arxiv: 2605.02287 · v2 · submitted 2026-05-04 · 💱 q-fin.TR · cs.CY· q-fin.GN

Recognition: no theorem link

Per-Market Information Leakage and Order-Flow Skill: Two Methodological Lenses on Informed Trading in Decentralized Prediction Markets

Maksym Nechepurenko

Authors on Pith no claims yet

Pith reviewed 2026-05-15 07:30 UTC · model grok-4.3

classification 💱 q-fin.TR cs.CYq-fin.GN

keywords informed tradingdecentralized prediction marketsinformation leakagesign-randomizationorder flow skilllifecycle heuristicmethodological comparisonPolymarket

0 comments

The pith

Three methodological approaches to informed trading in decentralized prediction markets function as distinct detection layers rather than interchangeable alternatives.

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

This paper compares three recent methods for spotting informed trading on platforms like Polymarket. It establishes that sign-randomization tests detect persistent directional skill at the account level, lifecycle heuristics flag potential insiders separately, and information leakage scores quantify front-loading per market. A reader should care because treating these as interchangeable risks misclassifying traders and overlooking how different market categories rely on distinct information sources. The analysis uses the 2026 U.S.-Venezuela enforcement case to show how the layers can stack for better precision in a combined approach.

Core claim

The paper claims that Mitts and Ofir's composite screen, Gomez-Cram et al.'s sign-randomization test classifying 3.14% as skilled winners and 1,950 as insiders, and the Information Leakage Score framework represent three separate layers of detection. Sign-randomization acts as an account-level test of persistent directional skill conditional on opportunity selection, not a direct insider trading test or per-market measure. The heuristic insider flag covers a population the skill classifier excludes and carries unknown precision. Pooling across politics, sports, and crypto categories makes platform-wide classifications mechanism-ambiguous. The January 2026 enforcement benchmark demonstrates a

What carries the argument

The layered detection pipeline separating account-level skill testing via sign-randomization, insider flagging via lifecycle heuristics, and per-market information front-loading via the Information Leakage Score.

If this is right

A combined use of the three methods improves precision by addressing different dimensions of informed trading.
Sign-randomization should not be interpreted as a direct measure of insider trading.
Per-market leakage scores can specify how much information enters individual contracts.
Market category differences require caution when generalizing skilled winner classifications across the platform.
The 2026 enforcement event provides a way to externally validate the distinct layers.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If validated, this layered approach could be applied to refine detection in other blockchain-based trading environments.
Researchers might develop hybrid models that sequence the layers to first filter accounts then score specific markets.
Regulatory bodies could use the separation to prioritize investigations based on which layer flags an account.

Load-bearing premise

The three detection methods capture independent dimensions of informed trading with little overlap or confounding across the diverse categories in the Polymarket data.

What would settle it

A high degree of overlap between the accounts identified by sign-randomization, the insider heuristic, and high information leakage scores in the same dataset would challenge the claim that they represent distinct layers.

read the original abstract

April 2026 saw notable methodological convergence in the academic study of informed trading on decentralized prediction markets. Three approaches surfaced almost simultaneously: Mitts and Ofir (2026) apply a composite screen to over 210,000 wallet-market pairs; Gomez-Cram et al. (2026) apply an event-level sign-randomization test to Polymarket's complete transaction history, classifying 3.14% of accounts as "skilled winners" and separately flagging 1,950 accounts as "insiders" via a lifecycle heuristic; Nechepurenko (2026) develops the Information Leakage Score (ILS) framework, which quantifies per-market information front-loading at an article-derived public-event timestamp. This paper provides a methodological comparison. The central claim is that these are three distinct layers of detection, not competing methods on a single layer. Sign-randomization is best understood as an account-level test of persistent directional skill conditional on opportunity selection -- not a direct test of insider trading, and not a per-market measure. The heuristic insider flag is separate from the skill classifier, applies to a population the classifier excludes by design, and has unknown precision. The Polymarket sample pools politics, sports, crypto, and other categories with different information technologies, so a platform-wide "skilled winner" classification is mechanism-ambiguous. The January 2026 U.S.-Venezuela operation cluster, where the DOJ indictment of Master Sergeant Gannon Van Dyke provides a rare external enforcement benchmark, illustrates how the layers stack: lifecycle heuristics identify suspicious accounts; legal investigation addresses non-public-information possession; per-market scoring would quantify how much information was leaked into each contract. A combined pipeline gains in precision because each layer filters a different dimension.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This paper frames three recent detection methods as complementary layers rather than rivals, with the new ILS adding a per-market leakage measure, but the independence claim rests on conceptual distinction without overlap tests.

