Pith. sign in

REVIEW 3 major objections 4 minor 26 references

Mangrove: Fast and Parallelizable State Replication for Blockchains

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Mangrove proposes that every smart contract runs its own consensus instance, so conflicting transactions are excluded by per-actor locks instead of a global total order.

desk verdict Per-contract consensus with a genuinely new locking design, but the slow path leans on a Quorum Consensus primitive that is asserted via a sketchy reduction rather than actually instantiated. read the letter →

arxiv 2509.06616 v1 pith:AWQCQSOD submitted 2025-09-08 cs.DC

Anton Paramonov , Yann Vonlanthen , Quentin Kniep , Jakub Sliwinski , Roger Wattenhofer This is my paper · ORCID
classification cs.DC MSC 68M1468M1268M15
keywords blockchainparallelexecutionstatereplicationoptimisticagreementByzantinefaulttolerancesmartcontractsper-actorconsensus
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Mangrove aims to show that a blockchain can support general smart contracts without maintaining a strict total order over all transactions. The paper's central claim is that a separate consensus instance per smart contract, coordinated by lightweight local locks at each user actor, can prevent double-spending across instances while letting different contracts make progress independently. If correct, this would give blockchain systems the parallel scalability of consensus-less payment protocols without giving up the ordering that smart contracts need. Under optimistic conditions, a transaction can be committed in two communication steps, matching the best known latency for this class of protocols.

What carries the argument

The central object is Parallel Optimistic Agreement (POA), a per-recipient consensus instance whose fast path consists of a leader broadcast followed by a single round of votes carrying fallback blocks. Its safety rests on Wait-Free Locking, two maps at each user actor entity — FP-locked for fast-path voting and SP-locked for slow-path proposals — that tie a sequence number to a single transaction, ensuring that no conflicting transaction can be adopted by enough honest validators in any POA instance. The slow path additionally relies on Quorum Consensus, a consensus primitive satisfying Quorum Validity, and Transaction Agreement, a per-sequence-number strong Byzantine agreement used retroactively to resolve conflicts after fast-path failure.

What would settle it

Run a concrete implementation or simulation of the slow path using a real Quorum Consensus algorithm satisfying Property 12, with $n=3f+2p+1$; if two conflicting transactions from the same user actor and sequence number can be decided in POA instances for different reactive actors in any asynchronous-before-GST schedule, the No Conflict property fails. Alternatively, under synchronous, honest-leader, at-most-$p$-faulty conditions, measure whether an honest user's UA-RA transaction is executed within three communication steps; exceeding that would contradict Fast UA-RA Transaction Execution.

Watch

Extended reading notes

Core claim

The paper claims that using a Replicated Actor Model, where user actors and reactive actors (smart contracts) each have their own agreement stream, eliminates the need for a global transaction order while preserving safety. The load-bearing result is the Parallel Optimistic Agreement (POA) primitive: each reactive actor runs its own POA instances, and conflicting transactions from the same user actor and sequence number cannot be simultaneously decided in any two instances, even when those instances belong to different actors. This is achieved by wait-free locking: a validator's user actor entity maintains fast-path and slow-path lock maps that bind a sequence number to one transaction, so conflicting transactions cannot gather the required quorum of votes. Under an honest leader, synchrony, and at most $p$ misbehaving validators, with $n\geq 3f+2p+1$ total validators, the fast path terminates in two communication steps for user-actor and reactive-actor transactions, and two or three steps for user-to-contract transactions. The slow path, invoked only on failure, uses a Quorum Consensus primitive with Quorum Validity followed by per-sequence-number Transaction Agreements to preserve safety and liveness.

Load-bearing premise

The slow path's safety rests on the existence of a Quorum Consensus primitive whose Quorum Validity holds at the assumed thresholds, and the paper reduces this to an existing theorem without supplying a concrete algorithm; if that primitive cannot be instantiated, the no-conflict and termination guarantees of the slow path do not follow.

