CHI Company Investigation · Finance
Polymarket
Competitive Economics, but Meaningful Governance Asymmetry.
Prediction-market exchange · International venue (polymarket.com) and Polymarket US
Standard company investigation · No score assigned · Research current through 6 October 2026
Polymarket’s pricing is not the problem. The concern is the gap between the control it holds over how markets are written and resolved, the integrity risk of some markets it lists, and the recourse customers have when things go wrong.
On fees, the treatment of winning traders and custody of funds, the evidence mostly runs in Polymarket’s favour. What survived falsification and a final adversarial verification is narrower: who decides contested outcomes and who carries the cost, which markets are listed on a venue without identity checks, and how the product was reportedly marketed.
The economics Mostly favourable
Market-rate fees, free makers, and no limiting of winners.
Polymarket earns from trading volume, not from customer losses. Its current fees sit at or below the main regulated comparator and well below sportsbook costs. Makers pay nothing, geopolitics markets are free, and Polymarket charges no deposit or withdrawal fees of its own.
Governance Control > responsibility
Polymarket writes the rules and holds an override. It does not carry the cost when they fail.
On the international venue it drafts and clarifies market rules and, in its published contract code, can resolve a market manually. It disclaims responsibility for outcomes and compensates disputed results only when it chooses, under no published policy.
This distinction is the finding. The everyday trading bargain is better for customers than the gambling alternatives Polymarket increasingly resembles. The concern sits in the asymmetry between platform control, market-integrity risk and customer recourse.
01Why we looked
Two businesses under one brand.
Users buy shares that pay $1 if an event happens, priced between 0 and $1, and trade them with each other on an order book. Polymarket does not take the other side of trades. Its revenue comes from taker fees, introduced in 2026, plus data licensing.
International Polymarket (polymarket.com) is operated under Panama law by Adventure One QSS Inc. It has no identity verification at sign-up; users keep positions in their own wallets on the Polygon blockchain; outcomes are decided through the UMA oracle. Polymarket US is a CFTC-designated exchange (QCX LLC) with full identity checks, custodial segregated accounts and a filed rulebook. Most findings on this page concern the international venue, and findings about one venue are not applied to the other.
The proposition tested was that Polymarket creates material customer-economic extraction, friction, asymmetry, opacity or risk beyond what is reasonably inherent in running a prediction market. CHI built the case across twelve research streams, ran an independent falsification pass against it, and then put the drafted dossier through a final adversarial verification. Several first-pass claims did not survive; the disposition moved down a step as a result.
One distinction runs through the page. Any exchange must decide outcomes, and regulated venues’ outcome decisions are also final. The question is not whether Polymarket has discretion, but whether the combination of discretion, disclaimed responsibility and thin recourse leaves customers carrying risk the operator controls.
02At a glance
Three findings survived. Each carries its own evidence status.
Resolution control without matching responsibility
Polymarket writes and clarifies the rules, co-controls who may propose outcomes and holds an admin override in its published code. Compensation for disputed outcomes is discretionary: given once, declined another time.
Published codeCompany statementsRead finding 1 ↓ Finding 2 · 2024–present, partly remediatedInsider-prone markets on a venue without identity checks
Markets on military operations, awards and corporate data can be known in advance by small groups. Several insider-trading cases have been charged; none has been adjudicated.
Allegations · charged casesReforms since 2026Read finding 2 ↓ Finding 3 · Dec 2025–May 2026Reported staged-win marketing
The Wall Street Journal reported creator videos showing bets that were never placed, filmed on a replica of the interface. No court or regulator has ruled on it.
Reported conduct · not a legal findingRead finding 3 ↓And one claim that did not survive. Pricing extraction is not supported as the core thesis. Fees moved from zero to market rate in 2026; they did not move to extractive levels. Section 03
03Pricing and fees
Pricing extraction is not supported. Fees ended a subsidy; they did not create extraction.
