Substituting Proof of Work in Blockchain with Training-Verified Collaborative Model Computation

Mohammad Ishzaz Asif Rafid; Morsalin Sakib

REVIEW 4 major objections 4 minor 30 references

Substituting Proof of Work in Blockchain with Training-Verified Collaborative Model Computation

T0 review · 4 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper claims that blockchain consensus can be secured by measured machine-learning training work, certified by a central server, instead of proof-of-work hash puzzles.

desk verdict The submission is an abstract about a centralized PoW replacement stapled to an unrelated math paper on fractional Laplacians, so the blockchain claims are unsupported and the preprint is incoherent. read the letter →

arxiv 2508.12138 v1 pith:IONTVPPC submitted 2025-08-16 cs.CR cs.AI

Mohammad Ishzaz Asif Rafid , Morsalin Sakib This is my paper

classification cs.CRcs.AI

keywords proofofworkblockchaincollaborativemachinelearningweightedlotterycertificateSHA-256energyefficiencyprivacy-preservingtraining

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

The paper proposes replacing Bitcoin-style proof-of-work with a hybrid system in which miners compete by training segments of machine-learning models. A central server measures each miner's contribution—number of parameters trained and reduction in model loss—and at the end of each training cycle runs a weighted lottery. The winner receives a digitally signed certificate that serves as the proof-of-work substitute and grants the right to append the next block. If the scheme works, blockchain mining energy would be redirected toward useful computation rather than hash puzzles.

What carries the argument

The training-verified certificate: a digitally signed record, issued by the central server, attesting that a miner trained a stated number of model parameters and achieved a stated loss reduction during a collaborative training cycle. The certificate functions like a PoW nonce—it is the token that enters the weighted lottery and authorizes appending a block.

What would settle it

Build a small test chain with the proposed server; have one miner honestly train on the provided dataset and another submit random or copied updates. If the server cannot reliably distinguish the honest contribution by loss reduction and parameter count—or if the dishonest miner can obtain a valid certificate—the substitution fails. Alternatively, check whether a certificate can be verified without trusting the central server; if not, the system has reintroduced a trusted party, which is exactly what proof-of-work avoided.

Watch

Extended reading notes

Core claim

The paper's central claim is that a verifiable certificate of measured machine-learning training effort can substitute for the cryptographic proof-of-work puzzle. The protocol is cyclic: horizontally scaled model segments are trained on preprocessed data; the server scores miners by parameters trained and loss drop; a weighted lottery selects a block appender; digital signatures and SHA-256 keep the chain intact. The authors assert this preserves blockchain integrity while removing proof-of-work's energy waste.

Load-bearing premise

The central server can correctly and irreversibly verify that a miner trained the claimed number of parameters and achieved the reported loss reduction, all while preserving privacy, and that the resulting certificate, lottery, and signature scheme are secure against adversarial miners.

Editorial extensions

If this is right

Miners' computational work becomes reusable: energy spent on training advances an ML model.
The weighted lottery replaces the hash puzzle as the block-selection mechanism.
Blockchain integrity would rest on digital signatures and SHA-256 hashing rather than on solving hash puzzles.
A central server becomes a trusted authority for measuring work, changing the trust model from fully decentralized to hybrid.
The design enables privacy-preserving collaborative training as part of the consensus process.

Reading between the lines

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

A concrete protocol for the server to verify parameter counts and loss reductions without trusting miners is not specified, which leaves the substitution claim open to testing.
Reintroducing a central server means the security guarantee is no longer the trustless property of proof-of-work; a realistic assessment would need to consider what the central server can be bribed or coerced into doing.
A Sybil attack—one miner faking multiple training identities to inflate lottery weight—would be a natural failure mode, and the abstract provides no defense against it.
If the verification gap can be closed, the same signed-certificate structure could be applied to other useful computations, such as scientific simulation or rendering, as proof-of-work replacements.

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

Here's the short version: the manuscript is not a paper about blockchain. The abstract describes a centralized 'training-verified' scheme meant to replace Bitcoin's PoW, but the full text is a mathematics paper, 'Hölder Extension for Fractional Laplacian' by Feng Li, with no connection to the abstract. So there is no technical content in the submission supporting the blockchain claims.

