Winner Fraction Heuristic

• Winner Fraction Heuristic is a unifying concept that quantifies how winning opportunities are apportioned in diverse systems such as blockchain, betting games, multi-winner voting, and graph-based processes.
• In blockchain applications, the heuristic ensures fairness by assigning a uniform selection probability among verifiably correct solvers without relying on stake or external weight.
• Across the Kelly criterion, multi-winner voting, and graph dynamics, the approach adapts to system-specific constraints—optimizing capital allocation, voter thresholds, and survival densities respectively.

“Winner Fraction Heuristic” is not introduced as a single, literal term across the relevant arXiv literature. Instead, the phrase denotes several distinct constructions in which a fraction associated with “winning” becomes the central analytic object: the per-block winning probability of a node in a deterministic hash-based lottery for a Proof-of-Useful-Work blockchain, the Kelly-optimal fraction of capital to allocate in a favorable game with variable pay-off, the $k$-dependent attainability threshold embedded in a multi-winner approval-voting heuristic, and the final density of surviving winners in a graph-based winner-takes-all process (Hoffmann et al., 2023, Pérez-Marco, 2014, Scheuerman et al., 2020, Krapivsky, 15 Jun 2026). The shared vocabulary conceals substantial differences in semantics: in one setting the fraction is a conditional probability, in another an optimal control variable, in another a cognitive scaling term, and in another an exact asymptotic density.

1. Terminological scope and conceptual unification

The literature does not present “Winner Fraction Heuristic” as a standardized technical label. In the blockchain paper, the winner-selection rule is described as an algorithmic, deterministic procedure rather than a heuristic, and the text explicitly notes that the paper does not introduce a named “Winner Fraction Heuristic” (Hoffmann et al., 2023). In the multi-winner approval-voting paper, the term is likewise absent, but the synthesis identifies a “Winner Fraction–style heuristic” in the modification of the Attainability–Utility with Threshold model, where the size of the winning set affects perceived attainability (Scheuerman et al., 2020). In the Kelly paper, the expression is used interpretively to denote a practical rule for how large a fraction of capital to allocate to a favorable but variable pay-off opportunity (Pérez-Marco, 2014). In the winner-takes-all graph process, by contrast, the winner fraction is an exact object of analysis: the fraction of vertices that remain active in the terminal state (Krapivsky, 15 Jun 2026).

This suggests that the phrase is best understood as a family resemblance across domains rather than a single theory. What remains common is the attempt to characterize how “winners” are distributed across a candidate set, an investment process, a ballot-selection rule, or a graph dynamics. The divergences are equally important: the relevant state variables are, respectively, the size of the correct-solvers set, the full pay-off distribution $p(b)$, the ratio encoded through $k$ and $m$, and the local topology of the graph.

2. Uniform winner fractions in deterministic blockchain winner selection

In "DFTWS for blockchain: Deterministic, Fair and Transparent Winner Selection" (Hoffmann et al., 2023), the winner-fraction idea is implied by a Proof-of-Useful-Work blockchain for High Energy Physics in which miners run Monte Carlo simulations rather than spam hashing operations. The Root Authority defines Monte Carlo block problems, including a seed derived from the previous block hash, a time interval for accepted solutions, and a commitment value $s$. Nodes that solve the problem upload solution data to RA servers and publicly broadcast a signed hash of the solution in a way that proves they solved the problem without revealing the solution itself.

After the submission interval closes, winner selection operates on a canonical list of correct solvers. The list contains entries $(\text{NodeID}, \text{Signature})$, ordered by NodeID alphabetically ascending. The RA first commits to randombytes via

$$s = \mathrm{Keccak256}(\text{PreviousBlockHash} \;||\; \text{randombytes}),$$

publishes $s$ with the block problem, and reveals randombytes only after the deadline. It then concatenates the Ed25519 signatures of the correct solvers,

$$c = \text{sig}_1 \;||\; \text{sig}_2 \;||\; \dots \;||\; \text{sig}_N,$$

computes

$$a = \mathrm{Keccak256}(c \;||\; \text{randombytes}),$$

takes the first 15 hex characters of $p(b)$0, converts them to an integer $p(b)$1, and defines

$p(b)$2

The winner is the node at position winnerindex in the alphabetically sorted list.

