# When One Leak Pays Forever: Context Binding and the Price of Deterring Collusion

Crypto.world

> In a threshold-encrypted mempool, users encrypt transactions to a committee key and the batch is decrypted only after ordering is fixed, which removes the information front-running exploits.

**arXiv**: 2609.36667

**Published**: 2026-09-29

**Authors**: Tingyi Lin, Shawn Yu, Ruoran Lai, Huanxi Zhang

## AI summary

## Overview

**Research area:** Algorithmic game theory applied to blockchain and cryptographic protocol design — specifically the economics of threshold-encrypted mempools, MEV (maximal extractable value), staking/slashing, and collusion deterrence.

**Technical level:** Intermediate. The core insight is intuitive (a secret that keeps working is worth more than a secret that expires), but the paper formalizes it with monotone access structures, discounted repeated games, supremum thresholds, and a covering linear program.

**Scope (one sentence):** The paper derives the exact penalty — such as slashable stake — needed to stop an authorized quorum from selling its decryption capability, and shows that the required penalty is set by how many future rounds a single leak can reach, not by the value of one round.

## What This Paper Is About

In a threshold-encrypted mempool, users encrypt transactions to a committee key and the batch is decrypted only after ordering is fixed, which removes the information front-running exploits. But any authorized quorum of the committee can reconstruct the decryption capability early and sell it to a front-runner. If that key lives for many epochs, one sale unlocks every later batch encrypted under it, so the colluding quorum is paid many times for one act of betrayal.

The paper asks a precise design question: how large must the one-time penalty (slashed stake, tracing liability, forfeited rewards) be so that no authorized coalition ever wants to leak? The answer is tied to a single new object — the set of future rounds that one leak still decrypts.

## Key Contributions

1. **An exact deterrence characterization (Theorem 4.1).** A coalition is deterred if and only if its exposure is at least the largest discounted value that a single leak, timed at any round, can reach. Because a leak is terminal and public, every dynamic deviation collapses to a choice of *when* to leak, so the repeated game enters only through the reachability set. Honest execution is coalition-safe exactly when the condition holds for every authorized coalition in a monotone access structure.
2. **A price for reuse (Theorem 4.5, Proposition 4.6, Corollary 4.7).** On a T-round instance with no discounting and unit stage values, full reuse requires exposure T while context binding requires exposure 1 — a separation of a factor of T, and the paper shows no instance separates the two designs by more than a factor of T. With a reuse window of at most ρ rounds, the bound is v̄(S)·Σ_{j=0}^{ρ(S)−1} δ^j, which is at most ρ(S)·v̄(S) for every δ ≤ 1.
3. **A cheapest way to split the penalty across parties (Corollary 4.8, Corollary 4.9).** With additive per-party exposure, deterring every authorized coalition is a set of covering constraints, so the minimum-total-stake design is a covering linear program. For a k-out-of-n committee with equal stakes and value depending only on coalition size, the smallest deterring per-party amount is max_{k≤m≤n} λ_m/m — the binding coalition is the size where reachable value per member peaks, not necessarily the smallest.
4. **Extensions and instantiations.** Probabilistic exposure (p(S)·f(S) ≥ sup_t Λ_t(S)) and time-distributed exposure (K_t(S) ≥ Λ_t(S) for every t) replace the one-time charge; a per-compromised-round charge of at least sup_r v_r(S) restores deterrence under unbounded reuse at every δ < 1. The framework is mapped onto Ferveo, Shutter, sealed-bid auctions, multi-authority voting, federated learning under a collective key, and threshold signatures, and calibrated on Ethereum front-running data.

