Beyond Byzantine: An Organizational Consensus Algorithm for Self-Interested Agents Under Information Asymmetry

Jianbo Liu; Jiawei Zhang

REVIEW 4 major objections 5 minor 21 references

Beyond Byzantine: An Organizational Consensus Algorithm for Self-Interested Agents Under Information Asymmetry

T0 review · 4 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read An organizational consensus protocol can cut coordination overhead by ~99% while keeping self-interested agents truthful, via event-triggered updates and penalties from delayed audits.

desk verdict A genuinely integrated framework sketch with plausible overhead savings, but the central truthfulness threshold is self-confirming and the convergence proof does not match the protocol; worth a serious referee only with major revision. read the letter →

arxiv 2607.28957 v1 pith:CBARF3AQ submitted 2026-07-31 cs.GT cs.DCcs.MA

Jiawei Zhang , Jianbo Liu This is my paper · ORCID

classification cs.GTcs.DCcs.MA

keywords organizationalconsensusmechanismdesignincompleteinformationgamestrategicagentscoordinationoverheaddelayedverificationBayesianNashequilibriumevent-triggeredcommunication

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

This paper argues that the binary honest-versus-Byzantine view of distributed consensus is the wrong lens for organizational decision-making, where departments are bounded-rational, self-interested, and information-asymmetric. It proposes the Organizational Consensus Algorithm (OCA), which replaces mandatory all-hands coordination with exception-triggered delta broadcasts and confidence-weighted aggregation, and pairs that fast path with a slow path of retrospective penalties tied to delayed verifiable outcomes like quarterly audits. The paper claims that this design simultaneously delivers large coordination savings (about 99% bandwidth reduction at N=20 in simulation), fast alignment (16 rounds), and truthfulness: above a computed penalty threshold λ_min = 0.5, strategic bias becomes unprofitable and an ε-truthful Bayesian Nash equilibrium emerges. A sympathetic reader should care because it suggests a principled way to align incentives in organizations without sacrificing efficiency, using ordinary audit signals rather than cryptographic assumptions.

What carries the argument

The load-bearing object is the pair (fast path, slow path): fast path uses exception-triggered delta broadcasts (only when local deviation exceeds threshold τ_i or version advances) and confidence-weighted aggregation x_j^{t+1} = Σ w_i x̂_i / Σ w_i with w_i = √c_i; slow path applies retrospective penalties p_i = λ||x̂_i − y_i||² from a delayed oracle and trust decay c_i^{(k+1)} = max(0, c_i^{(k)} − η p_i^{(k)}). The square-root weighting enforces diminishing returns on staked political capital; the quadratic penalty creates the incentive gradient that makes bias unprofitable above λ_min; and the trust decay couples past honesty to future influence, deterring one-shot rug pulls. This two-time

What would settle it

A concrete falsifying experiment would run OCA in the same simulation but with the delayed oracle disabled (or with oracle variance driven to infinity) and check whether system-average bias remains bounded for any λ; if bias grows monotonically with B_max/λ, the ε-truthfulness claim fails. For the convergence theorem, a simulation using a time-varying topology that is not strongly connected at every round (as event-triggering permits) that fails to reach consensus would falsify Theorem 3.

Watch

Extended reading notes

Core claim

OCA's central claim is that truth-telling can be induced by a retrospective quadratic penalty p_i = λ||x̂_i − y_i||² tied to a delayed verifiable oracle, plus a trust score c_i that sets voting weight w_i = √c_i and erodes with penalties. The paper proves an ε-truthful Bayesian Nash equilibrium when λ ≥ (B_max − B_min)/σ²_oracle (Theorem 1), a rug-pull deterrence condition (Theorem 2), and asymptotic consensus under a fixed strongly connected topology with self-loops (Theorem 3). Simulations show the predicted phase transition: at λ_min = 0.5 system bias collapses and malicious utility becomes negative, while honest agents keep high trust and positive utility.

