Lottery-Based Control Mechanisms

• Lottery-based control mechanism is a design approach that uses engineered randomization as the primary variable to determine allocations, participation, and state transitions.
• It finds applications in school choice, blockchain DApps, ride-sourcing, and network allocation, employing convex combinations and smooth mappings to manage stability and fairness.
• Its implementation bridges ex-ante welfare improvement with ex-post feasibility through advanced optimization techniques, secure randomness protocols, and distributed control systems.

Searching arXiv for the cited works and closely related lottery-mechanism literature. Lottery-based control mechanism denotes a class of designs in which randomization is not merely a tie-breaker or implementation detail, but the primary control variable used to govern allocations, participation, queue positions, payments, or state transitions. In the literature, this role appears in school choice through distributions over weakly stable matchings, in blockchain systems through smart-contract-managed lotteries, in network allocation through randomized throughput prospects evaluated under cumulative prospect theory, and in selection systems through smooth mappings from scores to marginal acceptance probabilities (Aziz et al., 11 Feb 2026, Long et al., 17 Aug 2025, Phade et al., 2018).

1. Conceptual scope and recurring design pattern

Across the surveyed work, a lottery-based control mechanism specifies a distribution over feasible outcomes and uses that distribution to shape ex-ante incentives or welfare. The controlled object varies by domain. In school choice, the object is a convex combination of weakly stable matchings. In admissions and funding, it is a vector of marginal selection probabilities summing to a quota. In ride-sourcing, it is a distribution over queue positions. In crowdsourcing and wagering, it is a randomized reward rule. In blockchain DApps, it is a smart-contract state machine that governs entry, drawing, and payout (Goldberg et al., 19 May 2026, Zhao et al., 2018, Liu et al., 29 Sep 2025).

Domain	Mechanism	Controlled object
School choice	“smart lottery” / PIRMES	Distribution over weakly stable matchings
Competitive selection	Clipped Linear Lottery	Marginal selection probabilities $p_i$
Ride-sourcing terminals	lottery-based queueing	Entry position in a virtual queue
Crowdsourcing	generalized lottrees	Lottery values $L(u)$ and rewards $R(u)$
Ethereum DApps	smart-contract lottery	Issuance, betting, drawing, prize distribution

A common structural feature is the separation between ex-ante and ex-post properties. A mechanism may improve ex-ante welfare through stochastic dominance, expected utility, or cumulative prospect value, while constraining ex-post realizations to remain stable, budget-balanced, or feasible. This suggests that lottery-based control is especially useful when deterministic mechanisms are blocked by impossibility results, hard thresholds, or severe fairness-efficiency tradeoffs.

2. Formal representations of lottery control

The formal language differs across subfields, but the underlying mathematical pattern is a controlled probability distribution over feasible outcomes. In school choice, a random matching is written as

$$p = \sum_{\ell=1}^k \lambda_\ell M^\ell,\quad \lambda_\ell \ge 0,\ \sum_\ell \lambda_\ell = 1,$$

and ex-post stability requires that each $M^\ell$ be weakly stable. In randomized selection, the mechanism is a map

$D: [0,1]^{n \times r} \to \Delta(\mathcal{K}),$

with marginals $p_i(X)=\Pr_{S\sim D(X)}[i\in S]$ and $\sum_i p_i(X)=k$. In resource allocation, each user receives a lottery over throughputs, and the system chooses among $k$ feasible allocations uniformly at random (Aziz et al., 11 Feb 2026, Goldberg et al., 19 May 2026, Phade et al., 2018).

In distributed settings, the mechanism is often modeled as a stochastic process with an explicit stopping rule. “Jointly Controlled Lotteries with Biased Coins” defines a mechanism as $M=(T,J,f)$, where $L(u)$0 is a stopping time, $L(u)$1 is a finite outcome set, and $L(u)$2 maps histories to outcomes. This representation makes robustness precise: the implemented distribution over $L(u)$3 can remain unchanged even if one of the two random generators is adversarially controlled (Solan et al., 2018).

In blockchain-based lottery DApps, the control representation is operational rather than convex-analytic. The contract stores a manager, a players array, and a winner, and implements the transition from open entry to draw to payout or refund through Solidity functions. The simple contract in the Ethereum gambling paper exposes a minimal control loop: $p_i(X)=\Pr_{S\sim D(X)}[i\in S]$1 followed by a draw routine and a refund routine restricted by onlyManager (Long et al., 17 Aug 2025).

