proportion to how much they care — so QV surfaces how much a voter cares, not just which side they are on, while making it ruinously expensive for any single party to dominate.

Its formal case was made by Steven Lalley and Glen Weyl in Quadratic Voting (2015): under price-taking behaviour QV is the only rule that is always efficient, and in a large electorate every symmetric equilibrium of the QV game converges on the efficient outcome, with the residual waste shrinking roughly as one over the number of voters — a direct answer to the "tyranny of the majority" that one-person-one-vote invites.

QV's roots are in corporate governance. Eric Posner and Glen Weyl first floated it in Quadratic Vote Buying, Square Root Voting, and Corporate Governance (2013), proposing that shareholders cast votes equal to the square root of their holdings so a large holder cannot simply steamroll the small ones, then developed the full argument in Quadratic Voting as Efficient Corporate Governance (University of Chicago Law Review, 2014): decoupling votes from shares and pricing them quadratically protects minority shareholders and, they argue, yields efficient outcomes "under reasonable conditions." Vitalik Buterin and Glen Weyl carried the idea into crypto in Liberation Through Radical Decentralization (2018), where a thousand votes cost a million credits — minority capture becomes prohibitively dear — but only if every voter is a verifiably separate person, a requirement squarely at odds with pseudonymous wallets. The idea has had real-world outings too: in 2019 the Colorado House of Representatives gave each legislator a budget of 100 vote credits to prioritize its appropriations backlog, and the quadratic cost forced them to reveal real priorities instead of rubber-stamping everything.

That identity assumption is QV's fatal flaw on-chain: split your tokens across many wallets and the quadratic penalty evaporates, so QV is only safe atop a working proof-of-personhood layer. The intensity it rewards can also be turned against it — Patty and Penn show in Uncertainty, polarization, and proposal incentives under quadratic voting (Public Choice, 2017) that once voters also set the agenda, QV rewards floating more divisive, higher-variance proposals; the same intensity-sensitivity that makes QV efficient on a fixed question pulls the menu of questions toward polarization. Its better-known cousin, quadratic funding (the engine behind Gitcoin Grants), applies the same square-root math to donations so that the number of backers matters more than the size of any one cheque.

There is a deeper, theory-side caveat to any mechanism that tries to read intensity rather than mere rank. Ehlers, Majumdar, Mishra and Sen prove in Continuity and Incentive Compatibility in Cardinal Voting Mechanisms (2016; Journal of Mathematical Economics, 2020) that, with no money changing hands, any cardinal voting rule that is both strategy-proof and continuous must collapse back to an ordinal one — it can use voters' rankings but not how strongly they feel. Cardinal schemes escape this only by attaching a genuine cost to intensity, which is exactly what QV's vote credits do and an ordinary score ballot does not.

Conviction voting

Conviction voting removes the deadline. Instead of a yes/no snapshot, voters stake tokens on the proposals they support and their conviction accumulates the longer the stake stays put; a proposal executes automatically once its built-up conviction crosses a threshold scaled to how much of the treasury it asks for. Support becomes a continuous signal rather than a moment, which makes last-minute, rented-capital swings useless — you cannot buy time-weighted conviction in one block.

The model has a rigorous backing. Michael Zargham of BlockScience recasts collective choice as a signal-processing problem in Sensor Networks and Social Choice (2018): individual preferences are continuously fused, like noisy sensor readings, into a smoothed estimate of what the community wants, so a vote becomes ongoing preference estimation rather than a discrete ballot — the formal reason conviction should grow with sustained stake. Pioneered by the Commons Stack and first run in production by 1Hive's Gardens, the approach has also been proposed for funding public goods straight out of protocol issuance — see the conviction-staked inflation funding discussion — on the argument that time-weighting raises the cost of bribery, since an attacker has to sustain a bribe rather than pay once. It suits continuous funding decisions but still rests on a capital-weighted base and is, by design, slow.

