# AMP: Arc Multi-Proposer Protocol with Bounded Inclusion Guarantees

Crypto.world

> In proposer-based BFT blockchains, one validator per block decides which transactions get included and in what order, giving that actor the power to censor and to extract MEV.

**arXiv**: 2605.23677

**Published**: 2026-05-22

**Authors**: Daniel Cason, Preston Vander Vos, João Sousa, Adi Seredinschi, Gordon Liao

## AI summary

## Overview

**Research area:** Distributed systems and blockchain consensus — specifically multi-proposer Byzantine Fault Tolerant (BFT) protocol design, transaction censorship resistance, and MEV (maximal extractable value) mitigation.

**Technical level:** Advanced. The paper assumes familiarity with BFT consensus (quorums, commit certificates, safety/liveness), partial synchrony and asynchrony models, Byzantine fault thresholds, and broadcast abstractions. It presents formal correctness claims and pseudo-code rather than tutorials.

**Scope (one sentence):** The paper specifies and proves correct AMP, a multi-proposer protocol that composes on top of existing vote-based BFT consensus algorithms to guarantee that any payload attested by more than 2f validators at one height is finalized in the next block.

## What This Paper Is About

In proposer-based BFT blockchains, one validator per block decides which transactions get included and in what order, giving that actor the power to censor and to extract MEV. The authors argue that rotating leaders and mempool-based designs do not fix this: censorship stays unprovable, every validator gets early visibility into pending transactions, and the single assembler's egress bandwidth caps throughput. AMP's goal is to constrain that power without replacing the consensus layer — spreading payload dissemination across many proposers and using validator vote extensions to force the next block to include any payload backed by more than 2f validators.

## Key Contributions

1. **A composable multi-proposer protocol.** AMP augments, rather than replaces, any BFT consensus algorithm in which explicit validator votes determine decisions and which supports vote extensions — application-defined data attached to votes. The paper states it supports agreement layers at the n > 3f, n > 5f, and n ≥ 5f−1 fault thresholds, under partial synchrony, synchrony, or asynchrony.
2. **The bounded inclusion guarantee.** Any payload attested by more than 2f validators at height h must appear in the block finalized at height h+1. Combined with a deterministic ordering function over finalized payloads, this bounds any single validator's discretion over both inclusion and ordering.
3. **Decoupling of dissemination from agreement.** Proposers collect transactions, bundle them into payloads, and broadcast them to all validators; consensus runs over concise payload identifiers, not full payloads. This removes the need for a shared mempool and eliminates the single-assembler egress bottleneck. AMP needs only two pre-consensus communication steps (user to proposer, proposer to all validators), versus four in Autobahn.
4. **A correctness argument.** The paper states it proves AMP preserves the underlying protocol's safety and liveness properties and that bounded inclusion holds within one consensus height. (The proof section itself, Section 5, is beyond the truncated content supplied here.)