read the letter

The core takeaway is that the three approaches—sign-randomization for account skill, lifecycle heuristics for insider flags, and the author's ILS for per-market front-loading—can be stacked instead of treated as alternatives. The paper uses the January 2026 Van Dyke enforcement event to show how they might filter different dimensions in the Polymarket data. That framing is the main contribution, and it is useful for anyone trying to build a detection pipeline rather than pick one tool. The synthesis of the three 2026 papers is clean and the per-market angle on ILS is a reasonable extension of the author's earlier definition. The writing keeps the distinctions clear without overclaiming empirical wins. The soft spot is the lack of any quantitative check on overlap or confounding. The sample pools politics, sports, and crypto markets that have different information flows, yet the paper reports no correlations between the sign-randomization flags and ILS scores, no precision numbers on the 1,950 insider accounts, and no robustness split by category. The external event is invoked as validation but without conditional probabilities or false-positive rates that would show the layers actually separate distinct information. This leaves the central claim plausible on paper but untested in the data. Readers working on prediction-market microstructure or surveillance tools will find the organizing framework worth reading, even if they have to supply their own overlap metrics. It is not a finished empirical result, but the methodological point is sharp enough to merit referee time with requests for those checks.

Referee Report

2 major / 2 minor

Summary. The paper claims that three recent approaches to detecting informed trading on decentralized prediction markets—Mitts and Ofir (2026) composite screening, Gomez-Cram et al. (2026) sign-randomization plus lifecycle heuristics, and the author's own Information Leakage Score (ILS)—constitute distinct, non-competing layers rather than alternative methods on a single dimension. Sign-randomization is characterized as an account-level test of persistent directional skill conditional on opportunity selection; the lifecycle heuristic flags a separate population excluded by the skill classifier; and ILS provides per-market quantification of information front-loading. The January 2026 Van Dyke enforcement event is invoked to illustrate how the layers can be stacked for improved precision in a pooled Polymarket sample spanning politics, sports, and crypto categories.

Significance. If the claimed orthogonality holds, the work would supply a useful organizing framework for combining detection methods and clarifying mechanism-ambiguous platform-wide classifications. The conceptual distinctions and the external enforcement benchmark are potentially valuable for future empirical pipelines, but the absence of quantitative overlap tests or precision metrics in the current manuscript limits immediate applicability.

major comments (2)

[Abstract and §4] Abstract and §4 (layer-stacking illustration): the central claim that the three methods measure independent dimensions with minimal overlap is asserted without any reported correlation, contingency table, or precision-gain statistic between sign-randomization flags, the 1,950 lifecycle insider flags, and ILS scores in the pooled sample; the January 2026 Van Dyke event is described qualitatively but supplies no conditional probabilities or false-positive rates demonstrating separation of layers.
[Abstract] Abstract: the distinctness of the per-market ILS layer rests on the definition introduced in Nechepurenko (2026); because the present manuscript supplies no independent validation or robustness check against category-specific information technologies, the non-competition claim is circular with respect to the author's prior work.

minor comments (2)