Editorial extensions

If this is right

  • Each smart contract maintains its own ordered chain, so congestion at one contract does not slow down unrelated contracts, and throughput can be scaled horizontally by assigning different actor entities to different machines.
  • Optimistic transactions execute in two communication steps for user-actor and reactive-actor transactions, and two or three steps for user-to-contract transactions, under the stated honesty and synchrony assumptions.
  • Conflicting transactions from the same user and sequence number can never both be committed, even when they target different reactive actors, without requiring a global order.
  • If a user misbehaves by issuing conflicting transactions, only the fast path's liveness is affected; honest transactions are still committed through the slow path, and the misbehavior is evidenced by two signed conflicting transactions.
  • The protocol matches the resilience bound $n\geq 3f+2p+1$ for fast Byzantine consensus with all users as proposers, which the paper argues is the best possible in that setting.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: if a practical Quorum Consensus with Quorum Validity can be instantiated, the per-actor lock pattern suggests a general recipe for replicated state machines: local single-writer locks at shared objects can replace global ordering for cross-object conflict prevention, beyond blockchains.
  • Editorial inference: a testable prediction absent from the paper is that, under heavy contention on one reactive actor, the throughput of all other actors remains flat; a benchmark comparing Mangrove-style per-actor ordering with sharding would directly test this.
  • Editorial inference: because UA-RA transactions pay an extra step when the leader is not the author, a leaderless variant of POA might reduce all transaction types to two steps at the cost of a larger validator set, echoing the $5f+1$ tradeoff already noted in the paper's discussion of Flutter.
  • Editorial inference: the model implies atomic composability, such as flash loans, must be reintroduced at the application layer via locking and execution contexts; whether this suffices for composability-critical decentralized finance under adversarial contention is a question the paper opens but does not answer.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. Mangrove proposes a replicated-actor blockchain execution model in which each smart contract (reactive actor) and each user actor is served by its own agreement instance, eliminating a global transaction order. For reactive actors the paper defines Parallel Optimistic Agreement (POA) with a two-step fast path and a slow path built from Quorum Consensus and Transaction Agreement; for user-actor-only transactions it uses Parallel Optimistic Broadcast. The paper claims an optimistic latency of 2-3 communication steps, a resilience bound of n >= 3f + 2p + 1, horizontal scalability per actor, and no cross-instance double spending. Appendices contain full pseudocode and proofs of Agreement, Validity, No Conflict, and system-level properties.

Significance. The contribution is conceptually interesting and, if the proofs go through, would be a meaningful advance: per-contract consensus without a global order, with a formal no-conflict cross-instance guarantee and fast-path latency comparable to consensus-less systems. The paper is honest about scope: it provides no implementation or benchmarks and explicitly acknowledges high message complexity. Its formal apparatus is detailed, and the high-level intersection arguments for Agreement and No Conflict are plausible. The main weakness is that the slow path is built on a Quorum Consensus black box whose existence is only argued by a terse reduction and an informal appendix sketch, and the central no-conflict claim depends on that primitive.