Load-bearing premise

The load-bearing assumption is that a delayed, verifiable ground-truth signal actually arrives (otherwise penalties are zero and the incentive guarantees collapse), together with a fixed strongly-connected communication topology that event-triggered updating can violate.

Editorial extensions

If this is right

If OCA is right, organizations can replace mandatory all-hands coordination with exception-triggered updates and still reach bounded-error consensus, cutting bandwidth by ~99% at N=20 and more at larger scales.
The penalty threshold λ_min gives a principled calibration rule: set λ ≥ (B_max − B_min)/σ²_oracle and no over-penalization is needed; excessively harsh penalties only hurt honest agents.
The protocol's domain is limited to numerical/mergeable states with bounded error; it is not a substitute for total-order replication when strict linearizability is required.
The formal guarantees include an ε-truthful Bayesian Nash equilibrium and a rug-pull deterrence condition, meaning the design is intended to be robust to strategic agents who might 'earn trust then exploit it.'
Asymptotic consensus holds when the communication graph is fixed, strongly connected, and has self-loops; convergence is guaranteed by Perron-Frobenius theory.

Reading between the lines

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

The paper's own limitation section concedes that if delayed verification never arrives, malicious agents face no consequences; a natural extension is to test how the mechanism degrades under audit frequency or oracle noise, which the current analysis does not quantify.
Because event-triggered suppression makes the effective interaction graph time-varying, Theorem 3's fixed-topology assumption is stronger than the protocol actually operates under; checking convergence bounds for time-varying graphs would be a direct stress test.
The ε-truthfulness guarantee depends on Gaussian oracle noise and concave utility; in non-convex negotiation spaces (e.g., discrete contract terms) the quadratic penalty may not induce truthfulness, suggesting a need for alternative loss functions.
The mechanism suggests a broader design principle: any organization with periodic, verifiable performance data can retrofit incentive alignment onto otherwise wasteful coordination processes, without cryptographic assumptions.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Quick take: the integration of event-triggered communication, trust-weighted averaging, and delayed-audit penalties is real, and the direct overhead simulation is plausible. The load-bearing incentive results are not: the BNE-threshold experiment in Section VI.G is circular, and Theorem 3 applies a static Perron-Frobenius argument to a protocol whose neighborhoods and weights change every round. I would not desk-reject, but I would send it back expecting major revision before it can be trusted.

What is genuinely new: framing organizational departments as strategic agents with trust stakes and episodic ground truth, then combining exception-triggered updates with quadratic penalties, is a useful framework sketch. The related work is current and the authors are disciplined about scope: they do not claim to replace linearizable consensus, and they explicitly flag the oracle-availability condition in Section VI.I. If y_i never arrives, the penalty is zero and the strategic guarantees collapse; that is an honest limitation, not a hidden one. The 99% bandwidth-saving numbers are the strongest part of the paper and are at least reproducible in principle from the protocol description.

Where it falls down. First, Theorem 3 states asymptotic consensus for a static strongly connected topology with self-loops, but Eq. (2) suppresses transmissions and Eq. (4) decays trust, so W_t changes over time. Perron-Frobenius on a fixed row-stochastic matrix does not cover that. This may be fixable by switching to products of stochastic matrices with the right conditions, but it is not what is written.

Second, the BNE threshold verification is self-confirming. The paper states the FOC as Bmax - 2*lambda*beta = 0, which gives beta* = Bmax/(2*lambda), independent of sigma^2_oracle and Bmin. Eq. (18) instead uses (Bmax - Bmin)/(2*lambda*sigma^2_oracle) and inserts a discontinuity at lambda_min; no optimization yields that. Since the simulated adversaries are assigned that externally imposed beta*(lambda), the cliff in Fig. 6 is an echo of the input. This is the exact circularity the stress-test flags, and I agree it lands on the paper. The paper's own Section VI.I says the full BNE proof lives in supplementary material; the submitted text does not contain it.