## Main Findings

- **One-shot reasoning is exact only under context binding.** The classical Becker rule — penalty ≥ gain from one act — is the special case where a leak reaches only its own round (Corollary 4.2, Corollary 4.4). Under full reuse it can be short by a factor of T (Theorem 4.5).
- **Full reuse diverges as patience grows.** With a key valid forever, constant per-round value v(S), and discount δ < 1, the exact threshold is κ(S) ≥ v(S)/(1−δ), which grows without bound as δ → 1 (Corollary 4.3).
- **No horizon-constant penalty deters unbounded reuse.** With unit values and no discounting, the exact threshold equals ρ(S), the largest number of rounds one leak reaches (Proposition 4.6). If ρ(S) grows without bound across a family of deployments, any fixed exposure fails.
- **Counting rounds is not enough in general.** Counting reachable rounds is exact only with unit values and δ = 1; otherwise the identity and timing of the reached rounds matter (Section 4, Corollary 4.7).
- **Key rotation caps the multiplier.** Rotating keys every ρ(S) rounds keeps every reachability set within ρ(S) rounds when no leak can include a later key, capping the multiplier at ρ(S) for every δ ≤ 1.
- **The cheapest deterring stake profile is a covering LP.** Per-party amounts need not be equal, and Theorem 4.1 needs neither additive nor symmetric exposure, so it covers heterogeneous slashing and weighted or hierarchical access policies (Corollary 4.9).
- **Liability that grows with use fixes the divergence.** A charge c ≥ sup_r v_r(S) per compromised round makes K_t(S) ≥ Λ_t(S) at every δ < 1; a fixed one-time loss does not remove the divergence (Section 5).
- **Ethereum calibration.** Using 196,691 insertion (sandwich) attacks from Torres et al., with per-attack profit of $24.07 at the median, $62.92 at the 75th percentile, and $20,084.01 at the maximum, and a key reused over 32 block contexts with δ = 1: required exposure ρv at ρ = 1 / ρ = 4 / ρ = 32 is $24.07 / $96.28 / $770.24 at the median, $62.92 / $251.68 / $2,013.44 at the 75th percentile, and $20,084.01 / $80,336.04 / $642,688.32 at the maximum.
- **Worked worksheet (Example 6.2).** A $50,000 cap on one epoch's leakage value with δ = 0.95 requires $1,000,000 under a permanent decoder, $50,000 under context isolation, and roughly $185,494 for a four-epoch reuse cap; tracing succeeding with probability at least 0.4 against a $500,000 fine gives expected exposure $200,000.
- **Key custody determines reachability, not committee overlap.** Heavy membership overlap without key reuse leaves a leak bound to its own round, while a key valid across epochs produces multi-round reuse even with little overlap. In Shutter, a coalition that produces an epoch key also holds a threshold of eon-key shares, so its exposure must cover the larger leak up to the next eon key generation; Ferveo's per-epoch distributed key generation gives a single-round reachability set.
- **Never-released data invert the inequality.** For ballots that are never published, a leak also opens past elections, which reachability sets do not count, so the Theorem 4.1 threshold becomes a lower bound on the exposure needed (Section 6.3).

## Methodology in Plain English

The authors build a repeated game with three ingredients. First, an authorized-coalition family (any k of n committee members, or a weighted policy). Second, a per-round value that a coalition could earn by selling early access to that round's hidden data. Third, a reachability set that records which rounds a buyer can still read after a leak at a given round — everything later for a long-lived key, only that round for a context-bound key.

They then exploit a structural simplification: a successful leak is public and ends the leaking parties' participation, so a coalition leaks at most once and its entire deviation payoff is what that single leak reaches, minus the one-time penalty. Every strategy is therefore just a choice of leak time, and deterrence becomes a comparison between the penalty and a maximum over leak times of a discounted sum. Proving the theorem reduces to showing that a coalition leaks exactly when some leak time yields more than the penalty, and that otherwise no deviation is profitable.