The paper’s fairness claim is that every node that solved the problem correctly and in time appears in the winner list exactly once, no list position is preferable, and the winner index is uniformly distributed modulo the list length, assuming Keccak-256 behaves like a random oracle (Hoffmann et al., 2023). Because NodeID ordering is fixed, Ed25519 signatures are deterministic, and the RA commits to randombytes before seeing the future signatures, each correct solver has probability $p(b)$3 of selection once the candidate set of size $p(b)$4 is fixed. The synthesis therefore identifies an effective winner-fraction heuristic: $p(b)$5 where $p(b)$6 is the candidate set for block $p(b)$7 and $p(b)$8. Across many blocks, if $p(b)$9 is the number of blocks won by node $k$0 and $k$1 is the set of blocks in which that node submitted a correct solution, then

$k$2

The significance of this construction is that no stake or external weight is used: the only “weight” is being in the candidate set. This distinguishes the scheme from traditional Proof-of-Work, where long-run block fraction is proportional to hash power, and from Proof-of-Stake, where it is proportional to stake. In the DFTWS design, long-run win fraction is approximately proportional to the fraction of blocks in which a node enters the correct-solvers set, while selection within that set is uniform, deterministic, fair, and transparent (Hoffmann et al., 2023).

3. Kelly winner fractions under variable pay-off

In "Kelly criterion for variable pay-off" (Pérez-Marco, 2014), the relevant winner fraction is the fraction $k$3 of current wealth allocated repeatedly to a favorable game. In the classical constant-pay-off setting, one bets fraction $k$4 of wealth each round; with probability $k$5 the round is won at pay-off $k$6-to-1, and with probability $k$7 the stake is lost. Expected log-growth is

$k$8

and the classical Kelly fraction is

$k$9

The paper generalizes to variable pay-off: win probability remains $m$0, but conditional on winning, the pay-off $m$1 is a non-negative random variable with distribution $m$2 on $m$3. Its mean is

$m$4

A round is favorable if

$m$5

Expected log-growth becomes

$m$6

and the maximizing fraction is the unique $m$7 satisfying the fundamental integral equation

$m$8

Within the “Winner Fraction Heuristic” interpretation, the central correction is that the true optimal fraction depends on the entire pay-off distribution, not only on its mean (Pérez-Marco, 2014). If one naively substitutes $m$9 into the classical formula,

$s$0

one obtains an upper bound, not the correct variable-pay-off solution. Corollary 3.2 states that

$s$1

with equality only when the pay-off is actually constant. The proof uses Jensen’s inequality for the strictly concave function

$s$2

The importance of this result is methodological. A winner-fraction rule based only on average pay-off is systematically too aggressive when pay-offs are random. The paper’s practical prescription is therefore not to collapse the pay-off distribution into a single average and apply the classical Kelly formula, but to estimate $s$3 and $s$4 and solve the integral equation numerically (Pérez-Marco, 2014). In this setting, the “heuristic” is not equal-chance selection among winners, but risk-aware calibration of the share of capital committed to a favorable opportunity.

4. Winner-set scaling in multi-winner approval voting

In "Modeling Voters in Multi-Winner Approval Voting" (Scheuerman et al., 2020), the winner-fraction idea appears in a descriptive model of voter behavior under approval voting with multiple winners. The paper studies four behavioral models—Complete, Take the $s$5 best, Attainability–Utility (AU), and Attainability–Utility with Threshold (AUT)—and compares them with an Optimal Baseline based on full expected-utility maximization. The key innovation is that AUT takes into account both the size of the winning set and human cognitive constraints.

The formal source of the winner-fraction dependence is a modification of the attainability baseline. In the original single-winner plurality version, attainability is

$s$6

For multi-winner approval voting with $s$7 winners and $s$8 candidates, the reference share $s$9 is replaced by

$(\text{NodeID}, \text{Signature})$0

If $(\text{NodeID}, \text{Signature})$1 is the total number of approvals so far, $(\text{NodeID}, \text{Signature})$2 is the number of missing ballots, and

$(\text{NodeID}, \text{Signature})$3

then multi-winner attainability is defined as

$(\text{NodeID}, \text{Signature})$4

The AUT model then assigns each candidate a candidate-level AU score,

$(\text{NodeID}, \text{Signature})$5

and predicts the ballot

$(\text{NodeID}, \text{Signature})$6

In practice the paper focuses on $(\text{NodeID}, \text{Signature})$7, so the score reduces to