Minor: experiments report medians without variance and no code is released. The sensitivity grid is nice but does not repair the threshold problem.

Bottom line: this is a framework worth discussing, not a result worth citing yet. The overhead experiment and the protocol idea justify referee time, but the strategic section needs to be rebuilt from the actual FOC and re-validated. I would accept it for review with a clear expectation of major revision.

Referee Report

4 major / 5 minor

Summary. The paper proposes the Organizational Consensus Algorithm (OCA), a mechanism-design framework for reaching bounded-error consensus among self-interested, information-asymmetric agents. OCA combines event-triggered delta broadcasts, confidence-weighted local aggregation, and retrospective quadratic penalties based on a delayed verifiable oracle. The manuscript claims an ε-truthful Bayesian Nash equilibrium (Theorem 1), a rug-pull deterrence condition (Theorem 2), asymptotic consensus via Perron-Frobenius (Theorem 3), and simulation evidence of 99% bandwidth savings, 16-round convergence, perfect correctness, and a BNE penalty threshold λ_min = 0.5 (Eq. 17, Fig. 6). The paper is explicit that its empirical claims are conditional on the simulation model and that a full equilibrium proof is deferred to supplementary materials.

Significance. If the formal guarantees were correct, OCA would be a useful contribution: it connects mechanism design to distributed consensus, proposes a practical incentive layer based on delayed audits, and gives a concrete simulation prototype. The paper also deserves credit for candidly stating its scope limitations (e.g., vector averaging only, oracle dependence, no general truthful-equilibrium claim). However, as it stands, the central theoretical and empirical pillars are not sound: Theorem 3 ignores the time-varying nature of the event-triggered communication graph, Theorem 1 is a single-agent optimization rather than a Bayesian Nash equilibrium proof, and the headline BNE threshold experiment is circular because the simulated adversary's bias is set by a formula that does not follow from the stated utility model. These issues undermine the paper's main claim that OCA simultaneously preserves truthfulness and reduces coordination overhead.

major comments (4)

[§VI.G, Eq. (18)] The rational-adversary bias formula β*(λ)=(Bmax−Bmin)/(2λσ²_oracle) for λ<λmin and 0 for λ≥λmin does not follow from the stated first-order condition. The FOC for maximizing Bmax·β−λβ² is Bmax−2λβ=0, giving β*=Bmax/(2λ), independent of both Bmin and σ²_oracle. The discontinuity at λmin is also not implied by any optimization. Because the simulation's adversarial bias is presumably generated from Eq. (18), the cliff-drop in Fig. 6 is imposed by construction rather than being an emergent equilibrium response. This invalidates the paper's empirical 'verification' of the BNE threshold and the claimed support for Theorem 1.
[§V.C, Theorem 3] The proof treats W as a fixed row-stochastic matrix and applies the Perron-Frobenius theorem to W^t. However, the update in Eq. (1) uses P_{j,t}, the neighbor set at time t, which changes over time because Eq. (2) suppresses broadcasts when local deviations are below threshold. The matrix W_t in the proof is time-varying. Even if each W_t is row-stochastic and the underlying graph is strongly connected at every time, convergence of the product of time-varying stochastic matrices requires additional conditions (e.g., repeated strong connectivity, bounded communication intervals); a fixed-matrix Perron-Frobenius argument is not valid. The theorem therefore does not prove asymptotic consensus for the protocol actually described.
[§V.A, Theorem 1] The proof analyzes a single agent's one-dimensional optimization and derives a condition on λ so that the optimal bias has norm below ε_tol. This is not a Bayesian Nash equilibrium: the paper does not define a game with all agents' strategies, beliefs, and payoffs, nor does it account for the fact that agents' reports enter the consensus update in Eq. (1), which may affect future penalties and benefits. The theorem's label 'ε-Truthful Bayesian Nash Equilibrium' is therefore unsupported as stated. The manuscript itself defers the full equilibrium proof to supplementary materials (§VI.I, §VII), but the main text should at least state precisely which game form is being solved.
[§VI, Tables II and Figs. 3–6] All experiments are summarized by medians over 30 Monte Carlo runs, with no variance, confidence intervals, or significance tests. This makes it impossible to assess whether reported differences (e.g., OCA Tε=16 vs. DeltaAntiEntropy Tε=12 in Table II; the precise positions of the cliff in Fig. 6) are meaningful. Since the BNE-threshold experiment is already circular, the statistical reporting further weakens the empirical contribution. Please provide distributions or confidence bands, or at least justify why median-only comparisons are sufficient for the claims made.

