Majority Judgment is a voting method in which each voter grades every candidate on one shared scale of words, such as Excellent down to Poor, instead of ranking the candidates or picking one. Michel Balinski and Rida Laraki set it out in A theory of measuring, electing, and ranking, published in the Proceedings of the National Academy of Sciences in May 2007. Their argument is that a ballot asking which candidate a voter prefers collects less information than one asking how good each candidate is, and that counting grades avoids some of the failures that follow from counting preferences.
The Radix DAO, which is being set up to take over governance of the Radix network from the Radix Foundation, has made Majority Judgment the default for every election it holds. The first is for the seven seats of the Permanent Radix Accountability Council, which takes over from the three-member Transition council once the DAO's founding sequence completes. Radix changes one term of the published method: the share of voting power that settles a candidate's grade is three-fifths, not one half. A Radix election is therefore not decided on the median grade.
How the count works
Every ballot gives each candidate exactly one grade. To count a candidate, start at the highest grade and add up the votes that placed the candidate at that grade or better, moving down one grade at a time. The first grade at which the running total reaches half the votes is the candidate's majority grade: the median of the grades that candidate received. Candidates are ranked by majority grade, best first, and a tie between two candidates on the same grade is broken by looking at the votes either side of it.
The Radix variant
The Radix rules are in the DAO's Proposal & Voting Framework, section 6.2.4, and its DAO Parameters Registry, sections 3.5 and 6B. Both sit in the framework's pending/ folder until the DAO ratifies them.
| Term | Value | What it does |
|---|---|---|
| Grade scale | Excellent, Very Good, Good, Acceptable, Poor | Five grades in a fixed order. An implementation stores them as 4 down to 0; the numbers encode the order and are never averaged |
| Grade Quantile | Three-fifths | The share of the voting power cast that must place a candidate at or above a grade for that grade to be theirs. It is written into the framework text rather than the registry, so changing it takes an amendment |
| Minimum Qualifying Grade | Good | The lowest grade at which a candidate can be seated |
| Quorum | 7% of eligible voting power | Participation needed for a round to count |
A candidate's grade under these rules is called the qualifying grade: the first grade, counting down, at which three-fifths of the voting power cast has placed the candidate at that grade or higher. The framework sets the fraction at three-fifths to match the 60% approval threshold of the DAO's other election mechanism, a two-stage vote that shortlists candidates and then confirms each one, so a candidate faces the same share-of-turnout test under either. It follows that a block holding more than two-fifths of the voting power cast can hold a candidate's grade down. The quantile and the floor move together: raising the quantile lowers every qualifying grade computed from the same ballots, so a floor left where it was becomes a higher bar.
Each voter's full voting power counts behind every grade they give. A ballot must grade every candidate, and the voting app has to refuse an incomplete ballot at the moment it is cast, so that a voter's mistake cannot turn into lost turnout. There is no Abstain grade: a voter who does not support a candidate grades them low. Because every valid ballot grades every candidate, each candidate is measured against the same total, and nobody gains from appearing on fewer ballots than a rival.
An exact split resolves upward. Where the running total reaches exactly three-fifths at a grade, that grade is the qualifying grade, so a candidate graded Excellent by exactly three-fifths of the voting power cast and Poor by the rest qualifies at Excellent. Some of the literature takes the lower grade in that case. The framework takes the higher one because it falls out of a single pass down the scale and can be checked against the published tallies without a second rule.
Seats, ties and reruns
Candidates are ranked by qualifying grade, and the ranking is the seating order, so there is no separate confirmation round. Where a candidate ranked within the open seats falls below Good, the seat passes to the next candidate at Good or better; if those run out first, the remaining seats go to the DAO's vacancy process. Candidates at Good or better who miss a seat go on a reserve list. Because of the floor, an election with no more candidates than seats is still a real vote: any candidate graded below Good is not seated.
Where two candidates share a qualifying grade and the tie decides a seat, the framework breaks it with the majority gauge. For each tied candidate it compares p, the share of voting power that graded them above the shared grade, with q, the share that graded them below it. A candidate with p greater than q ranks ahead of one with the two equal, who ranks ahead of one with q greater than p; within the first group the larger p ranks first, and within the last the smaller q. The framework uses the gauge rather than the iterative formulation of the method because it gives the same order in one pass over the published tallies, so any member can check it. A tie the gauge cannot separate is decided by the Accountability Council, or by a runoff where more than two candidates tie for a remaining seat.
A round that closes below quorum can be rerun once, and the rerun is not made easier: the same quorum, quantile and floor, the same snapshot of voters, and twice the voting period. The framework gives two reasons. A rerun on lower thresholds asks a different question rather than giving more time to answer the same one; and a lower quorum would make the rerun cheaper to control than the round it replaces, which rewards suppressing turnout in the first round. If the rerun also misses quorum, nobody is elected and every seat goes to the vacancy process.
On the ledger
The DAO's governance component is deployed on Radix mainnet. It runs a blueprint named Governance, version 1.0.0, whose methods include vote_on_majority_judgment_election, and its state keeps a store and a counter for Majority Judgment elections alongside those for temperature checks and proposals. Read on 19 September 2026, all three counters stood at 0: the contract has run nothing, and none of the framework is in force yet.
The parameter set the contract holds for elections, keyed dao-election, stores the Good floor in a field named minimum_median_grade. The name describes the published method, not the Radix one; under the framework the floor applies to the qualifying grade at three-fifths.
No election can run until the DAO ratifies the framework by vote, and the council election follows ratification and the DAO's legal formation. The ratification vote was delayed by a halt of the Radix network from 31 August to 11 September 2026. The election card on radix.wiki tracks where both stand.
How Caper approaches this
A caper puts the same question to its members with a ranked ballot rather than a graded one, and settles it in two stages rather than one. The ballot is a full permutation: a proposal carries at most five options, and the deployed logic rejects a ballot that leaves one unranked or ranks one twice. A voter says which option they prefer to which, and cannot say how good any of them is, which is the information Balinski and Laraki's argument rests on. Each ballot costs a fee, 100 $XRD on the deployment running today.
The count is positional. A ballot gives the option it ranks first m − 1 times the voter's weight, the second m − 2, and so on down to nothing for the last, where m is the number of options – a Borda count, which is the family Majority Judgment was proposed against. What differs is the weight behind each ballot. It is neither one member one vote nor tokens alone, but w = (t × v) / (V × T): the tokens the voter holds at the moment they cast, times the soulbound vote tokens they have accumulated, over the whole vote supply and the tokens in circulation when the proposal opened. Casting mints the voter exactly one vote token and a ballot is the only thing that mints one, so the second factor is earned by turning up and cannot be bought on the curve.
There is no minimum qualifying grade, because there are no grades. In its place the Borda leader has to clear a supermajority of the weight actually cast – 1.5 divided by the number of options, so 75% on a two-option proposal and 30% on a five-option one – and it must not be an option marked as doing nothing. A caper sets no quorum at all: a proposal nobody votes on fails on the weight test, and there is no rerun.
The absolute reading Majority Judgment collects on the ballot, a caper collects after it. A proposal that wins the vote does not execute. It locks the caper's trailing time-weighted price as a baseline and opens a window, and it executes only if the price integrated across that window closes at or above the baseline. Holding is consent and a tie ratifies, so blocking means selling, and selling hard enough to hold the price down for a real share of the window rather than for an instant. Both the window and the trailing average are stored fields on the live logic component, set short on the test deployment. The two stages are set out on proposals and voting.