## Main Findings

- **Bounded inclusion rests on quorum arithmetic.** A commit certificate carries at least n−f signed deciding votes, so every certificate contains extensions from at least n−2f correct validators. A validator may omit at most f attestations, so a payload attested by more than 2f validators at height h retains more than f attestations in every valid consensus value for height h+1 — and is therefore necessarily finalized there.
- **Consensus values are commit certificates.** In AMP the value decided at each height is the commit certificate of the previous height, validated by a `ValidCommit(h−1, v)` predicate. At the genesis height h₀, a distinguished empty certificate is accepted.
- **Extraction threshold is "more than f".** On a decision, the validator extracts payload identifiers attested by more than f validators (`ExtractIDs`), records them as ordered for that height, and removes them from the pending set. Since each such payload was attested by more than f validators, at least one correct validator holds the full payload and can retransmit it if a validator is missing it.
- **No shared mempool is required.** Proposers push payloads to all validators, and validators attest only through vote extensions; the paper argues this avoids giving every validator early visibility into pending transactions, which in mempool-based designs expands the MEV attack surface to the entire validator set.
- **Comparison with Autobahn.** Autobahn makes every validator a proposer and uses four pre-consensus steps (send to validator, batch and broadcast, acknowledge storage, disseminate proof of availability); AMP achieves a comparable "seamlessness" property in two steps by running consensus over identifiers.
- **Comparison with Constellation (Solana).** Constellation inherits Alpenglow's model tolerating f < n/5 Byzantine validators plus p < n/5 crashes, with roughly 1000 validators, about 16 proposers and about 256 attesters; the block assembler waits for half of all attesters while only a quarter of fragments are needed for reconstruction. The authors argue that if all Byzantine validators land in attester roles, no blocks finalize until timeouts cycle correct validators in. In AMP, any quorum of n−f validators can enforce inclusion.
- **Comparison with MCP.** MCP (Multiple Concurrent Proposers) achieves selective-censorship resistance and hiding, but guaranteeing liveness requires reducing fault tolerance to f < n/5; at the standard f < n/3 threshold its guarantees are probabilistic. AMP claims per-payload inclusion in the next block at standard fault tolerance.
- **Comparison with leaderless protocols.** HoneyBadgerBFT and DispersedLedger do not prevent censorship across rounds, and Prefix Consensus permits up to f slots to be censored even after Global Stabilization Time (GST). Cadence guarantees payload inclusion only after a recovery period following GST; AMP claims its guarantee holds regardless of network conditions.
- **Comparison with Ethereum's approaches.** Proposer–builder separation (PBS) redistributes control to builders and relays without constraining block contents; FOCIL gives stronger inclusion guarantees but depends on a separate inclusion-list committee's formation, dissemination, and availability. AMP introduces no committee — the consensus quorum itself determines what every valid decision must include.
- **Ordering policy is application-defined but must be deterministic.** The paper recommends sorting transactions across all finalized payloads by descending priority fee per unit of computation, with pseudo-random tie-breaking, and requires `Sort` to deduplicate transactions, keeping only the highest-ranked copy and pseudo-randomly selecting among ties.
- **Execution trade-off.** Because a validator entering height h+1 proposes the certificate for height h, it may reference payloads it has not yet received, producing "asynchronous execution." The paper notes that Bitcoin, Ethereum, and Solana do not use asynchronous execution and instead pre-execute blocks before ordering, and describes an option to remove it in AMP by waiting for all payloads attested by more than f validators before proposing.
- **No quantitative evaluation is reported in the supplied content.** There are no throughput, latency, or benchmark measurements in the available text; the validation offered is the formal correctness argument.

## Methodology in Plain English

The authors take an existing, widely deployed building block — BFT consensus where validators sign votes to decide a value — and add a lightweight layer on top rather than designing a new consensus protocol from scratch. Their lever is the "vote extension," a piece of application data a validator attaches to a vote before signing it, already used in CometBFT-based chains.

The design works like this. Anyone can act as a proposer: they collect user transactions, bundle them into a payload, and broadcast it to every validator. Validators keep a list of valid payloads they have received but that have not yet been ordered, and when they cast a vote that could lead to a decision they attach the identifiers of those pending payloads as a vote extension. The value that consensus decides on is the previous height's certificate of at least n−f signed votes. Since any decided certificate therefore carries attestations from a large majority, a payload that more than 2f validators vouched for cannot be squeezed out by the at most f faulty validators plus the at most f omissions a validator is allowed — so it lands in the next block.

To show correctness, the authors fix the system model (authenticated signatures, up to f Byzantine validators out of n, a weak best-effort broadcast that guarantees delivery only from correct senders), state the interface AMP expects from the consensus layer (`Propose`, `Decide` with a transferable commit certificate, `Validate`) and from the application (`ExtendVote`, `VerifyExtension`, `ValidPayload`, `Finalize`), give pseudo-code for a validator, and argue the bounded-inclusion property and the inheritance of safety and liveness from the underlying protocol. Notably, the design explicitly does not rely on broadcast guarantees for its correctness — only on the consensus layer and the protocol construction.