[Abstract] Abstract: the timeline reference to methodological convergence in 'April 2026' followed by discussion of the 'January 2026' enforcement event should be clarified to avoid reader confusion about chronology.
[Methods] The manuscript pools markets with heterogeneous information structures (politics/sports/crypto) yet does not discuss whether ILS thresholds or sign-randomization windows are calibrated separately by category; a brief note on this choice would improve transparency.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for these constructive comments, which correctly identify the need for stronger quantitative support of the layer-stacking claim. We address each point below and commit to revisions that add the requested metrics where data permit.

read point-by-point responses

Referee: [Abstract and §4] Abstract and §4 (layer-stacking illustration): the central claim that the three methods measure independent dimensions with minimal overlap is asserted without any reported correlation, contingency table, or precision-gain statistic between sign-randomization flags, the 1,950 lifecycle insider flags, and ILS scores in the pooled sample; the January 2026 Van Dyke event is described qualitatively but supplies no conditional probabilities or false-positive rates demonstrating separation of layers.

Authors: We agree that the current manuscript presents the orthogonality claim primarily on conceptual grounds. In the revised version we will add (i) pairwise correlations and a three-way contingency table between sign-randomization flags, lifecycle-insider flags, and binned ILS scores in the pooled sample, and (ii) any available account-level timing statistics from the public Van Dyke enforcement record. Because the Van Dyke episode is a single external benchmark rather than a statistical sample, we cannot compute reliable false-positive rates or conditional probabilities; the event will remain illustrative. revision: yes
Referee: [Abstract] Abstract: the distinctness of the per-market ILS layer rests on the definition introduced in Nechepurenko (2026); because the present manuscript supplies no independent validation or robustness check against category-specific information technologies, the non-competition claim is circular with respect to the author's prior work.

Authors: The manuscript is a comparative positioning paper that takes the ILS definition and its category-specific validations as established in Nechepurenko (2026). Its contribution is to show that ILS operates on a different dimension (per-market leakage) from the account-level skill test and the lifecycle heuristic. To address the circularity concern we will insert a short recap of the prior robustness checks across information-technology regimes and explicitly note that the pooled sample’s category heterogeneity is the reason a per-market measure is complementary rather than redundant. revision: partial

standing simulated objections not resolved

False-positive rates or conditional probabilities for layer separation cannot be estimated from the single Van Dyke enforcement event without a larger set of ground-truth cases.

Circularity Check

1 steps flagged

Self-citation load-bearing in claiming distinct detection layers via prior ILS definition

specific steps

self citation load bearing [Abstract]
"Nechepurenko (2026) develops the Information Leakage Score (ILS) framework, which quantifies per-market information front-loading at an article-derived public-event timestamp. This paper provides a methodological comparison. The central claim is that these are three distinct layers of detection, not competing methods on a single layer."

The assertion of ILS as an independent per-market layer (distinct from account-level sign-randomization and lifecycle heuristics) is justified solely by citation to the author's prior work; no new quantitative evidence of minimal overlap or confounding is supplied in this manuscript, rendering the separation self-referential by construction.

full rationale

The paper's central claim that sign-randomization, lifecycle heuristics, and per-market ILS constitute three orthogonal layers rests on the author's own 2026 definition of ILS without new overlap metrics, correlation checks, or external validation of independence in the pooled sample. This reduces the distinctness assertion to a self-referential framing rather than an independently derived result. The January 2026 enforcement benchmark is invoked illustratively but does not supply the required conditional probabilities or false-positive rates to demonstrate non-confounding. No other circular patterns (self-definitional equations, fitted inputs renamed as predictions, or ansatz smuggling) appear in the provided text.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim depends on the untested assumption that the three methods capture independent dimensions and that the cited external enforcement event cleanly validates the pipeline; no quantitative measures of overlap or precision are supplied in the abstract.

axioms (1)

domain assumption The three methodological approaches measure independent dimensions of informed trading
This underpins the claim that they are distinct layers rather than overlapping or competing.

invented entities (1)