A distinct formalization appears in pricing. “Pricing Randomized Allocations” models a lottery as $L(u)$4, where $L(u)$5 is a probability vector over items and $L(u)$6 is the price. The mechanism controls not only which outcome is realized but also how type space is partitioned by the menu of lotteries. In the buy-one model this control is strong enough to yield polynomial-time computation of the optimal envy-free lottery system for finite-support type distributions (0904.2400).

3. Matching and selection mechanisms

The most explicit use of the term in matching markets is the “smart lottery” for school choice. The baseline is student-proposing Deferred Acceptance after random tie-breaking of coarse school priorities, which induces a random matching

$L(u)$7

The control intervention is to optimize directly over the full set $L(u)$8 of weakly stable matchings and choose weights $L(u)$9 so that the resulting random matching is ex-post stable and stochastically dominates the baseline. The paper packages this as PIRMES, with the operational workflow: compute a baseline $R(u)$0, solve for an ex-post stable $R(u)$1 that sd-dominates $R(u)$2 and minimizes average rank, then decompose $R(u)$3 into a convex combination of weakly stable matchings and implement that lottery (Aziz et al., 11 Feb 2026).

This design changes what is being controlled. Standard DA with random tie-breaking is a lottery without control over correlations between schools’ tie-breakings. The smart lottery instead controls the correlation structure by selecting a distribution over stable matchings. In Example 1, the paper constructs

$R(u)$4

which eliminates third-choice probability for all four students, illustrating that optimizing the distribution over stable matchings can yield welfare gains unavailable to deterministic ex-post improvement cycles (Aziz et al., 11 Feb 2026).

A parallel development in competitive selection replaces discontinuous thresholding rules by smooth lottery maps. “Smooth Partial Lotteries for Stable Randomized Selection” formalizes smoothness as

$R(u)$5

and proposes the Clipped Linear Lottery

$R(u)$6

with $R(u)$7 chosen so that $R(u)$8. The mechanism always accepts above an upper threshold, always rejects below a lower threshold, and interpolates linearly in between. Its worst-case regret is bounded by

$R(u)$9

and for mean utility this matches the lower bound for any $$p = \sum_{\ell=1}^k \lambda_\ell M^\ell,\quad \lambda_\ell \ge 0,\ \sum_\ell \lambda_\ell = 1,$$0-smooth rule up to the factor $$p = \sum_{\ell=1}^k \lambda_\ell M^\ell,\quad \lambda_\ell \ge 0,\ \sum_\ell \lambda_\ell = 1,$$1 (Goldberg et al., 19 May 2026).

The contrast between these two literatures is instructive. The school-choice smart lottery preserves ex-post stability while improving ex-ante welfare through stochastic dominance. The Clipped Linear Lottery constrains the Lipschitz sensitivity of the score-to-probability map and optimizes the smoothness-regret tradeoff. In both cases the lottery is the control surface: one paper controls correlations across stable matchings, the other controls the geometry of marginal probabilities.

4. Distributed and automated lottery control

In blockchain applications, control is delegated to code. “From Fomo3D to Lottery DAPP” defines Ethereum gambling DApps as blockchain-based online lottery platforms in which smart contracts manage issuance, betting, drawing, and prize distribution. In the simple contract, participation is controlled by require(msg.value == 0.1 ether), winner selection is triggered through a manager-only draw, and the entire pot this.balance is transferred programmatically to winner. A refund branch returns exactly 0.1 ether to each recorded participant and resets the round (Long et al., 17 Aug 2025).

This architecture turns the smart contract into the control mechanism’s state machine. The paper explicitly identifies states equivalent to IDLE / NEW, [OPEN](https://www.emergentmind.com/topics/open), CLOSED_WIN, and CLOSED_REFUND, and emphasizes that decentralization, immutability, and transparency make the control logic auditable. Fomo3D illustrates a more elaborate variant in which a countdown timer, “Last Player Standing” rule, and airdrop mechanism shape user behavior by altering both jackpot timing and the probability of immediate rewards (Long et al., 17 Aug 2025).

The same section of the literature also exposes a central fragility of automated lottery control: poor randomness undermines the control rule itself. The Fomo3D airdrop seed is derived from publicly visible block variables and sender-dependent values, and the paper describes how attackers can reproduce the same environment to predict advantageous outcomes. Similarly, access-control modifiers such as isHuman based on extcodesize can be bypassed from a constructor, weakening the contract’s attempt to control participation (Long et al., 17 Aug 2025).