## Why This Matters

**Research impact.** AMP adds a data point to the multi-proposer literature that is deliberately minimal: rather than a new consensus protocol (HoneyBadgerBFT, DispersedLedger, Prefix Consensus, Autobahn, Cadence, MCP, Constellation), it is a gadget over existing vote-based BFT. If the claims hold, it gives protocol designers a way to obtain a per-payload, one-height inclusion guarantee at standard fault tolerance — something the paper argues the alternatives do not provide.

**Real-world applications:**

- **Institutional settlement chains.** The protocol originated as an internal Circle project for the Arc blockchain, aimed at blockchains settling institutional-grade financial flows where unconstrained inclusion and ordering power is treated as a market-integrity concern.
- **MEV mitigation in decentralized exchange and trading traffic.** Deterministic cross-payload ordering by priority fee removes a single validator's discretion to reorder or insert transactions.
- **Censorship-resistant transaction submission.** A user can send a latency-sensitive transaction to multiple proposers, and once more than 2f validators attest it, no assembler can defer it past the next block.
- **Consortium or permissioned BFT deployments** already running vote-extension-capable consensus, which can adopt AMP without swapping their consensus engine.

**Industry relevance.** The paper grounds its motivation in a Bank for International Settlements observation that block assemblers "can engage in activities that would be illegal in traditional markets such as front-running and sandwich trades," and notes that regulators have flagged the absence of such rules as a first-order concern. It is written by authors at Circle, the issuer of a major stablecoin, and targets chains intended for regulated financial settlement — a deployment context where auditability of inclusion rules matters commercially as well as technically.

## Future Directions

- **Handling faulty proposers.** The protocol description assumes all proposers are honest and defers the implications of Byzantine proposers to Section 4.4, which is not included in the supplied content; how the guarantee degrades and what redundancy users need remains a central open question.
- **Synchronous versus asynchronous execution.** The paper sketches an option to remove asynchronous execution by waiting for all payloads attested by more than f validators before proposing the next certificate, but the latency and liveness consequences of that trade-off are not quantified in the available text.
- **Spam and fee economics.** The authors state that discouraging spam — for instance by charging users for each proposer they submit to — is out of scope of this work, leaving the incentive design for multi-proposer submission open.
- **Empirical evaluation.** No throughput or latency measurements appear in the supplied content. Since one of the paper's stated motivations is that single-assembler egress bandwidth caps throughput and that larger blocks take longer to propagate, an experimental comparison against single-proposer and DAG-based baselines is a natural next step. The paper's Section 6 is described as concluding with directions for future work, but that section is beyond the truncated content provided.

## Target Audience

Consensus protocol researchers and graduate students working on BFT and multi-proposer designs; blockchain engineers at layer-1 and layer-2 projects evaluating censorship resistance and MEV mitigation; and technical staff at financial institutions, stablecoin issuers, or market-infrastructure firms assessing whether a settlement chain's inclusion and ordering rules are defensible. Readers without a background in quorum-based consensus will find the paper's guarantees difficult to evaluate, since the argument is stated in terms of n, f, commit certificates, and vote extensions throughout.

## Authors’ abstract

Blockchain systems that settle financial transactions face a structural tension: the validator that assembles each block holds unilateral power over transaction inclusion and ordering. Traditional markets curb this power through laws that prevent front-running and market manipulation. Regulators have flagged the absence of such rules as a first-order concern for blockchain-based financial infrastructure. To address this tension, we introduce AMP, Arc multi-proposer protocol…

[The full abstract on arXiv](https://arxiv.org/abs/2605.23677)

[Read the original paper](https://arxiv.org/abs/2605.23677)

---

Canonical: https://crypto.world/paper/amp-arc-multi-proposer-protocol-with-bounded-inclusion-guarantees-2605.23677
