Understanding Signed Voting Signature

• Signed voting signature is a signed measure that precisely describes binary majority vote aggregation using latent competence distributions and binomial variance scales.
• The framework shows that the complete odd-budget voting curve is captured by signed Hausdorff moments, uniquely encoding branch imbalances across variance scales.
• This approach enables analysis of nonmonotonic voting behaviors and aids in distinguishing estimation regimes in both statistical aggregation and cryptographic applications.

Searching arXiv for papers on "signed voting signature" and related voting-signature usages. Signed voting signature is a signed measure that gives an exact one-to-one description of binary majority-vote test-time aggregation under exchangeable repeated correctness. In the formulation of "When Can Voting Help, Hurt, or Change Course? Exact Structure of Binary Test-Time Aggregation," majority voting is governed not simply by a scalar competence level, but by a latent distribution of per-example correctness probabilities; the signed voting signature records, at each binomial variance scale, the excess latent mass above rather than below the majority threshold. The complete odd-budget voting curve and this signature are equivalent: the budget-to-budget increments are signed Hausdorff moments, and the full odd-budget curve recovers the signature uniquely (Liu, 7 May 2026).

1. Exchangeability, latent competence, and the voting curve

The construction begins with an infinite exchangeable stream of binary correctness bits $B_1,B_2,\dots$ for each random instance. By de Finetti’s theorem there is a latent random variable $Q\in[0,1]$, with law $\Pi(dq)$, such that conditional on $Q=q$ the $B$’s are iid Bernoulli$(q)$. In this representation, $\Pi$ is the population distribution of per-example success probabilities, described in the paper as a “random-effects” or “fixed-competence” law (Liu, 7 May 2026).

For odd voting budgets $M=2n+1$, the conditional probability that a majority of $M$ votes is correct is

$$P_n(q)=\sum_{j=n+1}^{2n+1}\binom{2n+1}{j}q^j(1-q)^{2n+1-j}.$$

Averaging over the latent law yields the population voting curve

$Q\in[0,1]$0

The baseline term $Q\in[0,1]$1 is the one-vote accuracy. This formulation separates the within-example binomial randomness from the across-example heterogeneity encoded by $Q\in[0,1]$2. A plausible implication is that voting behavior is intrinsically a property of the full latent law, not merely of the mean competence.

2. Definition of the signed voting signature

The signed voting signature is defined after reparameterizing competence by the binomial variance scale

$Q\in[0,1]$3

This coordinate identifies the two “branches” $Q\in[0,1]$4 and $Q\in[0,1]$5, which have the same variance scale. Orientation is then retained through the factor $Q\in[0,1]$6, which is positive for $Q\in[0,1]$7 and negative for $Q\in[0,1]$8. The signed voting signature is the signed measure $Q\in[0,1]$9 on $\Pi(dq)$0 defined by

$\Pi(dq)$1

for any Borel set $\Pi(dq)$2 (Liu, 7 May 2026).

If $\Pi(dq)$3 has density $\Pi(dq)$4, the paper gives the equivalent density-form expression

$\Pi(dq)$5

where

$\Pi(dq)$6

The signature therefore retains exactly the antisymmetric part of the latent law across the two branches. It does not store total mass at a variance scale; rather, it stores the branch imbalance. This is the sense in which it records excess latent mass above rather than below the majority threshold.

3. Moment representation and exact equivalence

The central algebraic identity is pointwise in $\Pi(dq)$7:

$\Pi(dq)$8

After averaging over $\Pi(dq)$9 and regrouping by $Q=q$0, this yields the signed Hausdorff-moment formula

$Q=q$1

Together with

$Q=q$2

the entire odd-budget curve $Q=q$3 becomes exactly the moment sequence of $Q=q$4 (Liu, 7 May 2026).

Because compactly supported signed measures are determined uniquely by their moments, the mapping

$Q=q$5

factors through $Q=q$6, and $Q=q$7 is one-to-one with the full odd-budget curve. This theorem sharpens the usual fixed-competence picture. The voting curve is not merely summarized by whether average competence lies above or below $Q=q$8; it is a complete moment transform of a signed measure on variance scales.

4. Nonmonotonicity, oscillation, and qualitative voting behavior

The paper’s motivating claim is that classical fixed-competence theory makes majority voting appear monotone: more votes help above the majority threshold and hurt below it. The exact-structure analysis shows that this picture is fundamentally incomplete. Even simple latent mixtures can generate sharply different voting curves, including nonmonotone behavior, and the paper gives an explicit construction with infinitely many trend changes (Liu, 7 May 2026).

Five three-atom mixtures are highlighted, all with $Q=q$9, but with qualitatively different odd-budget curves:

• “constant”: $B$0
• “fast drop”: mass near $B$1 plus $B$2 gives steep early loss
• “dip-then-surpass”: below-$B$3 mass and above-$B$4 mass at two $B$5-scales produce a curve that first dips below $B$6 and later rises above $B$7
• “rise-then-fall”
• “slow-rise”

In each case, the signature $B$8 has positive and negative atoms at distinct $B$9-values, and their competing $(q)$0 weights explain the sign changes of $(q)$1. The infinite-oscillation construction places alternating positive and negative atoms closer and closer to $(q)$2, so that $(q)$3 changes sign infinitely often.