The abstract does touch a real problem—Bitcoin's energy use—and the idea of redirecting mining compute to ML training is a legitimate research direction. But the claimed solution is listed only as assertions: a central server that verifies training by 'number of parameters trained' and 'loss reduction,' then issues a certificate that grants block rights. None of that is defined or defended. There is no mechanism for verifying training without releasing private data, no threat model, no analysis of how an adversarial miner might game the loss metric or the weighted lottery, and no consensus protocol showing how the certificate works with the rest of the chain.

The deeper issue is the manuscript itself. The body has no relation to the abstract. So every security property is not merely under-argued; it's absent. The reader's verdict of UNVERDICTED is right; the stress-test note is also right that no technical objection can be formulated because the content is missing.

Where credit is due: the math paper is presumably a real contribution in its own field. But as a submission to cs.CR, it's a different document. The abstract also doesn't cite prior proof-of-useful-work literature, so even as a position statement it undersells its own context.

The centralization concern the reader flags is real: the trusted server evaluates the work, issues the certificate, and the certificate is the ticket to append a block. That's not proof-of-work; it's proof-of-trust-in-the-server. The paper would need to show how the server doesn't become a single point of failure or control, and it doesn't.

Bottom line: desk reject. Sending this to referees would waste their time. If the authors actually intend the blockchain idea, they need to write the actual paper—specification, threat model, and security argument. If the math paper is what they meant to submit, it's misfiled.

Referee Report

4 major / 4 minor

Summary. The submitted manuscript consists of an abstract claiming a new blockchain consensus mechanism—replacing Proof of Work with a centralized cloud-based collaborative machine-learning training framework—and a full text titled "Hölder Extension for Fractional Laplacian," a mathematics paper on boundary regularity for fractional harmonic functions. The abstract describes miners training horizontal slices of models on private data, a central server measuring parameters trained and loss reduction, a weighted lottery selecting a winner, and a digitally signed certificate that "grants the right to append a block to the blockchain." The full text contains no blockchain content whatsoever: no description of the training protocol, no certificate scheme, no consensus rules, no threat model, no security proof, and no analysis of adversarial miners. The paper's central claim is therefore entirely unsupported by the body of the manuscript.

Significance. If the proposed architecture were developed rigorously, it could address a real concern about PoW energy consumption. However, the manuscript provides no mechanism, no implementation, no formal statement, and no security analysis. The abstract's assertions are not backed by any derivation or experiment. There are no machine-checked proofs, no reproducible code, and no falsifiable predictions in the submitted text. As a research contribution on the claimed topic, the paper has no evident technical content; the mathematical results in the full text, even if correct, are unrelated to the abstract's claims.

major comments (4)

[Abstract and full-text mismatch] The full text is a mathematical paper on the Hölder extension for the fractional Laplacian, with Sections 1–6 and references entirely about potential theory and PDE. It contains no mention of blockchain, proof of work, training, certificates, lottery, or consensus. The central claim of the abstract—that a training-verified certificate can substitute for PoW while preserving blockchain integrity—is thus asserted without any supporting design or analysis.
[Abstract, lines 4–9] The proposed architecture depends on a central server evaluating 'number of parameters trained' and 'reduction in model loss,' yet no mechanism is given for verifying these metrics without access to private data, nor is any error or incentive analysis provided. An adversary could report arbitrary training progress, and the abstract does not explain how the server's evaluation is sound or how the weighted lottery and certificate prevent double-spending or block-withholding attacks. These are load-bearing gaps.
[Abstract, lines 9–11] The claim that the certificate 'grants the right to append a block' is unsupported. There is no consensus protocol describing how other miners validate the certificate, no rule for resolving forks, and no security argument showing that the certificate cannot be forged or reused. The statement that SHA-256 hashing and digital signatures 'preserve blockchain integrity' is a non sequitur without a formal protocol.
[Abstract, centralized server role] The system places trust in a central server that both evaluates training and issues the certificate, which contradicts the decentralization premise of blockchain consensus. No argument is provided to show that this trust assumption is acceptable or that it does not simply reduce the system to a permissioned ledger. This circularity—where the same authority certifies work and authorizes block appends—is not addressed anywhere in the manuscript.

minor comments (4)