$(\text{NodeID}, \text{Signature})$8

The winner-fraction interpretation is explicit: as $(\text{NodeID}, \text{Signature})$9 increases, the baseline $$s = \mathrm{Keccak256}(\text{PreviousBlockHash} \;||\; \text{randombytes}),$$0 decreases, so more candidates become “attainable enough” to cross the threshold $$s = \mathrm{Keccak256}(\text{PreviousBlockHash} \;||\; \text{randombytes}),$$1 (Scheuerman et al., 2020). The model therefore predicts that voters approve more candidates when more seats are available, but does so without requiring exhaustive optimization over all $$s = \mathrm{Keccak256}(\text{PreviousBlockHash} \;||\; \text{randombytes}),$$2 approval subsets. This is the sense in which AUT incorporates both winner-set size and bounded rationality.

Empirically, the model is evaluated on behavioral data from 104 Mechanical Turk workers. All subjects face single-winner scenarios first, and are then randomized into a 2-winner condition with $$s = \mathrm{Keccak256}(\text{PreviousBlockHash} \;||\; \text{randombytes}),$$3 or a 3-winner condition with $$s = \mathrm{Keccak256}(\text{PreviousBlockHash} \;||\; \text{randombytes}),$$4. Using leave-one-out prediction over six observations per participant, AUT achieves mean prediction accuracy of $$s = \mathrm{Keccak256}(\text{PreviousBlockHash} \;||\; \text{randombytes}),$$5 in the 1-winner condition, $$s = \mathrm{Keccak256}(\text{PreviousBlockHash} \;||\; \text{randombytes}),$$6 in the 2-winner condition, and $$s = \mathrm{Keccak256}(\text{PreviousBlockHash} \;||\; \text{randombytes}),$$7 in the 3-winner condition, outperforming Optimal, AU, Complete, and Take $$s = \mathrm{Keccak256}(\text{PreviousBlockHash} \;||\; \text{randombytes}),$$8 best in each case (Scheuerman et al., 2020). The “Winner Fraction Heuristic” in this literature is therefore best understood as the AUT model’s $$s = \mathrm{Keccak256}(\text{PreviousBlockHash} \;||\; \text{randombytes}),$$9-sensitive attainability threshold.

5. Exact and heuristic winner fractions in winner-takes-all graph dynamics

In "The Winner Takes It All" (Krapivsky, 15 Jun 2026), the winner fraction is the final density of surviving active agents in a continuous-time stochastic elimination process on a graph. Initially, there is one active agent at each vertex. Each edge fires independently with rate 1; when an edge with two active endpoints is selected, one endpoint is chosen uniformly at random to lose and becomes permanently inactive. The process stops when no edge joins two active vertices. The remaining active agents are the winners.

On the infinite one-dimensional lattice $s$0, the paper derives the exact final winner fraction

$s$1

The dynamic derivation uses densities $s$2 of contiguous active segments of length $s$3, satisfying

$s$4

with solution

$s$5

From this one obtains the density of active segments surrounded by losers,

$s$6

and the active fraction

$s$7

The one-dimensional analysis is refined further by decomposing winners according to how many games they won. Since degree is 2, a winner can win at most two games, and the fractions $s$8 satisfy

$s$9

The paper also studies finite segments of length $$c = \text{sig}_1 \;||\; \text{sig}_2 \;||\; \dots \;||\; \text{sig}_N,$$0, deriving the mean number of losers $$c = \text{sig}_1 \;||\; \text{sig}_2 \;||\; \dots \;||\; \text{sig}_N,$$1, the asymptotic winner fraction

$$c = \text{sig}_1 \;||\; \text{sig}_2 \;||\; \dots \;||\; \text{sig}_N,$$2

the variance growth

$$c = \text{sig}_1 \;||\; \text{sig}_2 \;||\; \dots \;||\; \text{sig}_N,$$3

and the probabilities of reaching the final state with the minimum or maximum number of winners.