minor comments (5)

[§VI.C, Fast path] The equation x̂(t+1)_g = 0.4 x̂(t)_g + 0.6 x̄_w introduces ẑ_w without defining the weighted average x̄_w; this appears to be an informal summary of Eq. (1), but it should be stated precisely.
[§VI.F, Eq. (16)] The per-round utility formula includes min(β_i,1.0) and κ(1−c_i), but the relationship between this experimental utility and the theoretical utility in Eq. (3) is not fully specified. Clarify whether Eq. (16) is the actual simulation objective or a simplified proxy.
[§VI.H, Fig. 7] The caption lists only panels (h) and (i), while the text describes top-right and bottom-right panels for honest bias and honest utility. The panel labels are inconsistent and should be fixed.
[§I and §IV] The term 'bounded welfare loss' in the abstract is not precisely defined; Eq. (8) defines C_ε as a correctness score, not a welfare loss. Align terminology.
[§VI.I and §VII] The paper correctly notes that the retrospective penalty requires a delayed verifiable oracle and that without it 'malicious agents face no consequences.' This is a major scope assumption; it should be stated as a condition on all theorems (not only as a validity threat), since Theorems 1–2 depend on the oracle arriving.

Circularity Check

1 steps flagged · score 6.0 of 10

Fig. 6's BNE threshold is built into the adversary model (Eq. 18), making the central truthfulness validation circular.

self definitional [Section VI.G, Eq. (17)-(18), Fig. 6 and surrounding text]
"The first-order condition d/dβ[B maxβ−λβ 2] =B max −2λβ= 0 yields: β∗(λ) = ( Bmax−Bmin 2λσ2 oracle ifλ < λmin 0 ifλ≥λ min (18) ... When λ crosses λmin, the expected penalty exceeds the maximum achievable benefit, making bias unprofitable and truth-telling the dominant strategy."

By the paper's own displayed FOC, the optimal bias is β* = Bmax/(2λ) for a linear benefit Bmaxβ; no σ²_oracle or Bmin appears, and no discontinuity is implied. Eq. (18) instead inserts (Bmax−Bmin)/(2λσ²_oracle) and a hard cutoff at λmin=(Bmax−Bmin)/σ²_oracle. Since the simulated 'rational adversary' is defined by this externally imposed β*(λ), the cliff in Fig. 6 at λ=0.5 is a restatement of the input rule, not an empirical consequence of Theorem 1. Thus Fig. 6 cannot validate the ε-truthful BNE threshold; the threshold is an input to the experiment, not a prediction confirmed by it.

full rationale

The direct simulation comparisons (bandwidth savings, convergence latency, Cε) are self-contained and not circular. The problem is the central BNE-threshold experiment. In Section VI.G the paper states Eq. (18), the adversary's optimal bias, with a cutoff at λmin; but the FOC just above it gives β*=Bmax/(2λ), with no dependence on σ²_oracle or Bmin and no cutoff. The displayed β*(λ) is therefore an input assumption, and Fig. 6's observed cliff and utility inversion are programmed in by construction. Consequently the claim that λ≥0.5 deters bias is not independently verified. This is amplified by the paper's own admission in Section VI.I that the formal BNE proof is relegated to supplementary materials ('see docs/proofs/bne_equilibrium.tex') and that 'incentive compatibility is conditional on the stated model assumptions.' The overhead results remain non-circular, but the central truthfulness guarantee rests on a self-confirming experiment; hence score 6 rather than 0-2.

