Almost every DAO proposal is a yes/no question, and almost every DAO discussion about voting is about who gets weight – one token one vote, delegation, vote escrow. The ballot method is the other half of the machine: given the weights, what shape of preference may a voter express, and what arithmetic turns those expressions into a winner. Once a proposal has three or more live options – three budget levels, four candidate service providers, five parameter values – the counting rule stops being an implementation detail and starts picking outcomes.

This page covers the ordinal and set-valued families that DAOs reach for once yes/no runs out: approval, weighted, and the ranked rules (Borda, Condorcet/Copeland, instant-runoff), plus the multi-winner methods used to fill committees and fund grant slates. Quadratic voting and funding has its own page; it is a cardinal rather than an ordinal method and carries an identity dependency the others do not.

The counting rule is part of the mechanism

Which rule fits depends on what the vote is for. Cathy Barrera, founding economist of the consultancy Prysm Group, separates two reasons to hold one (What is Voting For?, 2018). Voting for preference combines members' different wants into a single choice, as when friends pick a restaurant. Voting on information pools what members know when they share a goal and disagree about the facts, as a jury does when it decides guilt. Most real votes mix the two, and she argues that the rules should follow from the mix: who may vote, how options reach the ballot, what threshold carries, and what voters need to know first. A criminal jury in the United States needs a unanimous verdict to convict because the system judges convicting an innocent person worse than freeing a guilty one. The two theorems below concern the first kind of vote: both are about combining rankings that express what voters want.

Two results from social choice theory bound what any ballot method can promise, and both are worth stating before comparing rules, because most governance-forum arguments about voting methods are really arguments about which of these two prices to pay.

Arrow's impossibility theorem establishes that no rule for aggregating individual rankings into a group ranking can simultaneously satisfy a short list of reasonable-looking conditions – unrestricted domain, unanimity, independence of irrelevant alternatives, and non-dictatorship – once there are at least three alternatives (Stanford Encyclopedia of Philosophy). Every ranked method in use is therefore a choice about which condition to give up, not a method that escaped the tradeoff.

Gibbard–Satterthwaite adds that over three or more options, any deterministic, non-dictatorial rule with a full range is manipulable: some voter, in some profile, does better by submitting a ranking that is not their honest one. A DAO cannot buy its way out of strategic voting by picking a cleverer tally – only relocate where the strategy shows up. See mechanism design for the general framing and the incentive-compatibility criterion these results constrain.

The single-winner families

Plurality (single choice). Each voter backs one option; the option with the most weight wins. Cheap and legible, and the default nearly everywhere. Its defect with three or more options is vote-splitting: two similar options divide the support that either alone would have beaten a third with, so the winner can be the option a majority actively opposes.

Approval. Each voter marks the subset of options they find acceptable – a threshold judgement, not a ranking. Approval sidesteps vote-splitting without asking for a full ordering, and it composes well with proposals whose options are not mutually exclusive. What it cannot express is intensity or order: as the Stanford Encyclopedia puts it, ranking a set of candidates and selecting the ones that are approved “are two different aspects of a voters overall opinion… They are related but cannot be derived from each other.”

Weighted. Each voter splits their voting power across options in whatever proportions they choose. This is the cardinal cousin of approval – it does capture intensity – and it is the natural fit for allocation questions (how to split a budget) rather than selection questions (which option wins).

Borda count. Voters submit a full ranking and each position pays fixed points: with n options, “give n−1 points to the candidate ranked first, n−2 points to the candidate ranked second, … and 0 points to candidate ranked last.” Because every ballot contributes the same total, Borda measures average standing across the whole field rather than first-place popularity, which is why it tends to elect broadly acceptable compromise options.

Condorcet methods, including Copeland. A Condorcet winner is an option that beats every other option in a head-to-head comparison. When one exists it is a strong claim to legitimacy; the difficulty is that it need not exist, because majority preference can cycle (A beats B, B beats C, C beats A). Copeland's rule resolves this by scoring each option's win–loss record across all pairwise matchups and electing the option that maximises it.

Instant-runoff (ranked choice, the Hare rule). Count first preferences; if no option clears 50%, eliminate the option with the fewest first preferences and transfer those ballots to their next surviving preference; repeat. This is the “ranked choice” type on Snapshot, described in its voting-types documentation.

Where each rule breaks

Every family has a signature failure mode, and knowing which one a DAO has adopted is more useful than arguing about which rule is best in the abstract.