The significance is structural rather than anecdotal. A voting curve can rise, fall, plateau, or change direction multiple times without contradicting exchangeability or de Finetti structure. This suggests that heterogeneous example difficulty, not merely noisy replication, is sufficient to produce behavior that a scalar-competence model would misclassify as anomalous.

5. Identifiability, branch symmetry, and estimation regimes

The signed voting signature determines the odd-budget curve exactly, but the converse statement for the full latent law fails. The paper emphasizes that the full latent law determines the curve, but the curve does not determine the law. The source of nonidentifiability is branch symmetry: $(q)$4 sees exactly the difference of mass on the two branches $(q)$5 and $(q)$6, while branch-symmetric additions leave the voting curve unchanged (Liu, 7 May 2026).

If $(q)$7 has density $(q)$8, the decomposition is

$(q)$9

The voting curve depends only on $\Pi$0; the symmetric part $\Pi$1 is a nuisance that cannot be recovered from votes alone. The paper describes this as branch-symmetric nonidentifiability.

This distinction leads to two estimation regimes. With per-example labels or calibrated $\Pi$2, one can estimate the full signature via the empirical signed pushforward

$\Pi$3

By contrast, with grouped labels at fixed repeat depth $\Pi$4, each example yields only $\Pi$5. From the count mixture one recovers raw moments $\Pi$6 only for $\Pi$7, and therefore only the finite prefix

$\Pi$8

Hence one identifies $\Pi$9 only up to the same truncation. Finite-depth NPMLE therefore fits $M=2n+1$0 only up to a finite signature-prefix; the remainder is pure extrapolation requiring either more repeats $M=2n+1$1 or extra smoothness assumptions.

6. Distinct cryptographic uses of signatures in voting

In a separate research lineage, “signature” in voting usually refers not to a signed measure over variance scales, but to a cryptographic mechanism for ballot authorization, anonymity, or verification. These usages are conceptually distinct from the signed voting signature of test-time aggregation.

Setting	Signature mechanism	Salient property
Privacy-preserving e-voting	RSA blind signature	Blindness, unforgeability, single-vote enforcement
Hybrid remote/postal voting	Blind signature on filled-in ballot encoding	Universal eligibility verifiability
Decentralized blockchain voting	One-time ring signature	Anonymity with linkability via key image
Absentee-ballot authentication	HMAC/PRF code	UF-CMA ballot-code verification
Designated-verifier voting	Lattice-based SDVS	Only designated verifier can check validity

In "Towards Better Privacy-preserving Electronic Voting System," the blind-signature-based voting scheme uses RSA blinding, blind signing, unblinding, anonymous submission to a blockchain, and verification through $M=2n+1$2; it is described as simple, stable, and scalable, but as requiring additional anonymous property in the communication with the blockchain (Yan et al., 2022). In the Belgian remote-voting setting, blind signatures are used to replace a single-use random token with a genuine RSA blind signature on the filled-in ballot encoding $M=2n+1$3, with the aim of curing the eligibility “single-use-token” weakness while preserving vote secrecy and enabling universal eligibility verifiability (Willemson, 2023).

In decentralized anonymous voting on Ethereum, a one-time ring signature produces $M=2n+1$4, where the key image $M=2n+1$5 enforces one-time use through linkability if it repeats. The system combines ring signatures with stealth-address encryption and a self-tallying workflow, while offloading verification to the tally phase to reduce gas cost (Wu, 2018). For absentee ballots, Bernhard’s protocol derives a per-voter secret with a PRF and computes a deterministic code $M=2n+1$6, treating ballot validation as a MAC-verification problem under UF-CMA security goals (Bernhard, 2018).

A more restrictive verifier model appears in LaSDVS, a post-quantum strong designated-verifier signature based on ideal lattices under Ring-SIS and Ring-LWE. In the proposed signed-voting instantiation, voters sign ballots, but only the election authority, as designated verifier, can validate them; the scheme is claimed to provide strong unforgeability under chosen-message attacks, non-transferability, non-delegatability, and signer anonymity, with signature size $M=2n+1$7 rather than $M=2n+1$8 (Poddar et al., 23 Apr 2025).

The cryptographic literature also illustrates failure modes. "Attacking the Diebold Signature Variant -- RSA Signatures with Unverified High-order Padding" analyzes a 1024-bit RSA scheme with public exponent $M=2n+1$9 in which the verifier checks only the low-order 160 bits of $M$0 against SHA-1$M$1. Because the upper 864 bits are ignored, the paper gives a mathematically simple forgery attack that enables signatures on arbitrary messages in a negligible amount of time (Gardner et al., 2024).

The coexistence of these usages makes disambiguation necessary. In aggregation theory, signed voting signature is an exact statistical object extracted from majority-vote behavior. In voting-system cryptography, signatures are authorization and verification primitives attached to ballots or voters. The shared terminology reflects a common concern with voting, but the mathematical objects, threat models, and inferential roles are entirely different.