For more than three years (c. 2022 to January 2026) ordinary trading on the international order book carried no platform fee. Taker fees arrived in stages between 5 January and 30 March 2026, then rose on sports in July. Polymarket US raised its single fee coefficient twice. These are real changes to existing customers’ economics, and CHI codes them in its pattern map.Section 09 They operate at levels at or below the regulated benchmark.S1S3
| Category | Rate | At 50¢ | At 90¢ |
|---|---|---|---|
| Crypto | 0.07 | $3.50 | $0.70 |
| Sports, economics, culture, weather | 0.05 | $2.50 | $0.50 |
| Politics, finance, tech, mentions | 0.04 | $2.00 | $0.40 |
| Geopolitics | 0 | $0 | $0 |
| Polymarket US | |||
| Single contracts (since 17 Sep 2026) | 0.0695 | $3.48 | $0.70 |
How the fee works. Fee = shares × rate × price × (1 − price), charged to takers only. It is highest in dollars per share at 50¢ and highest as a share of stake on longshots. The same curve is used by Kalshi. Worked examples are CHI’s arithmetic from the published schedule.S1
Read the ratio as indicative. These compare entry fees only. They exclude bid-ask spreads (not measured), a second taker fee on an early exit, and differences in how volume and handle are counted.S14
Earlier documentation appears to have understated the effective fee formula; the evidence does not establish that customers were actually charged a newly doubled rate.
Around 1 April 2026 the published fee formula changed in a way first read as a doubling. Polymarket’s own client code already used the corrected formula on 27 February 2026, and January 2026 press reporting quoted the corrected figure. The better reading is a documentation correction: for roughly ten weeks the published tables most likely showed less than the intended fee, and the correction was noted only in a technical changelog line. What was actually charged could not be checked on-chain, so this is recorded as a disclosure-quality issue, not a price action.S1S2Two other pricing facts belong beside the schedule. US customers pay more per trade on politics and sports (about $1.74 per 100 contracts at 50¢, against $1.00–$1.25 internationally) in exchange for regulatory oversight and segregated custody; the reason for the difference is not disclosed. And Kalshi pays interest on idle balances for US users, which Polymarket does not match outside a small holding-reward programme.S3S14
04Finding 1 · Resolution control
Polymarket controls more of the outcome than it takes responsibility for.
Polymarket writes and clarifies the rules that decide who is paid, co-controls who may propose outcomes, and keeps a manual-resolution override in its published code. It disclaims responsibility for outcomes and compensates only at its own discretion.
Some of this is inherent: real-world ambiguity cannot be designed away, and any venue must decide outcomes. What is not inherent is the combination of drafting power, an override, a disclaimer and discretionary compensation with no published policy.Polymarket writes the rules
It writes the rules into each oracle request and can post on-chain “additional context” clarifications; no code limits their content or timing. UMA’s governing standard places “the burden of correct construction on price requesters”, here Polymarket.S4
A manual-resolution path exists
An admin can flag a market and, after a safety period, resolve it manually. That period was 2 days until July 2025 and is 1 hour in the published adapter code; the 2026 oracle stack keeps an admin override its auditor says “bypasses” the normal reporter pipeline.S4
Not established: whether that code version is live, and whether or how often the override has been used.
Outcomes disclaimed; liability capped
Polymarket says it “cannot alter or reverse market resolutions” once UMA finalises them, which is true after finalisation. An earlier Terms of Use capped liability at the lesser of $50 or fees paid, which was $0 for a fee-free user; the 2026 terms reportedly cap it at $100 (weak source). Arbitration sits in Panama.S15
An open oracle process
Whitelisted proposers post outcomes with a bond; anyone can dispute; a second dispute goes to a vote of UMA token holders, with a 65% supermajority since April 2025.S4
Escalation is rare
UMA reports that 99.8% of its requests, across all integrations and not only Polymarket, settle without escalation. Disputes cluster in large, wording-dependent markets.S4
Manipulation was addressed
The 2025 reforms (higher vote threshold, a proposer whitelist, longer review windows for subjective proposals) target manipulation and inaccurate proposals. They do not address drafting ambiguity at source or compensation policy.S4
The compensation record: discretionary, and not consistent
Barron Trump / $DJT market (~$2.1M). Polymarket said “We firmly believe that UMA got this resolution wrong” and announced compensation for YES holders. Amount and completion are not established.S5
Compensation announcedUkraine minerals market (~$7M). Resolved against Polymarket’s own clarification. Polymarket: “because this wasn’t a market failure, we are not able to issue refunds.” A claim that a large UMA holder forced the outcome is an allegation.S5
DeclinedZelenskyy “suit” market (>$200M reported). Initial proposal disputed, final NO. No compensation found.S5
None foundMisconfigured markets. Refunds on a few misconfigured or pulled markets are reported in a single source and carry low weight.