[Title] The title 'Substituting Proof of Work in Blockchain with Training-Verified Collaborative Model Computation' does not match the full text, which is titled 'Hölder Extension for Fractional Laplacian.'
[References] The abstract cites blockchain and machine-learning references (e.g., [devries2018bitcoin], [truby2018decarbonizing], [li2017securing], [nakamoto2008bitcoin], [nist2015sha]) that do not appear in the bibliographic list of the full text; the bibliography contains only mathematics references.
[Terminology] Key terms such as 'training-verified certificate,' 'collaborative model computation,' 'weighted lottery,' and 'preprocessed datasets' are not defined anywhere in the manuscript.
[Related work] The manuscript does not compare the proposed approach with existing alternatives to PoW such as proof-of-stake, proof-of-burn, or verifiable delay functions, even at the level of the abstract.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation found; the blockchain security claim is unsupported because the full text is an unrelated mathematics paper.

full rationale

The abstract's central claim is that a central server issues a digitally signed certificate, and this certificate 'serves as a verifiable substitute for PoW and grants the right to append a block.' This is a proposed protocol design, not a first-principles derivation, and no equation or fitted quantity in the submitted text is reused as the output. In particular, the certificate-based right is defined by the protocol; a centralized trust model is not the same as a circular reduction. The provided full text ('Hölder Extension for Fractional Laplacian') contains no blockchain mechanism, threat model, or security proof, so the abstract's claim is unsupported and the preprint is internally inconsistent, but absence of support is a correctness/verifiability problem, not circularity. Citations such as [li2017securing] and [nist2015sha] are not self-citations bearing on a derived result. Consequently, under the hard rule requiring an exhibited reduction (Eq. X = Eq. Y by construction or fitted parameter renamed as prediction), no circular step can be identified, and the circularity score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 1 invented entities

The abstract relies on unproven assumptions about central server trustworthiness, the value of trained models, and the security of a lottery-based consensus. The manuscript body provides no support for these assumptions.

assumptions (3)

domain assumption A central server can accurately evaluate miner contributions using parameter count and loss reduction, in a privacy-preserving way.
Abstract states miners train model segments 'ensuring privacy' and the server evaluates using these two metrics, but no verification mechanism is given.
domain assumption Training models on preprocessed datasets constitutes socially valuable work.
Abstract says the approach 'redirecting energy toward productive computation', assuming the trained models are useful, which is not established.
domain assumption Digital signatures and SHA-256 hashing preserve blockchain integrity in the proposed hybrid.
Abstract mentions cryptographic primitives but provides no argument for how they secure the centralized training-and-lottery protocol.

invented entities (1)

Training-verified certificate
purpose: A digitally signed certificate that substitutes for proof-of-work and grants block-append rights to the lottery winner.
No protocol details, security proof, or implementation are provided. The certificate's unforgeability and verifiability are simply asserted.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Substituting Proof of Work in Blockchain with Training-Verified Collaborative Model Computation." pith.science (2026). https://pith.science/paper/IONTVPPC

@misc{pith2026250812138,
  author       = {Pith},
  title        = {Pith review of: Substituting Proof of Work in Blockchain with Training-Verified Collaborative Model Computation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IONTVPPC}},
  note         = {Machine review of arXiv:2508.12138}
}

read the original abstract

Bitcoin's Proof of Work (PoW) mechanism, while central to achieving decentralized consensus, has long been criticized for excessive energy use and hardware inefficiencies \cite{devries2018bitcoin, truby2018decarbonizing}. This paper introduces a hybrid architecture that replaces Bitcoin's traditional PoW with a centralized, cloud-based collaborative training framework. In this model, miners contribute computing resources to train segments of horizontally scaled machine learning models on preprocessed datasets, ensuring privacy and generating meaningful outputs \cite{li2017securing}. A central server evaluates contributions using two metrics: number of parameters trained and reduction in model loss during each cycle. At the end of every cycle, a weighted lottery selects the winning miner, who receives a digitally signed certificate. This certificate serves as a verifiable substitute for PoW and grants the right to append a block to the blockchain \cite{nakamoto2008bitcoin}. By integrating digital signatures and SHA-256 hashing \cite{nist2015sha}, the system preserves blockchain integrity while redirecting energy toward productive computation. The proposed approach addresses the sustainability concerns of traditional mining by converting resource expenditure into socially valuable work, aligning security incentives with real-world computational progress.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

30 extracted references · 30 canonical work pages

[1]

Motivation and purpose 1

work page
[2]

Weighted Sobolev spaces, Perron solutions, and harmonic measure 3 2.2

(Fractional) Perrons solutions and (fractional) harmonic measures 3 2.1. Weighted Sobolev spaces, Perron solutions, and harmonic measure 3 2.2. Fractional harmonic measure 5 2.3. Weighted p-capacity in RN+1 and Besov capacity in RN 7 2.4. Main results 11

work page
[3]

