Skip to content
Crypto.world

Research

Back to the Future: Rethinking Endorsement in Order-Execute Blockchains

Modern permissioned blockchains need flexible endorsement — the ability to define, per contract or state object, which parties must approve a transaction and its execution results — for regulatory…

Back to the Future: Rethinking Endorsement in Order-Execute Blockchains
arXiv
2604.27659
Published
2026-04-30
Authors
Rongji Huang, Yifeng Ye, Gerui Wang, Mingchao Wan, Yuxing Duan, Jingjing Zhang, Guangtao Xue, Shengyun Liu

AI summary

Overview

Research area: Permissioned blockchain architecture and Byzantine fault-tolerant (BFT) consensus — specifically, how to support flexible endorsement (per-contract/per-object approval policies) inside a classical order-execute blockchain rather than by adopting Hyperledger Fabric's execute-order-validate (EOV) model.

Technical level: Advanced. The paper assumes familiarity with state machine replication, PBFT/Tendermint round structure, quorum certificates, locking, and eventual synchrony (GST, Δ).

Scope in one sentence: The paper proposes an order-execute-endorse framework and its Tendermint-based instantiation, FlexTender, which embeds endorsement into consensus messages so that poorly endorsed transactions are deterministically removed without extra messaging in the normal case.

What This Paper Is About

Modern permissioned blockchains need flexible endorsement — the ability to define, per contract or state object, which parties must approve a transaction and its execution results — for regulatory compliance and governance. Hyperledger Fabric supports this by executing and endorsing transactions before ordering, but that design suffers high abort rates under the high-contention workloads typical of DeFi, hinders batching of cryptographic operations, and can threaten liveness when hot-spot state is repeatedly aborted. The authors instead keep the order-execute structure used by most systems (including their target platform, ChainMaker) and add endorsement after ordering and execution, solving the resulting problem of how to deterministically remove transactions that fail to get endorsed while preserving censorship resistance and decentralization for the rest.

Key Contributions

  1. The order-execute-endorse framework. A new architecture in which transactions are (1) batched and sequenced, (2) executed in the order given by consensus, (3) endorsed by the designated endorsers — with removal and re-execution/re-endorsement of subsequent transactions if any transaction fails endorsement — and (4) committed with the world state updated. The authors frame the design space through "Dilemma 1": either order before endorsement (avoiding conflicts but risking delays from pending endorsements) or endorse before ordering (avoiding delays but risking abort-causing conflicts). Fabric takes the second path; this paper takes the first.

  2. FlexTender, a Tendermint-based protocol. The authors embed endorsement into Tendermint's existing prevote/precommit message exchange. prevote messages carry endorsers' views on transactions, precommit messages signal that a block has collected sufficient endorsements or suggest removing a transaction that has not. Round advancement doubles as the re-execution/re-endorsement cycle, and timeouts double as prompts to remove unendorsed transactions. The authors state that, to their knowledge, FlexTender is the first BFT protocol supporting flexible endorsement while remaining conflict-free.

  3. Mechanisms against three specific liveness/fairness threats. (a) A refRound variable (paired with refValue) prevents unnecessary re-endorsement and tracks the evolution of examined proposals so that proposers pick either a properly endorsed QC value or the examined value with the fewest transactions; (b) partial endorsements are aggregated across a round, with two conflicting prevote messages from the same endorser jointly interpreted as endorsing all transactions; (c) censorship is resisted via rapid removal certificates, a CON flag letting endorsers broadcast endorsements even when they vote against the proposal at the consensus layer, and an immediate nil precommit when two differing proposals arrive in the same round. Removal requires at least f+1 nodes suggesting it (at least one correct; the paper notes this may rise to 2f+1 if needed).

  4. Implementation and evaluation on ChainMaker. FlexTender is implemented on top of ChainMaker and evaluated on Amazon EC2 against eov-sim, an EOV simulation built on the same codebase, using an Ethereum USDT transfer workload. Correctness proof and pseudocode appear in Appendices A and B, where modifications to Tendermint are also highlighted.

Main Findings

  • Throughput gain: Compared with eov-sim on the same platform, FlexTender achieves up to 10.6× speedup in throughput on the Ethereum USDT workload.
  • No extra messages in the normal case: The message exchange per round does not exceed Tendermint's normal procedure; under a correct leader, with all transactions endorsable and a synchronous network, FlexTender can commit a proposal (also) in three message delays.
  • Workload skew motivates the design: In the USDT workload derived from Ethereum between block heights 23,700,767 and 23,722,235, the top 0.01% of addresses are accessed by 39.23% of all transactions — the contention pattern that makes EOV's abort-and-retry behavior costly.
  • Fast removal path: FlexTender provides a fast path to remove an unendorsed transaction when there is proof it was vetoed by its endorsers ("rapid removal"), using Fabric-style monotone logical expressions over sets both to check endorsement and to check opposition.
  • Safety and liveness are stated as theorems: Theorem 1 (Safety with endorsement): every committed transaction must be properly endorsed. Theorem 2 (Liveness with endorsement): every properly endorsed transaction that cannot be rapidly removed will be committed eventually. Safety is independent of synchrony assumptions; synchrony is used for liveness and for censorship resistance through timely propagation of endorsements and conflicting messages.
  • Key invariant: Invariant 1 — if a proposal v is examined in round r (some correct node received n−f precommit messages for it), then either v is properly endorsed or at least one transaction in v can be removed in subsequent rounds. Invariant 2 — after GST, if enough endorsers endorse tx and tx cannot be rapidly removed, no correct node suggests removing tx.
  • Responsiveness: FlexTender guarantees responsiveness for both the normal and rapid-removal cases, meaning if enough endorsers progress at the pace of the actual network, the whole protocol does so as well.
  • Correctness reuses Tendermint: Because FlexTender relies largely on Tendermint's locking mechanism and has been widely adopted and analyzed, its safety argument inherits from a well-studied base.
  • Not reported in the provided excerpt: absolute throughput/latency numbers, node counts, EC2 instance types, workload sizes in transaction counts, and the full evaluation section (the text is truncated within Phase III of the protocol description).