Assumptions & free parameters 4 free parameters · 5 assumptions · 1 invented entities

The central formal results rest on standard mathematical assumptions (Perron-Frobenius, FOC) plus domain assumptions about oracle availability, Gaussian noise, concavity, and strong connectivity. The simulation adds hand-set parameters that directly determine the claimed BNE threshold, making that threshold fitted rather than independently predicted.

free parameters (4)

BNE threshold λmin = 0.5
Presented as a theoretical prediction, but Eq (17) sets it to (Bmax−Bmin)/σ²_oracle using hand-picked simulation parameters; Eq (18)'s β*(λ) formula does not follow from the stated FOC Bmax − 2λβ = 0, so the threshold is an ad hoc fit to the simulation.
Global update mixing coefficients = 0.4 / 0.6
Figure 3 describes x̂^{t+1}_g = 0.4 x̂^t_g + 0.6 x̄_w; this rule is not in the protocol Eq (1) and its coefficients are chosen by hand to produce smooth convergence.
Residual penalty rate ρ_r = 0.1
Section VI.D defines a penalty p_i = ρ_r·||x̂_i − x̂_g||², a different form than Eq (3)'s λ||x̂_i − y_i||²; introduced ad hoc in the welfare experiments.
Simulation noise parameters = σ=0.02, σ²_oracle=2.0
Hand-set inputs; the claimed λmin=0.5 is a deterministic function of these choices, so the 'predicted' threshold has no independent status.

assumptions (5)

standard math Perron-Frobenius theorem for nonnegative irreducible aperiodic matrices
Used in Proof of Theorem 3 to argue W^t → 1v^T; valid for a fixed row-stochastic matrix but not for the time-varying, event-triggered W_t defined by the protocol.
domain assumption Communication graph G is strongly connected and contains a self-loop at every round
Theorem 3 assumes it, but the protocol's Eq. (2) allows agents to remain silent when deviations are below τi, so edges can disappear and the matrix can become non-irreducible.
domain assumption Oracle signal y_i = x*_i + ε_i with ε_i ~ N(0, σ_y² I)
Assumed in Theorem 1 and simulations; real audits may be biased, discrete, or missing; paper notes this in Threats to Validity.
domain assumption Agent utility B_i is monotone increasing and concave, differentiable near x*
Used in Theorem 1's FOC; organizations may have non-convex preferences.
domain assumption Periodic availability of the delayed oracle
Necessary for penalties; acknowledged as limitation in VI.I.

invented entities (1)

Organizational trust stake c_i
purpose: Represents pledged budget/bonus/political capital; weights proposals and is slashed by penalties
A modeling abstraction; the paper gives no empirical calibration for the sqrt-weighting or the decay dynamics, and the same variable appears inconsistently (weight √c_i vs. r_i√c_i) in Eq (1) and text.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Beyond Byzantine: An Organizational Consensus Algorithm for Self-Interested Agents Under Information Asymmetry." pith.science (2026). https://pith.science/paper/CBARF3AQ