ReportedRule-interpretation cases are different. The January 2026 Venezuela-invasion market and the 2024 Venezuela election market both resolved under Polymarket’s written wording, through the UMA process, with user objections in the first case. They are not compensation cases and followed the written rules. They show that the drafter’s wording decides large markets.S6
Control > responsibility. When drafting is ambiguous, the side the final reading goes against carries the loss. Unless Polymarket chooses to compensate, the drafter carries none of it.
This finding rests on published contract code (deployment unverified) and dated company statements. It does not show that every disputed market is controlled by Polymarket, or that the override has been used against customers.05Finding 2 · Insider-prone markets
Some market categories create unusually strong insider-information risk.
Polymarket has listed markets whose answers small groups can know in advance, including military operations, award decisions and corporate data, on an international venue that has historically operated without identity verification at sign-up. The finding is not that Polymarket encourages insider trading. It is that these categories, on that venue, carry an information asymmetry ordinary traders cannot see.
US soldier, Maduro-operation market. Charged by federal prosecutors and the CFTC over bets reported at ~$33K staked and ~$400K profit. Pleaded not guilty.S7
Allegation · chargedGoogle engineer, Year in Search markets (~$1.2M). Charged by federal prosecutors and the CFTC. A corporate-data market of a kind regulated venues also list.S7
Allegation · chargedIsrael, June 2025 strike bets. Indictment of a reservist and a civilian.S7
Allegation · indictmentPolymarket says its own referrals started both US cases, and reports “100+” law-enforcement referrals.S8
Company statementThird-party analyses of “insider-like” volume and suspicious accounts are analyst interpretations, not findings. No conviction exists in any case above.
Analyst estimateWhat is distinctive
- Military and assassination contracts. Polymarket’s main regulated comparator reportedly does not list them, and a June 2026 CFTC proposal treats war, terrorism and assassination contracts as likely contrary to the public interest.
- No identity checks at sign-up on the international venue, where all three charged cases arose.
- No Polymarket compensation route for insiders’ counterparties; regulators seek restitution in the charged cases.
What is not distinctive, or has changed
- Kalshi lists award markets and has disciplined insiders of its own; corporate-data markets also trade on regulated venues.
- 2026 reforms: integrity rules for both venues (March), Chainalysis surveillance (April), account bans, and removal of the nuclear-detonation and downed-pilot markets. Polymarket US bans staff trading.
- Geopolitics markets are fee-free, so Polymarket earns no direct fee revenue from war markets.
Still open. War markets remained listed as of mid-2026, and no international staff-trading policy was located. Wash-trading estimates (peaking around December 2024) are historic; current prevalence is unknown.S9
06Finding 3 · Reported staged-win marketing
A reported gap between the marketed experience and the actual one.
The Wall Street Journal reported that creator videos promoting Polymarket showed bets that were never placed, filmed on a replica of the trading interface.
No court or regulator has found that Polymarket engaged in fraud or deceptive marketing. Litigation is at complaint stage. No retrieved Polymarket statement disputes the reporting; Polymarket announced a “comprehensive audit of active promotional content”.S10Reporting says Polymarket built the replica site and approved videos, and that clip makers were paid for audiences that were at least 60% US. Which venue the funnel promoted is not established.
NACA v. Blockratize (D.C. Superior Court), naming the CEO and CMO, plus a separate DC class action.S10
AllegationSenators’ letter to the CFTC on prediction-market marketing; a CFTC investigation of Polymarket was reported, scope unclear.
ReportedWhy it matters for CHI. On-chain analyses find that most Polymarket wallets (84–88%) show negative realised profit, which is what zero-sum trading less fees predicts and is not by itself evidence of harm. Marketing that shows wins which never happened widens the gap between the experience sold and the risk users actually carry.
07Counterevidence
The favourable evidence is substantial. It is why the disposition is only weakly supports.
These are not caveats. Several first-pass concerns failed on testing, and the record below is part of the result.
Not the house
Polymarket earns from volume, not from customer losses, and was not found to trade against its customers. A reported plan for an in-house market-making team (Dec 2025) is unconfirmed.
No winner-limiting found
No evidence of limiting, banning or surcharging profitable traders. Sportsbooks limit winners, and Betfair charges large winners a 20–40% premium.S14
Competitive fees
At or below Kalshi’s headline rate; well below sportsbook hold and PredictIt’s 10% of profit plus 5% on withdrawals.S1
Makers and geopolitics pay nothing
Makers pay no fee and share 15–25% of taker fees. Geopolitics markets carry no fee at all.S1
No Polymarket deposit or withdrawal fees
Gas is subsidised for default wallets. Third-party on-ramps such as card processors and bridges charge their own fees.