• Borda – burying. Borda is unusually easy to manipulate by ranking a strong rival last rather than honestly. The Stanford entry works a case where a voter who prefers a losing option can flip the result purely by reordering the options below her favourite, making “it in the interest of the third voter to misrepresent her preferences.” Borda is also sensitive to clones: adding near-duplicate options changes the point spread even if nobody's real preferences changed.
• Condorcet – cycles. Majority preference has no guarantee of transitivity, so a Condorcet method needs a documented tiebreak (Copeland, Schulze, ranked pairs) that will decide a nontrivial share of real elections. A DAO that adopts “the Condorcet winner wins” without specifying the cycle rule has not finished specifying its governance.
• Instant-runoff – monotonicity failure and the no-show paradox. The Hare rule violates monotonicity: gaining support can cost an option the win, because moving up in first preferences changes who is eliminated first. It also admits the no-show paradox, where voters would have preferred the outcome of an election in which they did not participate. Both are counterintuitive enough that a DAO discovering them after a contested vote will have a legitimacy problem, not just a maths problem.
• Approval – the threshold is strategic. Nothing in approval tells a voter where to set their acceptance bar, and the optimal bar depends on how everyone else sets theirs. Approving your second choice may elect it over your first; approving only your first may elect your last.
• Plurality – splitting. Covered above, and the reason most of the other rules exist.

Token weight changes which pathologies bite

Classical social choice assumes one voter one ballot. DAO ballots carry stake, and three differences follow that the textbook analysis does not cover.

The tally is computable in advance. On-chain and Snapshot votes are public while voting is open, so a large holder can see the running result and compute the exact ranking that maximises their outcome before the deadline. Strategic manipulation in the Gibbard–Satterthwaite sense is a coordination problem for ordinary voters and a spreadsheet exercise for a whale. This is the strongest practical argument for private and shielded voting, and it hits ranked rules harder than plurality because a ranked ballot has more degrees of freedom to exploit.

Small-electorate pathologies are the normal case. Monotonicity failures and Condorcet cycles are rare in large public elections and much less rare when a handful of delegates carry most of the weight. A DAO where five addresses decide the outcome is exactly the regime where these effects show up – see voter apathy for how concentrated real turnout is.

Ballot design measurably moves weight. Balietti, Saggese & Strohmaier's Voting Biases in Decentralized Autonomous Organization (DAO) Governance (July 2026) links voting-power distribution to three observable features of a proposal's choice list and finds all three carry a premium: choices selected by the proposal's own author (+58.8%), choices phrased as approval rather than rejection (+27.1%), and choices listed first (+7.7%). Their conclusion is the one that matters for anyone writing a governance framework – ordering, author signals and vote visibility “should be treated as institutional design choices, not neutral implementation details.”

What DAO platforms actually implement

Snapshot ships six voting types and its documentation is explicit that the type “calculates the outcome of the proposal” while the separate strategy calculates individual voting power. Two implementation details are load-bearing and routinely missed. Its approval type gives each approved choice the voter's full power – “if user selects two choices, each choice will receive the total voting power of the user” – so approval on Snapshot is not a budget the voter divides, and totals across choices exceed the electorate's weight. Its ranked choice type is instant-runoff, and Snapshot lists the limitation itself: it “will only determine one winner, doesn't work well to select 2 or more winning options.”

ENS built a custom pairwise tally for its Service Provider Program rather than use a stock type. Delegates drag providers above or below a NONE BELOW line – “projects above 'NONE BELOW' are approved; those below are disqualified” – then rank the approved ones by funding priority, with “the algorithm compares each project against the others to determine the final result” and providers able to submit a basic and an extended budget that can be ranked separately (EP 6.10). That is a Copeland-family pairwise comparison with an explicit no-funding threshold and a budget cap bolted on. It selected eight providers for Season 2 against a budget raised from $3.6M to $4.5M a year (SPP2 stream implementation thread, June 2025). See ENS DAO.

Polkadot uses the sequential Phragmén method, “a multi-winner election method introduced by Edvard Phragmén in the 1890s,” inside Nominated Proof-of-Stake “to elect validators based on their own self-stake and the stake that is voted to them from nominators” (Polkadot wiki). It was also used for Council elections under Governance V1, now deprecated. See Polkadot.

Radix's DAO has adopted Majority Judgment, a graded method Michel Balinski and Rida Laraki published in 2007, for every election it holds, starting with the seven seats of its Permanent Radix Accountability Council. Each voter grades every candidate on one scale from Excellent to Poor, and candidates are seated in order of the grade they qualify at. Textbook Majority Judgment takes a candidate's median grade; the Radix framework takes the highest grade that three-fifths of the voting power cast places the candidate at or above, to match the 60% approval threshold of its other election mechanism, and seats no one who qualifies below Good. The election contract is deployed on Radix mainnet and has not yet run. The Majority Judgment article sets out the variant term by term.