major comments (3)
  1. [§6.1, Claim 13] Quorum Consensus with Quorum Validity is the load-bearing primitive. The proof as written is a shorthand: it uses I^{n-f}, sim(c), and val(c') from [21] without defining how POA proposals correspond to [21]'s configurations, and it asserts that a block present in n-2f proposals belongs to val(c') for every c' in sim(c) without proving that an algorithm from [21] can be instantiated to decide exactly such blocks while preserving Agreement and Termination. Appendix B's direct-vote extension is only a sketch; it does not give pseudocode, invariants, or a proof that Narwhal/Tusk, Bullshark, Shoal, or Mysticeti can be modified to satisfy Quorum Validity (II). Since Lemma 18, the slow-path Agreement proof, Validity (II), and the No Conflict Property all invoke Quorum Validity (I)/(II), a gap here propagates to the paper's central claim. Please either replace this black box with a fully specified algorithm or give a complete reduction.
  2. [§5.2 and Properties 10/11] The common leader oracle is stated as an assumption ('we require this function to output an honest leader infinitely often'), and the suggested implementations are a common random source or round-robin. Neither is part of the stated GST model. The liveness properties (Termination, Multi-Termination, and the fast-termination results in Table 1) are conditioned on this oracle. This is acceptable as an explicit assumption, but the assumption should be stated in the formal system model and in the theorem statements so that readers can see it is not derived from the network model.
  3. [Appendix D, Agreement Property] The proof that every honest validator proposes the fast-path-decided block B to Quorum Consensus relies on Lemma 18, whose final case (SP-lock via a Quorum Consensus decision) uses Quorum Validity (II). Thus the Agreement proof is conditional on Claim 13. If the Quorum Consensus instantiation is only supplied later or at a different threshold, the Agreement and No Conflict proofs must be re-examined in full rather than merely cited as complete.
minor comments (4)
  1. [Abstract and Table 1] The abstract states 'a latency of 2 communication steps', but Table 1 reports 2delta for UA and RA-RA, and 2delta/3delta for UA-RA, while Property 27 states 3 communication steps for UA-RA. Please qualify the abstract or align the statements.
  2. [Remark 30] Remark 30 refers to 'Theorem 7', but no theorem with that number exists; the intended reference is likely a Property of reactive-actor validity or agreement. Please fix the cross-reference.
  3. [Algorithm 6, Part 3] Line 83 of Algorithm 6 uses the identifier 'B' where the context is the block decided in Quorum Consensus, B_qc. This is confusing and should be renamed for consistency with lines 82-94.
  4. [Lemma 32] The proof of Lemma 32 says 'In case tx is either UA-RA, RA-RA or RA'; the token 'RA' is not defined and should be either 'RA-RA' or a separate defined transaction type.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the No Conflict proof is derived internally from FP/SP-locking and imported consensus primitives, and the only terse existence claim (Claim 13) rests on an external theorem, not on the paper's own conclusion.

full rationale

I walked the derivation chain from the POA fast path through the slow path to Property 9. The fast-path agreement argument uses the quorum intersection of n-p vote sets; the slow-path No Conflict argument uses Lemma 18, proved from the FP/SP-locking protocol actions and from Quorum Validity (II), and then reduces cross-instance conflicts to the Agreement property of Transaction Agreement. None of these steps defines its conclusion into existence: the locking rules are concrete protocol actions, and the consensus primitives are stated as explicit assumptions with cited instantiations. Claim 13 is the only place where an existence proof is compressed, and it is a reduction to Theorem 5 and Definition 2 of [21], an external source with no author overlap; even if that reduction is terse and the appendix only sketches instantiations, this is an evidence/completeness gap rather than circularity. The self-citations in the paper ([41], [49], [36]) appear in related work and do not carry the load-bearing safety argument. Therefore no circular step can be quoted, and the score is 0.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

No numeric parameters are fitted to data; p, f, and Delta are model assumptions rather than fitted values. The protocol leans on imported consensus primitives and on the replicated actor model as stated assumptions, not on empirically fitted constants. No new physical or cryptographic entities, such as new particles, forces, or ledger entries, are introduced.

