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

Abstract

April 2026 saw notable methodological convergence in the academic

study of informed trading on decentralized prediction markets.

Three methodological approaches surfaced almost simultaneously,

each addressing the same body of documented cases on Polymarket but

operating at different methodological layers: Mitts and Ofir (2026)

apply a composite statistical screen to over 210,000 wallet–market

pairs and estimate $143 million in aggregate anomalous profit;

Gomez-Cram, Guo, Jensen, and Kung (2026) apply an event-level

sign-randomization test to the platform's complete transaction

history, classifying 3.14% of accounts as "skilled winners" who

drive most price discovery, and separately apply a single-event

lifecycle-and-conviction heuristic that flags 1,950 accounts as

"insiders"; Nechepurenko (2026) develops the Information Leakage

Score (ILS) framework, which quantifies per-market information

front-loading at the article-derived public-event timestamp.

This paper provides a methodological comparison and a sketch of how

the approaches combine. The central organizing claim is that these

are three distinct layers of detection, not three competing methods

on a single layer. Sign-randomization, in particular, 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

in Gomez-Cram et al. (2026) is methodologically separate from their

skill classifier, applies to a population (single-event,

recently-created accounts) that the skill classifier explicitly

excludes by design, and has unknown precision against an external

labelled set. The Polymarket sample on which all three approaches

are evaluated pools politics, sports, crypto, and other categories

with structurally different information technologies, so a

platform-wide "skilled winner" classification is mechanism-ambiguous,

and category-conditioned decompositions are required before the

methodology can be used as a surveillance layer.

The January 2026 U.S.–Venezuela operation cluster, where the U.S.

Department of Justice indictment of Master Sergeant Gannon Van Dyke

provides a rare external enforcement benchmark on at least one

alleged informed trader, illustrates how the layers stack:

account-level lifecycle heuristics identify a small set of

suspicious accounts with face-valid enforcement alignment; legal

investigation addresses whether a specific trader actually possessed

non-public information; per-market scoring would quantify how much

information was leaked into each contract before public observation.

None of the three layers subsumes the others, and a combined

surveillance pipeline gains in precision precisely because each

layer filters a different dimension of the problem.

Keywords: prediction markets, informed trading, market surveillance, sign-randomization, information leakage, methodological complementarity, Polymarket, regulatory technology JEL Classication: D82

JEL Classification: D82, G14, G18, G28, C58