Information Leakage Score (ILS) no independent evidence
purpose: Quantifies per-market information front-loading at article-derived public-event timestamps
New score developed to provide a per-market view of information leakage in prediction markets.

pith-pipeline@v0.9.0 · 5630 in / 1435 out tokens · 57423 ms · 2026-05-15T07:30:07.274247+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Manipulation, Insider Information, and Regulation in Leveraged Event-Linked Markets
q-fin.TR 2026-05 unverdicted novelty 7.0

Leverage scales market-price manipulation linearly while shifting outcome-manipulation thresholds and multiplying informed-trading rents in three distinct ways, calling for re-allocated regulatory attack surfaces rath...
A Taxonomy of Event-Linked Perpetual Futures: Variant Designs Beyond the Single-Market Binary Case
q-fin.TR 2026-05 unverdicted novelty 6.0

The paper organizes seven canonical variants of event-linked perpetual futures along four design axes, supplying payoff definitions, inheritance rules from prior work, and variant-specific constraints.

Reference graph

Works this paper leans on

41 extracted references · 41 canonical work pages · cited by 2 Pith papers

[1]

Journal of Economic Perspectives , volume =

Wolfers, Justin and Zitzewitz, Eric , title =. Journal of Economic Perspectives , volume =

work page
[2]

and Forsythe, Robert and Gorham, Michael and Hahn, Robert and Hanson, Robin and Ledyard, John O

Arrow, Kenneth J. and Forsythe, Robert and Gorham, Michael and Hahn, Robert and Hanson, Robin and Ledyard, John O. and Levmore, Saul and Litan, Robert and Milgrom, Paul and Nelson, Forrest D. and Neumann, George R. and Ottaviani, Marco and Schelling, Thomas C. and Shiller, Robert J. and Smith, Vernon L. and Snowberg, Erik and Sunstein, Cass R. and Tetlock...

work page
[3]

Surowiecki, James , title =

work page
[4]

Information Systems Frontiers , volume =

Hanson, Robin , title =. Information Systems Frontiers , volume =

work page
[5]

The Journal of Prediction Markets , volume =

Hanson, Robin , title =. The Journal of Prediction Markets , volume =

work page
[6]

and Servan-Schreiber, Emile and Tetlock, Philip and Ungar, Lyle and Mellers, Barbara , title =

Atanasov, Pavel and Rescober, Phillip and Stone, Eric and Swift, Samuel A. and Servan-Schreiber, Emile and Tetlock, Philip and Ungar, Lyle and Mellers, Barbara , title =. Management Science , volume =

work page
[7]

and Stiglitz, Joseph E

Grossman, Sanford J. and Stiglitz, Joseph E. , title =. American Economic Review , volume =

work page
[8]

Trading volume statistics , year =

work page
[9]

, title =

Kyle, Albert S. , title =. Econometrica , volume =

work page
[10]

and Milgrom, Paul R

Glosten, Lawrence R. and Milgrom, Paul R. , title =. Journal of Financial Economics , volume =

work page
[11]

and O'Hara, Maureen and Paperman, Joseph B

Easley, David and Kiefer, Nicholas M. and O'Hara, Maureen and Paperman, Joseph B. , title =. Journal of Finance , volume =

work page
[12]

and O'Hara, Maureen , title =

Easley, David and Kiefer, Nicholas M. and O'Hara, Maureen , title =. Review of Financial Studies , volume =

work page
[13]

Journal of Finance , volume =

Easley, David and Hvidkjaer, Soeren and O'Hara, Maureen , title =. Journal of Finance , volume =

work page
[14]

Journal of Financial Economics , volume =

Duarte, Jefferson and Young, Lance , title =. Journal of Financial Economics , volume =

work page
[15]

Flow Toxicity and Liquidity in a High-Frequency World , journal =

Easley, David and L. Flow Toxicity and Liquidity in a High-Frequency World , journal =

work page
[16]

and Bondarenko, Oleg , title =

Andersen, Torben G. and Bondarenko, Oleg , title =. Journal of Financial Markets , volume =

work page
[17]

and Ready, Mark J

Lee, Charles M.\ C. and Ready, Mark J. , title =. Journal of Finance , volume =

work page
[18]