assumptions (6)
  • domain assumption At most f validators are Byzantine, and at most p misbehave during the fast path; n >= 3f + 2p + 1.
    Stated in Section 3.1 and used in every intersection argument that supports Agreement, No Conflict, and fast termination.
  • domain assumption The network follows the GST partial-synchrony model with known maximum delay Delta.
    Section 3.3 defines the model; liveness and all fast-path timing bounds depend on message delivery after GST.
  • ad hoc to paper A common leader oracle outputs an honest leader infinitely often.
    Section 5.2 requires this for liveness. The paper suggests random choice with a common random source or round-robin, but the common randomness is not constructed in a Byzantine environment.
  • ad hoc to paper There exists a Quorum Consensus algorithm satisfying Quorum Validity (Property 12) and a Transaction Agreement primitive.
    Claim 13 asserts existence by reducing to [21], but no concrete algorithm or implementation is provided; Appendix B only sketches adaptations of existing protocols.
  • standard math Perfect Links provide reliable point-to-point message delivery between entities.
    Section 3.1 assumes the standard Perfect Links abstraction from [19] for both Inner and Outer Links.
  • domain assumption Users issue non-conflicting transactions with increasing sequence numbers and only spend objects they own.
    Section 3.2 states this responsibility; the Validity and No Conflict proofs assume honest users do not create conflicting transactions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mangrove: Fast and Parallelizable State Replication for Blockchains." pith.science (2026). https://pith.science/paper/AWQCQSOD

@misc{pith2026250906616,
  author       = {Pith},
  title        = {Pith review of: Mangrove: Fast and Parallelizable State Replication for Blockchains},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AWQCQSOD}},
  note         = {Machine review of arXiv:2509.06616}
}
read the original abstract

Mangrove is a novel scaling approach to building blockchains with parallel smart contract support. Unlike in monolithic blockchains, where a single consensus mechanism determines a strict total order over all transactions, Mangrove uses separate consensus instances per smart contract, without a global order. To allow multiple instances to run in parallel while ensuring that no conflicting transactions are committed, we propose a mechanism called Parallel Optimistic Agreement. Additionally, for simple transactions, we leverage a lightweight Byzantine Reliable Broadcast primitive to reduce latency. Mangrove is optimized for performance under optimistic conditions, where there is no misbehavior and the network is synchronous. Under these conditions, our protocol can achieve a latency of 2 communication steps between creating and executing a transaction.

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 16 canonical work pages

  1. [21]

    URL: https://forum.dfinity.org/t/ lament-a-tale-of-constant-struggle-of-what-its-like-trying-to-scale-on-icp/35829

    Accessed: 2024-10-11. URL: https://forum.dfinity.org/t/ lament-a-tale-of-constant-struggle-of-what-its-like-trying-to-scale-on-icp/35829. 48 Victor Shoup. Practical threshold signatures. InAdvances in Cryptology—EUROCRYPT 2000: International Conference on the Theory and Application of Cryptographic Techniques Bruges, Belgium, May 14–18, 2000 Proceedings 1...

    work page 2024

  2. [1]

    Revisiting Fast Practical Byzantine Fault Tolerance: Thelma, Velma, and Zelma

    1 Ittai Abraham, Guy Gueta, Dahlia Malkhi, and Jean-Philippe Martin. Revisiting fast practical byzantine fault tolerance: Thelma, velma, and zelma.arXiv preprint arXiv:1801.10022,

    work page Pith review arXiv

  3. [9]

    Mysticeti: Low-latency DAG consensus with fast commit path.arXiv preprint arXiv:2310.14821,

    12 Kushal Babel, Andrey Chursin, George Danezis, Lefteris Kokoris-Kogias, and Alberto Son- nino. Mysticeti: Low-latency DAG consensus with fast commit path.arXiv preprint arXiv:2310.14821,

    arXiv

  4. [10]

    Sui lutris: A blockchain combining broadcast and consensus.arXiv preprint arXiv:2310.18042,

    16 Sam Blackshear, Andrey Chursin, George Danezis, Anastasios Kichidis, Lefteris Kokoris- Kogias, Xun Li, Mark Logan, Ashok Menon, Todd Nowacki, Alberto Sonnino, et al. Sui lutris: A blockchain combining broadcast and consensus.arXiv preprint arXiv:2310.18042,

    arXiv

  5. [12]

    Towards scaling blockchain systems via sharding

    24 Hung Dang, Tien Tuan Anh Dinh, Dumitrel Loghin, Ee-Chien Chang, Qian Lin, and Beng Chin Ooi. Towards scaling blockchain systems via sharding. InProceedings of the 2019 international conference on management of data, pages 123–140,

    work page 2019

  6. [13]