From that comparison they derive closed-form thresholds for special reachability patterns (all later rounds, one round, a bounded window of ρ rounds), prove matching lower bounds to show the bounds are tight, and then solve the allocation question: when exposure is additive across members, the cheapest deterring stake profile is the solution of a covering linear program, which simplifies to a max-over-sizes formula for equal-stake threshold committees. Finally, they replace the one-time penalty with probabilistic and per-use charges, and map deployed systems (Ferveo, Shutter), plus auctions, voting, federated learning, and threshold signatures, onto the model's inputs.

## Why This Matters

**Impact on research.** The paper supplies a state variable — the reachability set — that separates deterrence analysis from the details of the market that follows a leak, and it shows that classic one-shot deterrence and the one-shot deviation principle of repeated games are special cases rather than general tools. It converts an economic security question ("is the stake big enough?") into a checkable inequality, and shows context binding has a one-line game-theoretic meaning: R_S(t) = {t}.

**Real-world applications:**

- **Threshold-encrypted mempools** (Ferveo, Shutter, BlindPerm): sizing slashable stake and choosing between per-epoch keys, reused keys, and windowed rotation.
- **Sealed-bid auctions:** servers holding a threshold key can open every later auction under that key unless bids are encrypted afresh per auction.
- **Federated learning under a collective key** (POSEIDON, MedCo): a coalition able to decrypt can sell a training round's model or aggregated updates; per-round key refresh bounds the exposure.
- **Threshold signatures and bridges:** a leaked signing capability stays valid in every later epoch for a long-lived bridge key, even with domain separation, but only one epoch for independent per-epoch keys or revocable per-context modules.

**Industry relevance.** For protocol designers, the results give a rule of thumb that contradicts a common budgeting instinct: a committee that sizes its stake to one round of MEV under a long-lived key is short by a factor of T. The paper also prices concrete alternatives — per-epoch keys, revocable modules that are replaced once a leak becomes public, per-use liability, and reuse caps — and its Ethereum calibration puts the gap between context isolation ($24.07 median) and 32-context reuse ($770.24 median) in dollar terms.

## Future Directions

- **Time-indexed access structures.** The paper notes that keys outliving a committee require time-indexed access structures, which its current monotone family over a fixed party set does not express.
- **Beyond transferable utility.** The exactness of Theorem 4.1 relies on transferable utility; without side transfers a coalition has no scalar objective, and the paper's limitations section (truncated in the supplied text) begins to address this gap.
- **State-dependent values for signature-like leaks.** A forged action that changes balances, nonces, or the set of feasible later actions makes values state dependent, which the model explicitly does not cover.
- **Certified empirical bounds on v_t(S).** The authors treat observed attacker profit as a stress benchmark rather than a certified upper bound, and call for certificates that aggregate every profitable response to early access and add tail and estimation margins.
- **Randomized and degraded-leak settings.** Propositions B.2 and B.4 extend the comparison to random reachability sets and degraded decoders; the constant per-round tracing hazard case is worked out in Corollary B.1, leaving richer detection dynamics open.

## Target Audience

Protocol designers and mechanism designers working on encrypted mempools, MEV mitigation, and proof-of-stake parameterization; cryptography engineers choosing between long-lived, per-epoch, and revocable-module key custody; and game-theory or algorithmic-economics researchers interested in collusion, coalition-proofness, and repeated games where a single deviation creates a stream of payoffs. Readers need comfort with discounted sums and linear programs, but the central result can be read as a single inequality.

## Authors’ abstract

A coalition that deviates once can profit many times when what it sells keeps working. In a threshold-encrypted mempool, a leading defense against maximal extractable value (MEV), a quorum of the decryption committee that sells its decryption capability to a front-runner exposes every later block that the capability still decrypts. We ask how large a penalty, such as slashable stake, deters this kind of collusion. In our repeated game, a single leak by any coalition in a…

[The full abstract on arXiv](https://arxiv.org/abs/2609.36667)

[Read the original paper](https://arxiv.org/abs/2609.36667)

---

Canonical: https://crypto.world/paper/when-one-leak-pays-forever-context-binding-and-the-price-of-deterring-2609.36667