Minimizers and De Giorgi class 13

work page
[4]

Proof of Theorem 2.4 14

work page
[5]

GHMD s σ⇒ LHMDs σ 17

Proof of Theorem 2.4 continued 16 5.1. GHMD s σ⇒ LHMDs σ 17

work page
[6]

Proof of Theorem 2.5 20 References 23

work page
[7]

H\"older extension for fractional Laplacian

M otiv ation and purpose Let s∈ (0, 1), N≥ 2, and consider the fractional elliptic equation (1.1)  (−∆)su = 0 in Ω u = g in RN\ Ω, in a bounded domain Ω∈ RN, where (−∆)su(x) = cN,sP.V . Z RN u(x)− u(y) |x− y|N+2s dy, where cN,s is a normalization constant. The solution to the Dirichlet boundary value problem in a bounded domain cor- responds with t...

work page Pith review arXiv 2025
[8]

We mainly refer to the results in [15, 19, 2]

M inimizers and De Giorgi class The contents in this section are standard. We mainly refer to the results in [15, 19, 2]. And [19, 2] are concerned with more general metric spaces; however, we can utilize their results directly, since weighted Sobolev spaces equipped with Muckenhoupt Ap-weights ( even with p-admissible weights) in Euclidean setting are in...

work page

Show all 30 references

[9]

Boundary regularity for p-harmonic functions and solutions of the obstacle problem on metric spaces

Bj ¨orn A, Bj ¨orn J. Boundary regularity for p-harmonic functions and solutions of the obstacle problem on metric spaces. Journal of the Mathematical Society of Japan, 2006, 58(4): 1211- 1232

2006
[10]

Nonlinear potential theory on metric spaces

Bj ¨orn A, Bj¨orn J. Nonlinear potential theory on metric spaces. European Mathematical Society; 2011

2011
[11]

Bj ¨orn, J., Boundary estimates and a Wiener criterion for the fractional Laplacian, arXiv:2107.04364

arXiv
[12]

Fat sets and pointwise boundary estimates for p- harmonic functions in metric spaces

Bj ¨orn J, MacManus P, Shanmugalingam N. Fat sets and pointwise boundary estimates for p- harmonic functions in metric spaces. Journal d’Analyse Math´ematique. 2001 Dec;85:339-69

2001
[13]

The boundary Harnack principle for the fractional Laplacian

Bogdan K. The boundary Harnack principle for the fractional Laplacian. Studia Mathematica, 1997, 123(1): 43-80

1997
[14]

The fractional Cheeger problem

Brasco L, Lindgren E, Parini E. The fractional Cheeger problem. Interfaces and Free Bound- aries, 2014, 16(3): 419-458

2014
[15]

Higher H ¨older regularity for the fractional p-Laplacian in the superquadratic case

Brasco, L, Lindgren E, Schikorra A. Higher H ¨older regularity for the fractional p-Laplacian in the superquadratic case. Advances in Mathematics 2018, (338): 782-846

2018
[16]

Regularity estimates for the solution and the free boundary of the obstacle problem for the fractional Laplacian[J]

Ca ffarelli L A, Salsa S, Silvestre L. Regularity estimates for the solution and the free boundary of the obstacle problem for the fractional Laplacian[J]. Inventiones Mathematicae, 2008, 171: 425-461

2008
[17]

An extension problem related to the fractional Laplacian

Ca ffarelli L, Silvestre L. An extension problem related to the fractional Laplacian. Communi- cations in partial differential equations. 2007 Aug 8;32(8):1245-60

2007
[18]

Ca ffarelli, L., Sire, Y . (2018). Bounds on the Green function for integral operators and frac- tional harmonic measure with applications to boundary Harnack. Proceedings of the American Mathematical Society, 146(3), 1207-1216

2018
[19]

A regularity condition at the boundary for solutions of quasilinear elliptic equations

Gariepy R, Ziemer WP. A regularity condition at the boundary for solutions of quasilinear elliptic equations. Archive for Rational Mechanics and Analysis. 1977 Mar;67(1):25-39

1977
[20]

Nonlinear potential theory of degenerate elliptic equa- tions

Heinonen J, Kipel ¨ainen T, Martio O. Nonlinear potential theory of degenerate elliptic equa- tions. Courier Dover Publications; 2018 May 16