@misc{pith2026260728957,
  author       = {Pith},
  title        = {Pith review of: Beyond Byzantine: An Organizational Consensus Algorithm for Self-Interested Agents Under Information Asymmetry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CBARF3AQ}},
  note         = {Machine review of arXiv:2607.28957}
}

read the original abstract

Traditional distributed consensus protocols classify nodes as either honest-but-faulty or actively malicious (Byzantine). However, in organizational structures, departmental agents rarely fit this binary. Instead, they exhibit bounded rationality and self-interested preferences while operating under asymmetric information. This paper presents the Organizational Consensus Algorithm (OCA), a mechanism design framework tailored for internal negotiation and decision coordination. OCA models inter-departmental conflict as an incomplete information dynamic game, integrating internal token staking, an exception-triggered challenge mechanism, and confidence-weighted consensus rules. Rather than enforcing instantaneous total ordering, OCA leverages a retrospective penalty system driven by delayed verifiable outcomes to deter structural bias and reduce exhaustive coordination overhead. A Python simulation prototype was developed to evaluate OCA. Across independent trials with varying organizational scales, OCA reports lower coordination overhead, higher informative reporting rates, and bounded welfare loss in noisy environments. Crucially, these results remain conditional on the stated simulation model and do not by themselves establish a general truthful equilibrium.

Figures

Figures reproduced from arXiv: 2607.28957 by the authors.

**Figure 3.** shows alignment latency Tε across protocols under an initial bias of xˆ (0) g = 0.5 (i.e., a 50% deviation from ground truth). OCA converges within a bounded number of iterations that scales sub-linearly with N. The admissibility-weighted aggregation in (1) accelerates convergence by prioritizing historically accurate departments—high-trust nodes exert greater influence on the global state via wi = ri √ ci , dampen… view at source ↗

**Figure 4.** Welfare correctness score Cε across update sparsity ratios ρ. Effective noise scales as σeff = σ · ρ. OCA maintains stable alignment (Cε ≥ 0.85) across all conditions, demonstrating robustness against bounded rationality and information asymmetry. v (t) i ≻ v last i ensures only causally fresh updates enter the aggregation. Confidence-weighted aggregation: Nodes with high historical accuracy (large ci) receive ampl… view at source ↗

Figures from the paper (3 more)

**Figure 5.** Figure 5: Strategic resilience under malicious node injection. (Left panel) Average trust score [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 6.** Figure 6: BNE critical threshold verification. (Left) System bias [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

**Figure 7.** Figure 7: Parameter sensitivity across κ × η grid (6 × 6). (Top-left) Malicious bias remains uniformly low across all parameter combinations. (Top-right) Honest bias stays near zero. (Bottom-left) Malicious utility is negative across the entire grid, confirming incentive compati…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 4 canonical work pages

[1]

The part time parliament,

L. Lamport, “The part time parliament,”ACM Transactions on Computer Systems, vol. 16, no. 2, pp. 133–169, May 1998, doi: 10.1145/279227.279229

arXiv 1998
[2]

In search of an understandable consensus algorithm,

D. Ongaro and J. Ousterhout, “In search of an understandable consensus algorithm,” inProc. USENIX Annual Technical Conference, 2014, pp. 305–319

2014
[3]

LowPaxos: State machine replication for low resource settings,

A. Mwotil, T. Anderson, B. Kanagwa, T. Stavrinos, and E. Bainomugisha, “LowPaxos: State machine replication for low resource settings,”IEEE Access, pp. 91272–91288, 2024, doi: 10.1109/ACCESS.2024.3421582

arXiv 2024
[4]

Hamava: Fault tolerant reconfigurable geo replication on heterogeneous clusters,

T. Mane, X. Li, M. Sadoghi, and M. Lesani, “Hamava: Fault tolerant reconfigurable geo replication on heterogeneous clusters,” inProc. IEEE Int. Conf. on Data Engineering, 2025, pp. 2024–2037, doi: 10.1109/ICDE65448.2025.00154

arXiv 2025
[5]

Avicenna: Masking slowdowns in replicated state machines with counterfactual evaluation,

C. Hodsdon, Z. Qin, K. Ngo, S. Sen, E. Katz Bassett, and W. Lloyd, “Avicenna: Masking slowdowns in replicated state machines with counterfactual evaluation,” inProc. European Conf. on Computer Systems, 2026, doi: 10.1145/3767295.3803615

arXiv 2026
[6]