Futarchy

Futarchy, proposed by economist Robin Hanson, splits a decision into two questions and routes each to the tool best suited to it: "vote on values, but bet on beliefs." Token holders democratically choose a metric of success (say, treasury value a year out); prediction markets then decide which proposal is most likely to raise that metric, and the market's verdict carries the decision. It puts factual forecasts in the hands of those willing to put money behind them, but depends on liquid, hard-to-manipulate markets and a welfare metric that cannot be gamed. The market-design fine print matters as much as the idea: experimental economics has long shown that the format of a market changes how people behave inside it. Coppinger, Smith and Titus found that Dutch and first-price sealed-bid auctions — formats theory treats as equivalent — drew different bidding in the lab, with bidders straying from the risk-neutral predictions the theory assumed (Incentives and Behavior in English, Dutch and Sealed-Bid Auctions, 1980). For a system that hands the decision to a market, that is the warning: the betting rules are part of the mechanism, not a neutral wrapper around it — one reason futarchy remains more experimented-with than adopted.

Ranked-choice and other preference methods

When a vote has more than two options, how you collect and tally preferences matters as much as who votes. The main families:

• Ranked-choice / instant-runoff (RCV/IRV). Voters rank the options. If no option has a majority of first choices, the lowest is eliminated and its votes transfer to their next preference, repeating until one option holds a majority. This extracts far more information than a binary ballot and rarely "wastes" a vote on a losing favourite.
• Condorcet methods. Elect the option that would beat every other in a head-to-head matchup, when such an option exists — the most resistant to spoilers and strategic ranking.
• Borda count. Award descending points by rank position and sum them; simple, but sensitive to irrelevant alternatives.
• Approval and score voting. Cardinal methods: approve any number of options (approval), or score each on a scale (score/range). Easy to tally and friendly to consensus candidates — and their incentives can be made sharp: Shah and Zhou show in Approval Voting and Incentives in Crowdsourcing (ICML 2015) that pairing approval ballots with the right strictly-proper scoring rule yields the unique "no-free-lunch" scheme under which honestly reporting the set of options you find plausible is the optimal strategy.

No ranked method is perfect, and that isn't a bug to be fixed: Arrow's impossibility theorem (Kenneth Arrow, 1951) proves that with three or more options, no ranked voting rule can satisfy a short list of obviously-desirable fairness conditions at once. The practical takeaway is comparative, not defeatist — some methods (Condorcet) fail those conditions far less often than others (plurality), so "least bad for this decision" is the right question.

Liquid democracy and vote-escrow

Liquid democracy lets each voter either vote directly or delegate their weight to someone they trust — and delegations are transitive and revocable, so expertise can pool without anyone permanently handing over their voice. The modern framing comes from Bryan Ford's 2002 Delegative Democracy. It fights apathy by letting busy holders lend their weight to active delegates — though, as DAO delegation data shows, it can also quietly re-concentrate power in a handful of large delegates.

Vote-escrow (ve) models, popularized by Curve's veCRV, tie weight to commitment: lock tokens for up to four years and your voting power scales with how long you lock, then decays as the lock runs down. The aim is to hand the steering wheel to long-term aligned holders rather than mercenary, transient capital — at the cost of liquidity and, in practice, a secondary market in "bribes" for locked votes.

Off-chain and on-chain voting

Where a vote runs is a design choice in itself:

• Off-chain (e.g. Snapshot). Votes are wallet signatures over a past token snapshot — gasless and high-turnout, but the result is only a signal until someone (usually a multisig) executes it. Cheap participation, with a trust assumption at the end.
• On-chain (Governor-style). Proposals, votes, a timelock, and execution all live in the contract — fully trustless, but slower and more expensive to participate in.
• Bridges. Optimistic execution (post a passed off-chain result on-chain, executable unless challenged) and holographic consensus (a prediction market "boosts" high-signal proposals so a large DAO can decide without full quorum on everything) try to keep the cheapness of off-chain with the finality of on-chain.