2018
[21]

Existence results for fractional p-Laplacian problems via Morse theory

Iannizzotto A, Liu SB, Perera K, Squassina M. Existence results for fractional p-Laplacian problems via Morse theory. Advances in Calculus of Variations 9.2 (2016): 101-125

2016
[22]

Global H ¨older regularity for the fractional p- Laplacian

Iannizzotto A, Mosconi SJ, Squassina M. Global H ¨older regularity for the fractional p- Laplacian. Revista Matem´atica Iberoamericana. 2016 Dec 16;32(4):1353-92

2016
[23]

The Wiener criterion for non-local Dirichlet problems

Kim M, Lee KA, Lee SC. The Wiener criterion for non-local Dirichlet problems. Communica- tions in Mathematical Physics. 2023 Jan 12:1-43

2023
[24]

Regularity of quasi-minimizers on metric spaces

Kinnunen J, Shanmugalingam N. Regularity of quasi-minimizers on metric spaces. manuscripta mathematica. 2001 Jul;105:401-23

2001
[25]

Fractional superharmonic functions and the Perron method for nonlinear integro-differential equations

Korvenp ¨a¨a J, Kuusi T, Palatucci G. Fractional superharmonic functions and the Perron method for nonlinear integro-differential equations. Mathematische Annalen. 2017 Dec;369(3):1443- 89

2017
[26]

Uniformly fat sets

Lewis JL. Uniformly fat sets. Transactions of the American Mathematical Society. 1988;308(1):177-96

1988
[27]

Asymptotic behavior of nonlocal p-Rayleigh quotients

Li, F. Asymptotic behavior of nonlocal p-Rayleigh quotients. ArXiv./abs/1907.08032v3. 2022

1907 arXiv
[28]

Boundary H¨older regularity for elliptic equations

Lian Y , Zhang K, Li D, Hong G. Boundary H¨older regularity for elliptic equations. Journal de Math´ematiques Pures et Appliqu´ees. 2020 Nov 1;143:311-33

2020
[29]

Perron’s Method and Wiener’s Theorem for a non-local Equation

Lindgren E, Lindqvist P. Perron’s Method and Wiener’s Theorem for a non-local Equation. Potential Analysis. 2017 May;46(4):705-37

2017
[30]

G., Regularity at the boundary of solutions of elliptic equations and conformal mapping, Dokl

Maz’ya, V . G., Regularity at the boundary of solutions of elliptic equations and conformal mapping, Dokl. Akad. Nauk SSSR 150, 1297-1300 (1963) (Russian). English transl.: Soviet Math. Dokl. 4, 1547-1551 (1963)

1963

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

Motivation and purpose 1

work page

[2] [2]

Weighted Sobolev spaces, Perron solutions, and harmonic measure 3 2.2

(Fractional) Perrons solutions and (fractional) harmonic measures 3 2.1. Weighted Sobolev spaces, Perron solutions, and harmonic measure 3 2.2. Fractional harmonic measure 5 2.3. Weighted p-capacity in RN+1 and Besov capacity in RN 7 2.4. Main results 11

work page

[3] [3]

Minimizers and De Giorgi class 13

work page

[4] [4]

Proof of Theorem 2.4 14

work page

[5] [5]

GHMD s σ⇒ LHMDs σ 17

Proof of Theorem 2.4 continued 16 5.1. GHMD s σ⇒ LHMDs σ 17

work page

[6] [6]

Proof of Theorem 2.5 20 References 23

work page

[7] [7]

H\"older extension for fractional Laplacian

M otiv ation and purpose Let s∈ (0, 1), N≥ 2, and consider the fractional elliptic equation (1.1)  (−∆)su = 0 in Ω u = g in RN\ Ω, in a bounded domain Ω∈ RN, where (−∆)su(x) = cN,sP.V . Z RN u(x)− u(y) |x− y|N+2s dy, where cN,s is a normalization constant. The solution to the Dirichlet boundary value problem in a bounded domain cor- responds with t...

work page Pith review arXiv 2025

[8] [8]

We mainly refer to the results in [15, 19, 2]

M inimizers and De Giorgi class The contents in this section are standard. We mainly refer to the results in [15, 19, 2]. And [19, 2] are concerned with more general metric spaces; however, we can utilize their results directly, since weighted Sobolev spaces equipped with Muckenhoupt Ap-weights ( even with p-admissible weights) in Euclidean setting are in...

work page