Conflict free replicated data types,

M. Shapiro, N. Pregui c ¸a, C. Baquero, and M. Zawirski, “Conflict free replicated data types,” inStabilization, Safety, and Security of Distributed Systems, 2011, pp. 386–400, doi: 10.1007/978-3-642-24550-3.29

work page doi:10.1007/978-3-642-24550-3.29 2011
[7]

Efficient state based CRDTs by delta mutation,

P. S. Almeida, A. Shoker, and C. Baquero, “Efficient state based CRDTs by delta mutation,” inProc. Int. Conf. on Networked Systems, 2014, pp. 62–76, doi: 10.1007/978-3-319-26850-7.5

work page doi:10.1007/978-3-319-26850-7.5 2014
[8]

Efficient synchro- nization of state based CRDTs,

V . Enes, P. S. Almeida, C. Baquero, and J. Leit ˜ao, “Efficient synchro- nization of state based CRDTs,” inProc. IEEE Int. Conf. on Data Engineering, 2019, pp. 148–159, doi: 10.1109/ICDE.2019.00022

arXiv 2019

Show all 21 references

[9]

ConflictSync: Bandwidth efficient synchronization of divergent state,

C. Baquero, P. S. Gomes, and M. B. Rodrigues, “ConflictSync: Bandwidth efficient synchronization of divergent state,” inProc. Int. Workshop on Principles and Practice of Consistency for Distributed Data, 2025, doi: 10.1145/3806077.3806697

2025
[10]

Practical rateless set reconciliation,

L. Yang, Y . Gilad, and M. Alizadeh, “Practical rateless set reconciliation,” inProc. ACM SIGCOMM, 2024, doi: 10.1145/3651890.3672219

2024
[11]

Distributed Kalman filtering with adaptive communication,

D. Selvi and G. Battistelli, “Distributed Kalman filtering with adaptive communication,”IEEE Control Systems Letters, pp. 15–20, 2025, doi: 10.1109/LCSYS.2025.3550401

2025
[12]

Mechanism design for optimal consensus problems,

D. Bauso, L. Giarr ´e, and R. Pesenti, “Mechanism design for optimal consensus problems,” inProc. IEEE Conf. on Decision and Control, 2006, pp. 3381–3386, doi: 10.1109/CDC.2006.377206

2006
[13]

Distributed consensus resilient to both crash failures and strategic manipulations,

X. Bei, W. Chen, and J. Zhang, “Distributed consensus resilient to both crash failures and strategic manipulations,” arXiv:1203.4324, 2012

2012 arXiv
[14]

An incentive scheme for sensor fusion with strategic sensors,

K. Chen, D. G. Dobhakhshari, V . Gupta, and Y .-F. Huang, “An incentive scheme for sensor fusion with strategic sensors,”IEEE Transactions on Signal Processing, pp. 6342–6351, 2019, doi: 10.1109/TSP.2019.2954974

2019
[15]

Faithful and privacy preserving implementation of average consensus,

K. Teranishi, K. Kogiso, and T. Tanaka, “Faithful and privacy preserving implementation of average consensus,” inProc. American Control Conf., 2025, pp. 2937–2942, doi: 10.23919/ACC63710.2025.11107548

2025
[16]

Incentive compatible recovery from manipulated signals, with applications to decentralized physical infrastructure,

J. Milionis, J. Ernstberger, J. Bonneau, S. D. Kominers, and T. Rough- garden, “Incentive compatible recovery from manipulated signals, with applications to decentralized physical infrastructure,” arXiv:2503.07558, 2025, doi: 10.48550/arXiv.2503.07558

2025 doi arXiv
[17]

A tunable incentive mechanism for binary aggregation without verification,

C.-C. Chen and W. Golab, “A tunable incentive mechanism for binary aggregation without verification,” arXiv:2606.30974, 2026

2026 arXiv
[18]