On infinite regular trees with degree $$c = \text{sig}_1 \;||\; \text{sig}_2 \;||\; \dots \;||\; \text{sig}_N,$$4, i.e. Bethe lattices with coordination number $$c = \text{sig}_1 \;||\; \text{sig}_2 \;||\; \dots \;||\; \text{sig}_N,$$5, branch independence yields an exact formula for the fraction of winners: $$c = \text{sig}_1 \;||\; \text{sig}_2 \;||\; \dots \;||\; \text{sig}_N,$$6 For large $$c = \text{sig}_1 \;||\; \text{sig}_2 \;||\; \dots \;||\; \text{sig}_N,$$7,

$$c = \text{sig}_1 \;||\; \text{sig}_2 \;||\; \dots \;||\; \text{sig}_N,$$8

The paper then formulates a winner-fraction heuristic for more general sparse or high-dimensional graphs: start from the Bethe-lattice approximation and use local degree and tree-likeness as the primary structural parameters (Krapivsky, 15 Jun 2026). This produces conjectural high-dimensional asymptotics

$$c = \text{sig}_1 \;||\; \text{sig}_2 \;||\; \dots \;||\; \text{sig}_N,$$9

Unlike the previous literatures, this setting treats the winner fraction as an exact observable first and a heuristic only second. The exact solutions on $$a = \mathrm{Keccak256}(c \;||\; \text{randombytes}),$$0 and on Bethe lattices function as anchor points from which tree-based approximations for more complex graphs are proposed.

6. Comparative interpretation, assumptions, and recurring misconceptions

The four uses of “winner fraction” are mathematically non-equivalent. In the blockchain setting, the relevant quantity is a conditional selection probability inside a correct-solvers set of size $$a = \mathrm{Keccak256}(c \;||\; \text{randombytes}),$$1 (Hoffmann et al., 2023). In the Kelly setting, it is the optimal capital fraction $$a = \mathrm{Keccak256}(c \;||\; \text{randombytes}),$$2 that maximizes expected log-growth under a variable pay-off distribution (Pérez-Marco, 2014). In multi-winner approval voting, it is the $$a = \mathrm{Keccak256}(c \;||\; \text{randombytes}),$$3-dependent scaling of attainability through the term $$a = \mathrm{Keccak256}(c \;||\; \text{randombytes}),$$4, coupled to a threshold rule (Scheuerman et al., 2020). In the winner-takes-all graph process, it is the density of terminal survivors and, for general graphs, a tree-based approximation to that density (Krapivsky, 15 Jun 2026).

A common misconception is to treat these objects as if they all encoded the same notion of fairness or optimality. The papers do not support that reading. The blockchain scheme defines fairness as exact symmetry among correct and timely solvers, with transparency and auditability enforced by public commitments, deterministic signatures, and recomputation of the winner index (Hoffmann et al., 2023). The Kelly framework defines optimality by asymptotic expected log-growth, not by equalization across agents or events (Pérez-Marco, 2014). AUT is a descriptive behavioral model of human approval decisions under uncertainty and cognitive constraints, not a normative social-choice rule (Scheuerman et al., 2020). The winner-takes-all graph process is a stochastic elimination model whose winner fraction depends on graph topology, not on strategic choice or resource allocation (Krapivsky, 15 Jun 2026).

The assumptions also differ sharply. The blockchain design assumes honest and transparent identity registration by the Root Authority, secure Keccak-256 and deterministic Ed25519 signatures, and sufficient network synchrony for broadcasting before deadlines (Hoffmann et al., 2023). The Kelly analysis assumes iid rounds, stationarity of $$a = \mathrm{Keccak256}(c \;||\; \text{randombytes}),$$5 and $$a = \mathrm{Keccak256}(c \;||\; \text{randombytes}),$$6, frictionless reinvestment, and perfect knowledge of the relevant probabilities (Pérez-Marco, 2014). The AUT model assumes that voters use threshold-based candidate-level evaluation rather than exhaustive expected-utility maximization, and its empirical performance is established by cross-validated prediction accuracy rather than closed-form rational-choice derivation (Scheuerman et al., 2020). The graph-theoretic analysis relies on exact solvability in one dimension and on tree factorization for Bethe lattices; its broader heuristic extensions depend on local tree-likeness and degree concentration (Krapivsky, 15 Jun 2026).

This suggests that “Winner Fraction Heuristic” functions best as an umbrella descriptor for a recurrent modeling move: identify the quantity that governs how winning opportunities are apportioned, and express it as a fraction whose structure captures the dominant constraints of the system. In one literature that fraction is uniform over a candidate set, in another it is reduced by pay-off variability, in another it is scaled by the size of the winner set, and in another it is induced by elimination dynamics on a graph.