    URL: https:// aptosfoundation.org

    Accessed: 2024-10-11. URL: https:// aptosfoundation.org. 29Sui Foundation. Sui blockchain,

    work page 2024

  7. [14]

    URL: https://sui.io

    Accessed: 2024-10-11. URL: https://sui.io. 30 Davide Frey, Lucie Guillou, Michel Raynal, and François Taïani. Process-commutative distributed objects: From cryptocurrencies to byzantine-fault-tolerant crdts.Theoretical Computer Science, 1017:114794,

    work page 2024

  8. [15]

    The consensus number of a cryptocurrency

    18 Mangrove: Fast and Parallelizable State Replication for Blockchains 32 Rachid Guerraoui, Petr Kuznetsov, Matteo Monti, Matej Pavlovič, and Dragos-Adrian Seredin- schi. The consensus number of a cryptocurrency. InProceedings of the 2019 ACM Symposium on Principles of Distributed Computing, pages 307–316,

    work page 2019

Show all 26 references
  1. [16]

    Pi- lotfish: Distributedtransactionexecutionforlazyblockchains.arXiv preprint arXiv:2401.16292,

    36 Quentin Kniep, Lefteris Kokoris-Kogias, Alberto Sonnino, Igor Zablotchi, and Nuda Zhang. Pi- lotfish: Distributedtransactionexecutionforlazyblockchains.arXiv preprint arXiv:2401.16292,

    arXiv

  2. [17]

    38 Petr Kuznetsov, Andrei Tonkikh, and Yan X Zhang

    URL: https://arxiv.org/abs/2309.12715, arXiv: 2309.12715. 38 Petr Kuznetsov, Andrei Tonkikh, and Yan X Zhang. Revisiting optimal resilience of fast byzantine consensus. InProceedings of the 2021 ACM Symposium on Principles of Distributed Computing, pages 343–353,

    2021 arXiv

  3. [18]

    Fast leaderless byzantine total order broadcast.arXiv preprint arXiv:2412.14061,

    42 Matteo Monti, Martina Camaioni, and Pierre-Louis Roman. Fast leaderless byzantine total order broadcast.arXiv preprint arXiv:2412.14061,

    arXiv

  4. [20]

    Groundhog: Linearly-scalable smart contracting via commutative transaction semantics.arXiv preprint arXiv:2404.03201,

    46 Geoffrey Ramseyer and David Mazières. Groundhog: Linearly-scalable smart contracting via commutative transaction semantics.arXiv preprint arXiv:2404.03201,

    arXiv

  5. [22]

    Shoal: Improving DAG- BFT latency and robustness.arXiv preprint arXiv:2306.03058,

    51 Alexander Spiegelman, Balaji Arun, Rati Gelashvili, and Zekun Li. Shoal: Improving DAG- BFT latency and robustness.arXiv preprint arXiv:2306.03058,

    arXiv

  6. [23]

    Paramonov, Y

    A. Paramonov, Y. Vonlanthen, Q. Kniep, J. Sliwinski, R. Wattenhofer 19 52 Alexander Spiegelman, Neil Giridharan, Alberto Sonnino, and Lefteris Kokoris-Kogias. Bull- shark: DAG BFT protocols made practical. InProceedings of the 2022 ACM SIGSAC Conference on Computer and Communi...

