Vote: A Multifaceted Framework

• Vote is a structured mechanism for aggregating preferences, defined by axiomatic rules and applied in democratic elections and computational models.
• Strategic turnout models illustrate how cost-sensitive participation and multi-round voting can improve decision-making in diverse electoral settings.
• Digital voting systems employ cryptography and blockchain for verifiable, privacy-preserving counts, while vote signals enhance decentralized learning algorithms.

Searching arXiv for recent and foundational papers on voting procedures, e-voting, STV, and vote-aware optimization. Voting is a cornerstone of democracy, but contemporary research treats “vote” more broadly as a formal act of expressing and aggregating preferences over alternatives, candidates, projects, or generated outputs. In the literature considered here, voting appears as an axiomatized choice rule, a strategic participation game, a preferential multi-winner count, a privacy-preserving digital protocol, a participatory-budgeting mechanism, and a vote-aware signal for machine learning and decentralized optimization (Poplawski, 2018, Gersbach et al., 2017, Cho et al., 2024).

1. Formal structure and aggregation principles

At its most abstract, a voting rule is a family of functions $f_i : \prod_{v \in V} A \to A$ defined for electorates of size $|V|=i$, where $A$ is the set of alternatives and $L \in A$ is a distinguished tie outcome. In the axiomatic treatment of “self-consistency,” the support of alternative $a$ in profile $x$ is $P_a(x)=|\{v \in V \mid x_v=a\}|$, and axioms $C1$–$C5$ require determinism, symmetry with respect to voters and alternatives, invariance to adding an $L$-voter, and stability when a new voter supports the current social choice. Under these axioms, whenever the rule selects a non-tie alternative, that alternative must have strictly more supporters than any other single alternative; the paper therefore characterizes a family of majority vote rules through consistency rather than through May’s positive responsiveness (Poplawski, 2018).

A complementary formalization appears in “Propose or Vote,” where agents with single-peaked quadratic utilities $|V|=i$0 first choose whether to become proposers or voters, and then majority voting is applied to proposals. With appropriate default points and majority voting over two randomly selected proposals, the procedure can implement the Condorcet winner with only one round of voting if a Condorcet winner exists. In the one-dimensional single-peaked setting, the median ideal point is central: with odd $|V|=i$1, there exists an equilibrium in which only the median agent proposes and all others vote; with a modified procedure using an Artificial Agent and sequential elimination, the equilibrium becomes unique (Gersbach, 6 Jun 2025). This suggests that, in formal models, voting is not only an aggregation rule but also a mechanism for disciplining agenda formation.

2. Strategic turnout, costly participation, and repeated rounds

In costly-turnout models, voting is valuable only when the pivotality benefit exceeds the cost $|V|=i$2. “Assessment Voting” studies a binary choice between $|V|=i$3 and $|V|=i$4, with $|V|=i$5, $|V|=i$6, voting cost $|V|=i$7, and a Poisson population in which a deterministic Assessment Group of size $|V|=i$8 votes in round 1 while the remaining citizens form a Poisson($|V|=i$9) population in round 2. Assessment Group members receive a subsidy equal to $A$0, so round-1 voting is costless and sincere; the first-round vote difference is $A$1, where $A$2 records whether the voter prefers $A$3 or $A$4. After the first-round aggregate $A$5 is announced, second-round citizens decide whether to vote or abstain. The central theorem shows that for every $A$6 there exists $A$7 such that, for all $A$8, with probability at least $A$9: all first-round citizens vote for their preferred alternative, no second-round citizen votes, and alternative $L \in A$0 is chosen (Gersbach et al., 2017).

The mechanism rests on a threshold effect. For sufficiently large $L \in A$1, the only equilibrium in the second-round game is the no-show equilibrium $L \in A$2, because pivotality probabilities become too small relative to $L \in A$3. This yields a sharp contrast with one-round costly voting, where voluntary turnout produces bounded turnout and asymptotically equal winning probabilities for both alternatives, while compulsory voting implements the majority’s preferred alternative at the cost of universal participation. The welfare comparison in the paper states that there exist $L \in A$4 and $L \in A$5 such that, if $L \in A$6 and $L \in A$7, then $L \in A$8 (Gersbach et al., 2017).

A different two-round idea appears in “Repeat Voting,” which proposes two identical rounds in which every eligible voter may vote in each round, the first-round result is made public, and the final result is computed by summing votes across rounds, $L \in A$9. The argument is explicitly conceptual and proposal-oriented rather than theorem-driven: the first round acts as a “giant poll” whose votes actually count, potentially increasing participation by giving marginal voters a second chance after observing a real result, and reducing dependence on inaccurate pre-election polls. The paper argues that this can increase total participation and yield more representative outcomes, but it does not present explicit equilibrium conditions or mathematical proofs (Hart, 2022).