A more abstract solution to adversarial randomization appears in “Jointly Controlled Lotteries with Biased Coins.” There the mechanism observes outputs from two finite alphabets and implements a target distribution $$p = \sum_{\ell=1}^k \lambda_\ell M^\ell,\quad \lambda_\ell \ge 0,\ \sum_\ell \lambda_\ell = 1,$$2 over a finite set $$p = \sum_{\ell=1}^k \lambda_\ell M^\ell,\quad \lambda_\ell \ge 0,\ \sum_\ell \lambda_\ell = 1,$$3. The paper proves a finite-length strong secure $$p = \sum_{\ell=1}^k \lambda_\ell M^\ell,\quad \lambda_\ell \ge 0,\ \sum_\ell \lambda_\ell = 1,$$4-implementation and an exact weak secure implementation: under unilateral deviation by one device, the outcome distribution remains within $$p = \sum_{\ell=1}^k \lambda_\ell M^\ell,\quad \lambda_\ell \ge 0,\ \sum_\ell \lambda_\ell = 1,$$5 of $$p = \sum_{\ell=1}^k \lambda_\ell M^\ell,\quad \lambda_\ell \ge 0,\ \sum_\ell \lambda_\ell = 1,$$6 in the strong mode, or exact implementation is preserved up to non-termination and identification of the faulty device in the weak mode (Solan et al., 2018).

This distributed perspective suggests that lottery-based control can be centralized, as in a school district or platform, or decentralized, as in a smart contract or multi-device protocol. What remains invariant is that randomness is engineered, not merely sampled.

5. Incentives, budget consistency, and strategic robustness

A major reason to introduce lotteries is to reshape incentives subject to hard budget or strategy constraints. In the NBA draft context, the Carry-Over Lottery Allocation draft mechanism replaces single-season regular-season records with a multi-year lottery index $$p = \sum_{\ell=1}^k \lambda_\ell M^\ell,\quad \lambda_\ell \ge 0,\ \sum_\ell \lambda_\ell = 1,$$7. Every non-playoff team receives the same increment $$p = \sum_{\ell=1}^k \lambda_\ell M^\ell,\quad \lambda_\ell \ge 0,\ \sum_\ell \lambda_\ell = 1,$$8, with the baseline $$p = \sum_{\ell=1}^k \lambda_\ell M^\ell,\quad \lambda_\ell \ge 0,\ \sum_\ell \lambda_\ell = 1,$$9, tickets that do not win a top pick carry over, and playoff success or winning a top pick diminishes the index. Because extra losses after elimination do not change $M^\ell$0, the mechanism is designed to remove incentives to lose additional games once playoff qualification is impossible (Highley et al., 2 Feb 2026).

COLA also illustrates that lottery-based control can be stateful. The state variable is the carry-over ticket stock, not the current-season record. In exceptional draft years, the paper introduces a truth-elicitation layer using Non-Binary Robust Bayesian Truth Serum to decide whether the lottery boundary should be moved upward to include some playoff teams, thereby preserving the anti-tanking logic when teams might otherwise prefer lottery access to playoff participation (Highley et al., 2 Feb 2026).

In crowdsourcing, generalized lottrees make the budget constraint explicit. Participants form a solicitation tree rooted at the crowdsourcer, with contribution $M^\ell$1 at node $M^\ell$2. The mechanism defines lottery values $M^\ell$3 and rewards $M^\ell$4, and requires budget consistency: $M^\ell$5 The paper gives three variants—1-Pachira, K-Pachira, and Sharing-Pachira—and proves that the first-is-root rescaling satisfies Budget Consistency, Continuing Contribution Incentive, Continuing Solicitation Incentive, Value Proportional to Contribution, Unprofitable Solicitor Bypassing, and Unprofitable Sybil Attack (Zhao et al., 2018).

Behavioral mechanism design enters this space through cumulative prospect theory. “Web3 Meets Behavioral Economics” studies a crypto lottery in which each participant pays entry fee $M^\ell$6, the operator takes rake $M^\ell$7, and the prize pool is

$M^\ell$8

Under expected utility, participant expected profit is $M^\ell$9, so profitable lotteries would attract no rational entrant. Under CPT, however, the shape of the prize vector matters. Among the four mechanisms evaluated, top-$D: [0,1]^{n \times r} \to \Delta(\mathcal{K}),$0 linear weighting with $D: [0,1]^{n \times r} \to \Delta(\mathcal{K}),$1 yields the highest utility, becoming attractive at about $D: [0,1]^{n \times r} \to \Delta(\mathcal{K}),$2 for $D: [0,1]^{n \times r} \to \Delta(\mathcal{K}),$3 and $D: [0,1]^{n \times r} \to \Delta(\mathcal{K}),$4, whereas Winner-Take-All requires about $D: [0,1]^{n \times r} \to \Delta(\mathcal{K}),$5 and the three-bands rule about $D: [0,1]^{n \times r} \to \Delta(\mathcal{K}),$6 (Toyoda, 2022).