, title =

Hawkes, Alan G. , title =. Biometrika , volume =

work page
[19]

Quarterly Journal of Economics , volume =

Budish, Eric and Cramton, Peter and Shim, John , title =. Quarterly Journal of Economics , volume =

work page
[20]

, title =

Platt, John C. , title =. Advances in Large Margin Classifiers , publisher =. 1999 , pages =

work page 1999
[21]

Proceedings of the 22nd International Conference on Machine Learning (ICML) , pages =

Niculescu-Mizil, Alexandru and Caruana, Rich , title =. Proceedings of the 22nd International Conference on Machine Learning (ICML) , pages =

work page
[22]

arXiv preprint arXiv:2508.03474 , year =

Saguillo, Oriol and Ghafouri, Vahid and Kiffer, Lucianna and Suarez-Tangil, Guillermo , title =. arXiv preprint arXiv:2508.03474 , year =

work page arXiv
[23]

2026 , howpublished =

Mitts, Joshua and Ofir, Moran , title =. 2026 , howpublished =

work page 2026
[24]

2020 IEEE Symposium on Security and Privacy (S&P) , pages =

Daian, Philip and Goldfeder, Steven and Kell, Tyler and Li, Yunqi and Zhao, Xueyuan and Bentov, Iddo and Breidenbach, Lorenz and Juels, Ari , title =. 2020 IEEE Symposium on Security and Privacy (S&P) , pages =

work page 2020
[25]

and Tetlock, Philip E

Schoenegger, Philipp and Tuminello, Peter S. and Tetlock, Philip E. and Karger, Ezra and Atanasov, Pavel , title =. Science Advances , volume =. 2024 , doi =

work page 2024
[26]

, title =

Leetaru, Kalev and Schrodt, Philip A. , title =. Annual Convention of the International Studies Association , year =

work page
[27]

2026 , month = apr, note =

Nechepurenko, Maksym , title =. 2026 , month = apr, note =

work page 2026
[28]

2026 , month = may, note =

Nechepurenko, Maksym and Shuvalov, Pavel , title =. 2026 , month = may, note =

work page 2026
[29]

Soros, George , title =

work page
[30]

, title =

Schelling, Thomas C. , title =

work page
[31]

, title =

Manski, Charles F. , title =. Economics Letters , volume =

work page
[32]

2026 , note =

Tsang, Kwok Ping and Yang, Zichao , title =. 2026 , note =

work page 2026
[33]

and Peng, L

Ng, H. and Peng, L. and Tao, Y. and Zhou, D. , title =. 2026 , note =

work page 2026
[34]

and Huang, Taeyoung , title =

Clinton, Joshua D. and Huang, Taeyoung , title =. 2025 , note =

work page 2025
[35]

, title =

Murphy, Allan H. , title =. Journal of Applied Meteorology , volume =

work page
[36]

, title =

Brier, Glenn W. , title =. Monthly Weather Review , volume =

work page
[37]

, title =

Gneiting, Tilmann and Raftery, Adrian E. , title =. Journal of the American Statistical Association , volume =

work page
[38]

2026 , month = may, note =

Nechepurenko, Maksym , title =. 2026 , month = may, note =

work page 2026
[39]

2026 , month = apr, note =

Gomez-Cram, Roberto and Guo, Yunhan and Jensen, Theis Ingerslev and Kung, Howard , title =. 2026 , month = apr, note =

work page 2026
[40]

Soldier Charged With Using Classified Information To Profit From Prediction Market Bets , howpublished =

U.S. Soldier Charged With Using Classified Information To Profit From Prediction Market Bets , howpublished =. 2026 , month = apr, day =

work page 2026
[41]

2026 , month = apr, day =

work page 2026