    2022

  7. [24]

    Stingray: Fast concurrent transactions without consensus.arXiv preprint arXiv:2501.06531,

    53 Srivatsan Sridhar, Alberto Sonnino, and Lefteris Kokoris-Kogias. Stingray: Fast concurrent transactions without consensus.arXiv preprint arXiv:2501.06531,

    arXiv

  8. [25]

    Cryptocon- currency: (almost) consensusless asset transfer with shared accounts

    55 Andrei Tonkikh, Pavel Ponomarev, Petr Kuznetsov, and Yvonne-Anne Pignolet. Cryptocon- currency: (almost) consensusless asset transfer with shared accounts. InProceedings of the 2023 ACM SIGSAC Conference on Computer and Communications Security, pages 1556–1570,

    2023

  9. [26]

    Rapidchain: Scaling blockchain via full sharding

    56 Mahdi Zamani, Mahnush Movahedi, and Mariana Raykova. Rapidchain: Scaling blockchain via full sharding. InProceedings of the 2018 ACM SIGSAC conference on computer and communications security, pages 931–948,

    2018

  10. [2008]

    44 Ray Neiheiser, Arman Babaei, Giannis Alexopoulos, Marios Kogias, and Eleftherios Koko- ris Kogias

    Accessed: 2025-01-20. 44 Ray Neiheiser, Arman Babaei, Giannis Alexopoulos, Marios Kogias, and Eleftherios Koko- ris Kogias. Pythia: Supercharging parallel smart contract execution to guide stragglers and full nodes to safety.Workshop on Scalability & Interoperability of Blockc...

    2025

  11. [2017]

    Asynchronous bft asset transfer: Quasi-anonymous, light, and consensus-free.arXiv preprint arXiv:2405.18072,

    6 Timothé Albouy, Emmanuelle Anceaume, Davide Frey, Mathieu Gestin, Arthur Rauch, Michel Raynal, and François Taïani. Asynchronous bft asset transfer: Quasi-anonymous, light, and consensus-free.arXiv preprint arXiv:2405.18072,

    arXiv

  12. [2018]

    Good-case latency of byzantine broadcast: A complete categorization

    2 Ittai Abraham, Kartik Nayak, Ling Ren, and Zhuolun Xiang. Good-case latency of byzantine broadcast: A complete categorization. InProceedings of the 2021 ACM Symposium on Principles of Distributed Computing, pages 331–341,

    2021

  13. [2019]

    Shoal++: High throughput DAG BFT can be fast!arXiv preprint arXiv:2405.20488,

    11 Balaji Arun, Zekun Li, Florian Suri-Payer, Sourav Das, and Alexander Spiegelman. Shoal++: High throughput DAG BFT can be fast!arXiv preprint arXiv:2405.20488,

    arXiv

  14. [2021]

    Pod: An optimal-latency, censorship-free, and accountable generalized consensus layer.arXiv preprint arXiv:2501.14931,

    8 Orestis Alpos, Bernardo David, and Dionysis Zindros. Pod: An optimal-latency, censorship-free, and accountable generalized consensus layer.arXiv preprint arXiv:2501.14931,

  15. [2022]

    On the validity of consensus

    21 Pierre Civit, Seth Gilbert, Rachid Guerraoui, Jovan Komatovic, and Manuel Vidigueira. On the validity of consensus. InProceedings of the 2023 ACM Symposium on Principles of Distributed Computing, pages 332–343,

    2023

  16. [2023]

    Fast transaction scheduling in blockchain sharding.arXiv preprint arXiv:2405.15015,

    4 Ramesh Adhikari, Costas Busch, and Miroslav Popovic. Fast transaction scheduling in blockchain sharding.arXiv preprint arXiv:2405.15015,

    arXiv

  17. [2024]

    Chainspace: A sharded smart contracts platform.arXiv preprint arXiv:1708.03778,

    5 Mustafa Al-Bassam, Alberto Sonnino, Shehar Bano, Dave Hrycyszyn, and George Danezis. Chainspace: A sharded smart contracts platform.arXiv preprint arXiv:1708.03778,

    arXiv

  18. [2025]

    Sharper: Sharding permissioned blockchains over network clusters

    9 Mohammad Javad Amiri, Divyakant Agrawal, and Amr El Abbadi. Sharper: Sharding permissioned blockchains over network clusters. InProceedings of the 2021 international conference on management of data, pages 76–88,

    2021

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.