Randomization also resolves impossibility in wagering. “Randomized Wagering Mechanisms” defines lottery wagering mechanisms that convert deterministic net payoffs into a winner-takes-all lottery with identical expected payoffs. Applied to the Weighted Score Wagering Mechanism, the resulting LWS satisfies IR, SIC, EBB, sybilproofness, anonymity, neutrality, no arbitrage, and Pareto optimality. The paper also introduces surrogate wagering mechanisms, including RP-SWME, which use randomized surrogate outcomes and random partitioning to preserve incentive properties and robustness to noisy ground truth (Chen et al., 2018).

Across these cases, lottery-based control is not solely about welfare maximization. It is also a way to meet joint desiderata—budget consistency, anti-tanking, sybil-resistance, or Pareto optimality—that deterministic designs cannot simultaneously satisfy.

6. Optimization, empirical evidence, and computational limits

Lottery-based control mechanisms frequently shift difficulty from implementation to optimization. In school choice, the core problem of finding an ex-post stable random matching that sd-dominates a baseline and minimizes average rank is infinite-dimensional if each weakly stable matching is treated as a variable. The paper proves that checking constrained-sd-efficiency is coNP-complete, that computing the uniform average of all DA outcomes over all tie-breakings is NP-hard, and then solves the design problem with integer programming and column generation. The master problem chooses mixture weights over a current set of stable matchings; the pricing problem searches for a new weakly stable matching with positive reduced cost (Aziz et al., 11 Feb 2026).

The empirical results are correspondingly mixed: computationally demanding, but substantively large. In synthetic instances with 80 students and $D: [0,1]^{n \times r} \to \Delta(\mathcal{K}),$7, improving students gain on average $D: [0,1]^{n \times r} \to \Delta(\mathcal{K}),$8 rank positions under DA-PIRMES-CG versus $D: [0,1]^{n \times r} \to \Delta(\mathcal{K}),$9 under EE. In the 2015 Harku, Estonia kindergarten data under NOSIB-DIST3, EE improves 11% of students relative to DA by about $p_i(X)=\Pr_{S\sim D(X)}[i\in S]$0 rank on average, whereas DA-PIRMES-CG improves 69% of students and yields around a sixfold increase in both fraction of improving students and average improvement compared to EE in some priority structures (Aziz et al., 11 Feb 2026).

Ride-sourcing terminals exhibit a different optimization pattern. The lottery-based queueing paper models a driver’s entry decision through the expected waiting time induced by a distribution over queue positions, proves threshold equilibrium, and reformulates the infinite-dimensional, non-smooth design problem as a tractable bi-level program. Theoretical comparison shows that the lottery mechanism can achieve higher or equal social welfare than FIFO-queue-based dynamic pricing. Numerically, in unconstrained markets it only narrowly trails dynamic pricing in profit and significantly outperforms static pricing, while under commission fee caps it can surpass dynamic pricing in profitability (Liu et al., 29 Sep 2025).

Network resource allocation under CPT exposes another limit. For any fixed permutation profile, the inner allocation problem is convex and admits a market-based equilibrium with equilibrium prices. But the full system problem over permutations is NP-hard, and the primal and dual can exhibit a duality gap. The paper therefore studies a relaxation through doubly stochastic matrices and an average system problem, then derives a qualitative structure: under an inverse-S-shaped weighting function, the optimal lottery tends to flatten the lower tail so that a user receives the same minimum throughput in a large fraction of outcomes, with only a small top fraction carrying bonuses (Phade et al., 2018).

A broader methodological extension appears in worst-case VCG redistribution mechanism design. There, the “lottery ticket hypothesis” is used as a selection rule over neural subnetworks implementing Groves terms: a large MLP is repeatedly pruned into tiny subnetworks, each “ticket” is worst-case-trained using MIP-discovered violating type profiles, and repeated draws are guided away from previously bad tickets. This suggests that lottery-based control can also operate over mechanism search spaces, not only over final allocations (Guo, 2023).

Taken together, these results identify a persistent pattern. Lottery-based control mechanisms enlarge the feasible design space, often strictly improving ex-ante welfare, smoothness, or incentive compatibility. They also introduce nontrivial combinatorial and statistical burdens: column generation, bi-level optimization, MIP-based certification, and hard questions about robustness of randomness sources. The literature therefore treats lotteries not as a simple randomization layer, but as a high-dimensional design object whose distributional structure is itself the mechanism.