Methodology in Plain English

The authors start from an architectural observation: Fabric's flexible endorsement works because transactions are endorsed before they are ordered, so a pending endorsement never blocks anything else. But that ordering forces a validation phase that aborts conflicting transactions — painful in DeFi, where many transactions touch the same accounts.

Their alternative keeps transactions in the familiar order-execute pipeline: consensus fixes the order, nodes execute in that order, and then the designated endorsers sign off on the results. Since removing one transaction can change the execution results of the transactions after it, removal triggers re-execution and re-endorsement of the successors. The obvious danger is that this removal-and-retry loop never terminates, or that malicious nodes exploit it to censor specific transactions.

Their key trick is to notice that Tendermint already has a round-by-round loop with timeouts. Instead of building a separate endorsement subsystem (which they tried, and found produced awkward nested rounds — an "inner loop" for consensus and an "outer loop" for endorsement), they reuse the existing loop: each round's prevote carries endorsements, each round's precommit carries either "this block is fully endorsed" or "please drop this transaction," and timeouts handle endorsers that are slow or offline. To make this safe they add bookkeeping (refValue/refRound, lockedValue/lockedRound updated only when endorsements accompany a quorum certificate) so a proposal can only get shorter across rounds, never gain new transactions, which prevents infinite add/remove loops. They then close three attack avenues: an endorser trickling its signature late each round, an endorser splitting endorsements so no single node sees a full set, and a malicious proposer using a withheld endorsement as a pretext to strip an unrelated transaction. Finally, they implement this on ChainMaker and compare against an EOV simulation sharing the same codebase, using real USDT transfer patterns from Ethereum as the workload.

Why This Matters

For research, the paper reframes flexible endorsement as something that does not require inverting the order-execute model, which is the assumption behind a long line of Fabric improvement work. It shows that endorsement and consensus can share one message pattern, and gives a formal handle on the problem through "examined" proposals, two invariants, three requirements, and a safety/liveness theorem pair. It also identifies a structural limit of EOV that is not merely a performance detail: repeated aborts on hot-spot state can undermine liveness.

Real-world applications the paper's design supports:

  • Central-bank / commercial-bank minting: A coin-minting transaction must be endorsed by one node representing the central bank and another representing a commercial bank.
  • Cross-border transfers: The relevant banks must endorse the transfer, including its execution results.
  • Regulated securities and derivatives: An interest-rate swap must involve an auditor to confirm the swap is priced at fair value, which fluctuates as market conditions change — which is why endorsements must attach to execution results, not just to the transaction.
  • Regulator approval of specific operations: A swap transaction must be endorsed by the securities regulator, while unrelated transactions remain unaffected by any endorser's behavior.

Industry relevance: the work is aimed at ChainMaker, an enterprise-grade platform developed in part by some of the authors and deployed across dozens of state-owned enterprises in China, and it targets the regulatory climate signalled by the U.S. GENIUS Act and the Hong Kong Stablecoins Ordinance. Because FlexTender is a modification of Tendermint — the base for Cosmos, BNB Beacon Chain, and ChainMaker — the approach is applicable well beyond a single codebase, and the "no additional messaging overhead in the normal case" property matters for operators who cannot afford extra network cost.

Future Directions

  • Full evaluation beyond throughput: The excerpt reports only relative throughput against eov-sim; latency, abort rates, behavior under varying contention, and scaling with node count are natural next measurements.
  • Combining with accountability: The authors explicitly suggest pairing the protocol with accountability mechanisms to detect faulty processes and alleviate the censorship problems they describe.
  • Fair ordering and separated block building: The paper notes that censorship risk is exacerbated when a fair-ordering component determines the initial proposal or when block creation is delegated to a separate role — how FlexTender composes with those settings is left open.
  • Policy design and decentralization of endorsers: The authors state that endorsement policies "should themselves be (partially) decentralized," but leave this to application designers; how to enforce or measure that remains unaddressed.

Target Audience

Consensus-protocol researchers and graduate students working on BFT and permissioned blockchain architecture; engineers building or operating ChainMaker-, Cosmos-, or Tendermint-based chains who need compliance-grade approval workflows without migrating to Fabric; and enterprise architects in regulated finance (stablecoins, cross-border payments, tokenized securities) evaluating whether flexible endorsement can be retrofitted onto an existing order-execute stack. Readers should be comfortable with quorum certificates, view/round changes, and locking before the protocol section.

Authors’ abstract

Due to regulatory compliance and governance management, modern (permissioned) blockchains require flexible endorsement, which allows the endorsement policy for each contract or state object to be individually defined. To enable flexible endorsement, Hyperledger Fabric employs an execute-order-validate (EOV) paradigm, in which transactions first undergo speculative execution and endorsement, and are only then ordered and validated. Meanwhile, most blockchain systems,…

The full abstract on arXiv
Read the original paper