Multi-winner selection is a different problem

Choosing a five-member security council, a grants slate, or a set of funded service providers is not the same question as choosing one option, and running a single-winner rule repeatedly gets it wrong in a specific way: it tends to hand every seat to the largest coherent bloc. A 51% majority that runs the table on five sequential single-winner elections takes all five seats.

The multi-winner family exists to prevent that. Single transferable vote transfers surplus above a quota as well as votes from eliminated candidates, so a bloc holding 40% of the weight elects roughly 40% of the seats rather than none. Sequential Phragmén attacks the same goal by spreading each voter's “load” as evenly as possible across the winners it elects; Janson's Phragmén's and Thiele's election methods is the standard modern treatment. Where the constraint is a budget rather than a seat count – the usual DAO case – the selection becomes a knapsack problem, which is what ENS's approve-then-rank-then-cap procedure is solving.

The practical rule for a DAO: if the proposal fills more than one slot, do not reach for the ranked-choice type. It was built to produce exactly one winner, and its own documentation says so.

How Caper approaches this

Caper does not offer a menu of voting types. Every proposal is settled by one rule, fixed in the contract: a Borda count over complete rankings, weighted by each voter's vote weight. A ballot has to rank every option on the proposal – partial ballots are rejected, as are duplicate rankings – and position i in a ballot of n options contributes (n−1−i) times that voter's weight, the classical Borda scale, with the last-ranked option contributing nothing.

Two consequences fall out of that arithmetic, and the first is easy to misread. Because every ballot spends the same number of points however it ranks them, an option's share is a share of the Borda pool rather than a share of the electorate – and the most any option can take of that pool is 2/n, reached only when every voter ranks it first. The pass threshold of 1.5/n is therefore three-quarters of unanimity at every ballot size, not a bar that softens as options are added. Stated in members rather than points: where the voters who do not back an option rank it last, it takes 75% of the weight ranking it first to carry, on a two-option ballot and on a five-option ballot alike.

The second is that “none of these” is not something a proposal author opts into. Caper appends a single no-op option, titled Do nothing, to every proposal as its last option; the proposal form shows it as a fixed row marked “Always included” and offers no equivalent flag on the author's own options. So n is always one more than the author wrote, the smallest possible ballot is one proposed action against Do nothing, and if the Borda leader is that option the proposal settles with no winner and nothing to execute. Rejection is a result the count produces, not a quorum it fails to reach.

The count is only the first of two gates, and that changes what the ballot method is for. A tally that clears the threshold does not execute anything; it makes the proposal triggerable. trigger_proposal then locks the caper’s trailing time-weighted average price as a baseline and opens the market window, and resolve_proposal integrates the price over that window and passes only if the window’s own TWAP is at or above the baseline. Every execution path binds the resulting flag. So the counting rule decides what is on the table and the market decides whether it happens – which is why a compromise-seeking rule is a defensible choice here in a way it would not be if the tally were final. See optimistic governance and execution.

A third outcome joined the two above on 5 September 2026: a DEBATE option (action kind 6) carries no recipient, amount or target and executes nothing by construction, and trigger_proposal refuses one outright – “DEBATE options are terminal at the tally” – rather than opening a window there is nothing to veto. Two of the things a Caper ballot can elect therefore move no value at all: the appended Do nothing, and a position the members wanted on the record. Only the remaining five typed kinds reach phase two.

The tradeoff is the honest one from the sections above. Borda is compromise-seeking and manipulable by burying, and Caper’s ballots are public while voting is open – so the strategic reading a whale can compute applies here as anywhere. What the design buys in exchange is that a winning executable option is the transaction, with no committee between the two gates and the treasury; see voting mechanisms.

References

• Stanford Encyclopedia of Philosophy, Voting Methods – Borda, Condorcet, Copeland, monotonicity failure, the no-show paradox, approval.
• Stanford Encyclopedia of Philosophy, Arrow's Theorem.
• Stefano Balietti, Pietro Saggese & Markus Strohmaier, Voting Biases in Decentralized Autonomous Organization (DAO) Governance (arXiv:2607.09435, 10 July 2026).
• Svante Janson, Phragmén's and Thiele's election methods (arXiv:1611.08826).
• Snapshot, Voting types.
• ENS DAO, EP 6.10 – Service Provider Program Season 2 selection and the SPP2 stream implementation thread.
• Polkadot, Sequential Phragmén method.
• Cathy Barrera, What is Voting For? (Prysm Group, April 2018).