DV-PMI · Market integrity
Eight market-integrity tests on the 670M-trade Polymarket on-chain panel. Patterns reported are statistical, not legal conclusions. Of the eight indices we publish, 6 are live and four are in development. Three classic FINRA tests (spoofing, layering, quote stuffing) require order-book and cancellation data that the on-chain trade record does not contain; we list them below the live indices rather than approximate them.
Cumulative snapshot through 2026-02-28 · Per-wallet detail in downloads
Each test reports patterns consistent with the named conduct, not legal conclusions about any wallet. No public page names a specific wallet address. Per-wallet detail is available in the downloadable CSVs.
Wallets whose accuracy cannot be explained by price-following alone. Binomial orthogonality test at p<0.01 with Holm-Bonferroni and Benjamini-Hochberg corrections.
How close to market resolution flagged wallets place their trades, versus the non-flagged population. Lower ratios are consistent with pre-resolution information leakage.
Effective spread differential between markets with and without flagged-wallet activity. The textbook Glosten-Milgrom prediction is wider spreads; on Polymarket the pattern is the opposite, consistent with algorithmic market makers crowding into information-rich markets.
Excess accuracy of flagged wallets in markets that resolved opposite to the consensus pre-resolution price. Flagged wallets are positive in every surprise tercile; the population is negative in every tercile.
Self-matched trades, the FINRA Rule 6140 mechanical analog. Polymarket's CLOB does not permit self-matching at scale; Tier 1 is essentially zero. Network-based cluster-wash detection (Sirolly et al, Columbia 2025) supersedes the structural screens for the harder coordinated-washing question and is cited from the page.
Per-market Herfindahl-Hirschman Index of participation. High HHI is the precondition under which one trader can move the consensus price alone, not direct evidence that they did. The index flags structure, not intent. Threshold cutpoints map onto DOJ horizontal-merger guideline categories.
These indices are scheduled for subsequent releases. Aggregator design and data feasibility are confirmed; the bottleneck is implementation time.
Wallet pairs and small clusters that trade primarily with each other and rarely with the broader market. Pending a network-cluster implementation consistent with Sirolly, Ma, Kanoria, and Sethi (Columbia, 2025). The earlier structural-persistence screen has been demoted; its raw output remains downloadable for reproducibility.
Concentration of one-sided trading in the final hours before market resolution moving the settling price. Limited by minute-grain timestamps in on-chain data; intraday precision not possible.
Three classic FINRA-style surveillance tests cannot be run on the on-chain Polymarket panel. We list them explicitly rather than approximate them.
Requires cancelled-order data, which is not on-chain.
Requires order-book depth snapshots, which we do not collect.
Requires sub-second order submission rates, which the on-chain trade record does not contain.
All eight indices report patterns consistent with the named conduct rather than legal conclusions about any wallet's intent. No public page names a specific wallet address. Per-wallet detail is available in downloadable CSVs subject to the usage notice on the methodology page. We benchmark against the leading heuristic approach in the literature (Mitts and Ofir, 2026); the concordance is reported on the PII page.
The four live indices are surfaces of empirical results from Della Vedova (2026), Detecting Informed Trading in Prediction Markets: An Orthogonality Test. Each is mapped onto the four Daubert Rule 702 factors (testability, known error rate, peer-reviewed methodology, general acceptance) in the source paper and on the consulting page.
The dashboard is read-only. It does not produce enforcement recommendations or advisory opinions and is not a substitute for forensic review by a qualified expert. For engagement inquiries, see /expert-witness.