The contract can also defend the clock itself. Wait-for-quiet voting — proposed by Dominic Williams and used in the Internet Computer's network governance — pushes a proposal's deadline back every time the leading outcome flips, so a large holder cannot lurk until turnout is thin and snipe the result in the final block; the vote closes only after a stretch of quiet in which the lead holds. It is the timing-side complement to reading voting weight from a past snapshot block — the fix DAOs adopted after Beanstalk.

Sybil resistance and identity

Every mechanism that moves away from pure capital weighting — quadratic, one-person-one-vote, reputation — needs a way to stop one actor from minting many identities. Proof-of-personhood aims to certify "one human, one account" without full doxxing: social-graph analysis (BrightID), curated registries with vouching (Proof of Humanity), or biometric uniqueness (World ID). It is the precondition for egalitarian voting on a permissionless network, and the reason token voting persists as the default: tokens give you Sybil resistance for free, while every alternative has to build an identity layer first.

Choosing a mechanism

There is no universally best rule. The choice is a trade-off between decentralization, resistance to manipulation, and broad participation — push hard on any one and the others give. Token voting buys manipulation-resistance and simplicity at the cost of equality; quadratic and identity schemes buy equality at the cost of needing a trustworthy identity layer; conviction and ve-models buy resistance to rented capital at the cost of speed and liquidity. Recent research even argues that no purely wallet-weighted rule can fully escape plutocracy against a rational attacker — which is why mature designs increasingly combine stake with something money can't cheaply buy: time, participation, or verified identity.

Voting on Caper

Caper runs two of the families above in sequence, and a proposal has to clear both. A proposal puts between two and five ranked options to the caper's members – at least one of which must be "Do nothing". Every ballot ranks all of them, and the contract folds the ballots with a Borda count, weighting each by (t·v)/(V·T): the voter's token balance multiplied by their accumulated vote tokens, over the vote-token supply times the circulating supply frozen when the proposal was raised. The leading option carries the tally only if it holds at least 1.5 / N of all weight cast – 75% with two options, 30% with five – and only if it is not the "Do nothing" option.

A ballot that passes does not execute anything. Where the winning option is an executable action it authorises someone to open the optimistic window – the market window – which locks the caper's trailing time-weighted average price as a baseline. The action goes ahead only if the price averaged over that window is at or above the baseline. Holding is consent; selling, hard enough and for long enough to hold the average down, is the veto.

One class of winner never reaches that second phase. Since 5 September 2026 an option can be typed DEBATE, which carries no action at all: it is ranked with the rest, it can win the count outright, and the proposal ends there. trigger_proposal refuses a DEBATE winner rather than opening a window with nothing to execute at the end of it, and creation asserts such an option carries no execution fields (contracts/logic/src/lib.rs). A slate of debate positions plus the mandatory "Do nothing" is therefore a caper putting a question to its members, settled by the same ranked count and never handed to the market – so a passing ballot no longer implies that anything happens. The full kind list is in Proposals.

Each half borrows a different family's problems. The ballot is a Borda count, so it inherits Borda's known weakness – sensitivity to irrelevant alternatives – which is why the option ceiling is five rather than open-ended. The threshold’s absolute figure does fall as options are added – 75% of cast weight on a two-option ballot, 30% on a five-option one – but the bar itself does not soften, because the most a Borda leader can take falls in exactly the same proportion. Under complete rankings each ballot contributes n(n − 1)/2 of weight and a unanimous first place takes n − 1 of it, so a leader’s ceiling is 2/n at every ballot size. A threshold of 1.5/n is therefore three-quarters of that ceiling, and half again an even split, whether the slate holds two options or five (apply_borda in contracts/common/src/lib.rs; settle_proposal in contracts/logic/src/lib.rs). Its weighting is neither pure token-weighted voting nor quadratic: multiplying stake by an earned, untransferable participation record means capital alone never carries a vote, though capital is not shut out of earning that record either, since vote tokens accrue on trades as well as on ballots.