Decentralized information elicitation without verification,

K. Chen, C. Huang, and J. Huang, “Decentralized information elicitation without verification,”IEEE Transactions on Networking, pp. 4387–4402, 2026, doi: 10.1109/TON.2026.3674784

2026
[19]

Scalable Byzantine fault tolerant state machine replication on heterogeneous servers,

M. Eischer and T. Distler, “Scalable Byzantine fault tolerant state machine replication on heterogeneous servers,”Computing, pp. 97–118, 2018, doi: 10.1007/s00607-018-0652-3

2018 doi
[20]

Tolerating slowdowns in replicated state machines using copilots,

K. Ngo, S. Sen, and W. Lloyd, “Tolerating slowdowns in replicated state machines using copilots,” inProc. USENIX Symposium on Operating Systems Design and Implementation, 2020, pp. 583–598

2020
[21]

Gossip consensus,

D. Cason, N. Miloˇsevi´c, Z. Miloˇsevi´c, and F. Pedone, “Gossip consensus,” inProc. ACM Middleware, 2021, doi: 10.1145/3464298.3493395

2021

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.

[1] [1]

The part time parliament,

L. Lamport, “The part time parliament,”ACM Transactions on Computer Systems, vol. 16, no. 2, pp. 133–169, May 1998, doi: 10.1145/279227.279229

arXiv 1998

[2] [2]

In search of an understandable consensus algorithm,

D. Ongaro and J. Ousterhout, “In search of an understandable consensus algorithm,” inProc. USENIX Annual Technical Conference, 2014, pp. 305–319

2014

[3] [3]

LowPaxos: State machine replication for low resource settings,

A. Mwotil, T. Anderson, B. Kanagwa, T. Stavrinos, and E. Bainomugisha, “LowPaxos: State machine replication for low resource settings,”IEEE Access, pp. 91272–91288, 2024, doi: 10.1109/ACCESS.2024.3421582

arXiv 2024

[4] [4]

Hamava: Fault tolerant reconfigurable geo replication on heterogeneous clusters,

T. Mane, X. Li, M. Sadoghi, and M. Lesani, “Hamava: Fault tolerant reconfigurable geo replication on heterogeneous clusters,” inProc. IEEE Int. Conf. on Data Engineering, 2025, pp. 2024–2037, doi: 10.1109/ICDE65448.2025.00154

arXiv 2025

[5] [5]

Avicenna: Masking slowdowns in replicated state machines with counterfactual evaluation,

C. Hodsdon, Z. Qin, K. Ngo, S. Sen, E. Katz Bassett, and W. Lloyd, “Avicenna: Masking slowdowns in replicated state machines with counterfactual evaluation,” inProc. European Conf. on Computer Systems, 2026, doi: 10.1145/3767295.3803615

arXiv 2026

[6] [6]

Conflict free replicated data types,

M. Shapiro, N. Pregui c ¸a, C. Baquero, and M. Zawirski, “Conflict free replicated data types,” inStabilization, Safety, and Security of Distributed Systems, 2011, pp. 386–400, doi: 10.1007/978-3-642-24550-3.29

work page doi:10.1007/978-3-642-24550-3.29 2011

[7] [7]

Efficient state based CRDTs by delta mutation,

P. S. Almeida, A. Shoker, and C. Baquero, “Efficient state based CRDTs by delta mutation,” inProc. Int. Conf. on Networked Systems, 2014, pp. 62–76, doi: 10.1007/978-3-319-26850-7.5

work page doi:10.1007/978-3-319-26850-7.5 2014

[8] [8]

Efficient synchro- nization of state based CRDTs,

V . Enes, P. S. Almeida, C. Baquero, and J. Leit ˜ao, “Efficient synchro- nization of state based CRDTs,” inProc. IEEE Int. Conf. on Data Engineering, 2019, pp. 148–159, doi: 10.1109/ICDE.2019.00022

arXiv 2019