3. Preferential and multi-winner voting

A large part of the modern voting literature concerns preferential and multi-winner systems, especially Single Transferable Vote (STV). In the Australian context, STV is a preferential, multi-winner system in which each ballot is initially allocated to the highest-ranked candidate, a quota is calculated, candidates at or above quota are elected, surpluses are transferred, and the lowest candidate is excluded when necessary. The paper “To whom did my vote go?” describes a demonstration system for past Australian Federal Senate elections that lets a voter enter a ballot and trace how it is transferred across counts, including surplus transfers, exclusions, and fractional values. The example from the 2025 Victorian Senate election shows a ballot that starts with Steph HODGINS-MAY at weight $a$0, has $a$1 of its value used to elect her at count 259, transfers at value $a$2 to Fiona PATTEN, and after Fiona’s exclusion at count 284 transfers, still at $a$3, to Michelle ANANDA-RAJAH, helping her win the last seat. The same paper also highlights a counter-intuitive phenomenon from the 2016 Federal Senate election for Tasmania: a vote can contribute a negative amount to a candidate’s tally under particular surplus-transfer and rounding rules (Conway et al., 16 Sep 2025).

The vote package in R operationalizes several electoral systems—plurality, two-round runoff, approval, score, STV, and Condorcet winner/loser identification—with special emphasis on STV for small multi-winner elections such as committee and council elections and the selection of multiple job candidates. For STV, the package uses a Droop-style quota, supports complete round-by-round tables and visualizations, and implements two notable extensions: STV with equal preferences and STV with group constraints. The equal-preference variant splits a ballot’s weight among candidates tied at the highest rank among remaining candidates, and the authors state that the package implements STV with equal preferences “for the first time in a software package, to our knowledge.” The group-constraint variant enforces a minimum number of winners from a specified group, which changes the set of candidates who can be eliminated or elected at certain stages (Raftery et al., 2021).

These preferential systems clarify a recurring misconception: a vote in a ranked or transferable system is not a fixed scalar object. It may be partially used to elect one candidate, later transferred at reduced weight, or, under some rule variants, even appear as a negative contribution in an intermediate tally. This suggests that “one person, one vote” is operationally implemented through complex transfer rules rather than by a single immutable ballot contribution (Conway et al., 16 Sep 2025).

4. Verifiability, privacy, and digital voting infrastructures

End-to-end verifiable voting systems treat a vote not only as a preference expression but also as a cryptographic object that must be cast as intended, recorded as cast, and tallied as recorded. In vVote, an adaptation of Prêt à Voter for the 2014 Victorian State election, each ballot has a randomized candidate order and a QR code; the voter uses an Electronic Ballot Marker, receives a preference receipt signed by the Private Web Bulletin Board, and can later verify that the corresponding encrypted vote appears on the Public Web Bulletin Board. The system publishes a publicly verifiable list of decrypted votes and a proof that they correspond to the encrypted votes on the bulletin board, while using threshold ElGamal, re-encryption mixnets, randomized ballot generation, and multiple audit steps. At the same time, the paper is explicit that the vVote portion of the election is universally verifiable in a way the paper-ballot portion is not, and that complex STV tallying remains a special challenge (Culnane et al., 2014).

A more infrastructural line of work appears in vSPACE, a proof-of-concept centered on TrueElect[Anon] [Creds]. It combines Hyperledger AnonCreds SSI, a Kubernetes confidential cluster inside an Enterprise-Scale Landing Zone, continuous authentication, multi-party computation, and DLT-backed audit trails. The stated goal is to provide DLT-based, end-to-end auditable elections without leaking who voted for whom, using anonymous credentials to prove eligibility and uniqueness without exposing identity, and to publish combined certifiable ZKPs and audit artifacts into public DLTs. The protocol claims the classical properties of eligibility, uniqueness, privacy, universal anonymity, fairness, accuracy, individual and universal verifiability, robustness, self-tallying, and scalability, together with certifiable confidentiality, DIDs/SSI continuous authentication, consensus-AI hybridization, and integrity audit autonomy (Elnour et al., 2024).

Blockchain-oriented systems take different architectural positions. FASTEN uses an Election Commission for off-chain voter authentication and one random token per eligible voter, a set of wardens who submit encryption keys and later decryption keys, and an Ethereum smart contract that stores candidate lists, token hashes, encrypted votes, and tallies. The protocol explicitly targets voter anonymity, vote concealment during the voting period, vote immutability, and double voting inhibition; encrypted votes are grouped by “encryption-id,” and only after the voting interval closes do wardens submit the corresponding decryption keys (Damle et al., 2021). VoteMate, by contrast, is a dApp on EVM-based blockchain that implements the SBvote protocol with booth smart contracts, NIZK-based validation of 1-out-of-$a$4 encrypted ballots, a fault-recovery phase for non-responsive voters, and on-chain verification of per-group and global tallies. The paper presents this as a practical route to transparent, tamper-resistant, end-to-end verifiable voting on EVM-based infrastructure, while explicitly acknowledging that scalability remains a primary challenge when candidate participation is high (Homoliak et al., 21 May 2025).