The second half is optimistic governance – the decision is assumed to stand and the work is in objecting – and the usual contrast – that those systems hand the veto to a guardian – does not survive reading them. Aragon's Optimistic Token Voting plugin permissions who may propose, not who may object: the veto sits with token holders and carries when objections reach minVetoRatio of supply. The one that did put objection behind a bond was UMA's oSnap, and UMA retired it on 15 December 2025, since when it executes nothing at all. So the difference Caper turns on is not who holds the veto but what using it costs: a holder veto is a signature, while selling is a trade at a price, which is the futarchy instinct applied to one price series rather than to a pair of conditional markets. The trade is legibility for cost: there are no extra markets to bootstrap and no liquidity to fragment, but the signal is coarser, since a price that moved for unrelated reasons during the window is indistinguishable from a deliberate veto.

Two properties follow that are unusual enough to state directly. There is no quorum and no turnout floor anywhere in the sequence: the threshold is a share of the weight actually cast, not of the weight that exists, so a single ballot can carry a proposal to the window if nobody else votes – the deliberate inverse of the designs surveyed in quorum and threshold design. That is survivable only because the second phase does not need turnout to work: silence there resolves as consent, but so does every member who reads the proposal and declines to sell.

And objecting in the second phase is the same act as leaving: a member who would have exited over a proposal blocks it on the way out, at a real price rather than a free signature. What the design rests on is the cost of moving a price and holding it there – which is why both the baseline and the verdict are time-weighted averages rather than readings from a single moment. For the on-chain governance landscape this sits in, see DAO governance models and the cornerstone What is a DAO?.

References

Quadratic voting and cardinal mechanisms:

• Steven P. Lalley & E. Glen Weyl, Quadratic Voting — working paper (2015); condensed as "Quadratic Voting: How Mechanism Design Can Radicalize Democracy," AEA Papers & Proceedings 108 (2018).
• Eric A. Posner & E. Glen Weyl, Quadratic Vote Buying, Square Root Voting, and Corporate Governance — Harvard Law School Forum on Corporate Governance (2013).
• Eric A. Posner & E. Glen Weyl, Quadratic Voting as Efficient Corporate Governance — University of Chicago Law Review 81 (2014).
• Vitalik Buterin & E. Glen Weyl, Liberation Through Radical Decentralization (2018).
• John W. Patty & Elizabeth Maggie Penn, Uncertainty, polarization, and proposal incentives under quadratic voting — Public Choice 172 (2017): 109–124.
• Lars Ehlers, Dipjyoti Majumdar, Debasis Mishra & Arunava Sen, Continuity and Incentive Compatibility in Cardinal Voting Mechanisms — Université de Montréal (2016); Journal of Mathematical Economics 88 (2020).

Preference methods, auctions, and conviction:

• Nihar B. Shah & Dengyong Zhou, Approval Voting and Incentives in Crowdsourcing — Proceedings of ICML (2015).
• Vicki M. Coppinger, Vernon L. Smith & Jon A. Titus, Incentives and Behavior in English, Dutch and Sealed-Bid Auctions — Economic Inquiry 18 (1980).
• Michael Zargham (BlockScience), Sensor Networks and Social Choice (2018) — the signal-processing basis for conviction voting.
• Conviction-staked inflation funding — Ethereum public-goods funding discussion.
• Dominic Williams, Using "wait-for-quiet" voting — deadline extension as a defence against last-minute swings.

Further reading:

• Aragon, A taxonomy of voting methods (2021).
• Vitalik Buterin, Moving beyond coin voting governance (2021).
• CoinDesk, Attacker Drains $182M From Beanstalk (2022) — the flash-loan governance attack.
• The Colorado Sun, Colorado's experiment with quadratic voting (2019).
• Robin Hanson, Futarchy: Vote Values, But Bet Beliefs – Hanson's original statement of the proposal, live at GMU (re-checked 30 August 2026).
• Bryan Ford, Delegative Democracy (2002).
• Curve, Voting Escrow (veCRV).
• Wikipedia, Arrow's impossibility theorem.