Electryo brings Selene-style tracker verifiability to the polling-station setting. Each paper ballot contains a QR code with an encrypted identity and signature, the scanner posts encrypted tuples and proofs to a bulletin board, and after mixing and decryption the public tally is displayed as $a$5 pairs. The voter uses a receipt code and a tracker-retrieval mechanism based on $a$6 and $a$7 to recover the tracker corresponding to her vote. The paper is explicit that this design requires stronger assumptions on ballot privacy than normal paper voting, because the paper record contains an encrypted link to the voter’s identity, but it also argues that this yields good auditability, dispute resolution, and support for comparison risk-limiting audits (Roenne et al., 2021). The privacy–verifiability trade-off is therefore not incidental but structural.

5. Voting behavior, experimentation, and democratic design

Research on voting also studies how citizens search, interpret, and react to political information before they cast a ballot. A representative Swiss survey conducted in early January 2024, roughly two months before the federal popular votes on the “13th OASI pension payment” and the “Pensions initiative,” collected 6,427 search queries that respondents said they would use to find information on search engines. About 89.7% of queries were neutral, only 10.3% carried explicit sentiment, 15.6% asked about consequences, and 15.1% about interpretations. The paper reports no statistically significant relationship between voting intention and query sentiment, hence no direct evidence of selective exposure in query formulation for these retirement-policy votes; by contrast, undecided and non-voters were substantially more likely to search for nuanced information such as consequences and interpretations (Vziatysheva et al., 2024).

VoteLab addresses a different empirical problem: how to run controlled experiments that compare voting rules with real participants. Its architecture separates an Android voter interface, a PostgreSQL-based backend with a vote processor, and a .NET web dashboard for campaign design. The proof-of-concept study used four methods—majority voting, combined approval voting, score voting, and modified Borda count—on four COVID-19 questions, each with five options. The study operationalized “consistency” as agreement across methods on the ranked position of options and found that the protection question had the highest consistency, while the vaccine question had the lowest average consistency across methods (Kunz et al., 2023). This suggests that the choice of voting rule can be outcome-sensitive even when the underlying respondents and options are fixed.

Participatory budgeting work pushes this further from observation to institutional redesign. In Aarau’s “Stadtidee,” each citizen received 10 points to distribute across at least 3 projects under a fixed public budget of CHF 50,000. The paper contrasts a standard utilitarian greedy method with the Method of Equal Shares (MES), a proportional participatory-budgeting rule in which each voter is assigned a notional equal share of the budget and projects are funded when their supporters can jointly “pay” for them. The reported outcome is that alternative preferential methods such as cumulative voting and MES can produce more winning projects with the same budget and broader geographic and preference representation, particularly for voters who used to be under-represented, while causal analyses indicate that citizens prefer proportional voting methods and associate them with strong legitimacy even without very technical specialized explanations (Pournaras et al., 20 May 2025). A plausible implication is that empirical work on voting must be read jointly as behavioral measurement and as institutional engineering.

6. Vote beyond elections: preference data and decentralized learning

In adjacent computational literatures, “vote” is no longer a ballot in a civic election but a multi-annotator or peer-selection signal. Vote-based Preference Optimization begins from the observation that preference datasets used in RLHF and DPO are typically generated by collecting multiple votes or scores per pair of outputs, yet standard methods binarize these into hard labels. VPO introduces a Bayesian MMSE estimator for the latent preference probability $a$8 in a Beta–Binomial model, yielding the soft target

$a$9

This target then replaces hard labels in generalized preference objectives, producing vote-aware variants such as VDPO and VIPO. The paper reports that these variants outperform their base algorithms on SHP and UltraFeedback Binarized, and shows that vote-derived soft targets mitigate the reward-divergence behavior observed in standard DPO (Cho et al., 2024).

S-VOTE uses the term in a different computational sense: a local voting protocol for client selection in decentralized federated learning. Clients compute cosine similarity between their own model and neighbors’ models,

$x$0

vote for sufficiently similar neighbors, and perform local training if the number of received votes satisfies $x$1, with $x$2 in the reported experiments. An adaptive spontaneous-training rule increases the probability $x$3 of local training for underutilized clients according to $x$4. Under non-IID data, the paper reports lower communication costs by up to 21%, 4–6% faster convergence, local performance improvements of 9–17% in some configurations, and 14–24% energy-consumption reduction compared with baseline methods (Sánchez et al., 31 Jan 2025).

Across these computational settings, the vote becomes a calibrated, count-sensitive, or peer-generated signal rather than a one-shot civic act. This suggests a unifying abstraction: a vote is a structured unit of support whose informational content depends on the aggregation rule, the strategic environment, and the verification machinery attached to it.