Users control their funds
Polymarket holds no user keys. Independent exits exist at the contract level: self-custody wallets, an admin-free redemption path and a wallet exit hatch.S11
No single-user freeze found
No function to freeze a specific user’s funds was found in the reviewed contracts. Restricted jurisdictions moved to close-only, so users there can exit.S11
A public ledger
On-chain settlement and open contracts let researchers, journalists and prosecutors detect wash trading, insider patterns and the staged influencer trades.
Referrals and surveillance
Polymarket says its referrals started both US insider prosecutions. Integrity rules and Chainalysis surveillance followed in 2026.S8
Sometimes used for users
It publicly disagreed with UMA in 2024 and announced compensation, and pledged full refunds after a June 2026 third-party front-end attack.S5
Later reforms
Responsible-gambling tools (October 2026), a reported 30-day arbitration opt-out, per-position fee reporting and a shorter taker delay.S12
Polymarket US
CFTC-designated, with identity checks, segregated customer funds and publicly filed fee changes. Federal probes of the international venue closed in 2025 without charges, as reported.S16
Many frictions are compelled or sector-wide. Geoblocking follows foreign regulators; state lawsuits and under-21 allegations target the whole industry; low review-site scores are common across the sector; contested resolutions and insider cases also occur on Kalshi.
08Smaller items
Weaker concerns, held at their weight.
- Global admin levers. Since 2026, Polymarket can upgrade and pause its collateral token, upgrade default wallets and set the fee at match within an adjustable cap. Consumer messaging located does not describe these; no misuse was found.S11
- Fee disclosure. Published fee tables most likely understated fees for about ten weeks in early 2026 and were corrected with a technical note.S1
- Thin international recourse. Panama arbitration, a low liability cap, and chat, Discord and email support with no telephone line. A formal complaints programme was still being built in 2026.S15
- Late responsible-gambling tools. Self-exclusion and deposit limits arrived on 1 October 2026, about 4.5 months after Polymarket US opened generally; Kalshi’s came earlier.S12
- US withdrawal backlog (allegation). The WSJ reported, on anonymous sourcing, a February 2026 card-fraud attack, withdrawal slowdowns and a later leadership exit at Polymarket US. Duration was not quantified; controls were reportedly back to normal by May.S13
- Security incidents. A December 2025 email-login wallet drain was attributed to a third-party provider; reimbursement is not established.
09CHI pattern map
Two patterns supported, one weakly, one not. None of them is the central finding.
CHI tests every company against four recurring patterns of customer-economic change, applied as frozen. Each action counts once.
Charging for something users already had.
SupportedFee-free taker trading on the international order book became metered between 5 January and 30 March 2026, one fact pattern rolled out in stages. Ordinary first monetisation at market rates; confidence medium.S1
Getting more from the same users.
SupportedNarrowSports taker rate 0.03 → 0.05 and maker rebate 25% → 15% (10 July 2026); Polymarket US coefficient 0.05 → 0.0695 in two filings. Against it: a crypto rate cut and volume rebates.S3
Withdrawing older products or architecture.
Weakly supportedThe USDC.e collateral was retired for Polymarket-issued pUSD on 28 April 2026, clearing resting orders. Conversion is 1:1, with no fee, and audited.S11
Passing outside costs on to users.
Not supportedNo fee or benefit change was linked to an external cost. Gas is subsidised, the reverse of pass-through.
Outside the four patterns. The early-2026 fee-documentation correction is not coded as a price action. The page’s central findings, resolution-risk allocation, market selection, marketing conduct and recourse, fall outside SP001–SP004. They are recorded as cross-cutting findings; no new pattern is proposed.
10What remains unresolved
These limit certainty. They do not overturn the findings.
- How often the manual-resolution override has been used, and in whose favour.
- Whether the override-capable contract version is live, including the 1-hour safety period.
- What fees were actually charged during the early-2026 documentation discrepancy.
- Whether the June 2026 refund commitments were completed in full.
- The current international Terms of Use, including the liability cap and forfeiture clauses.
- The Polymarket US Rulebook, complaint process and withdrawal rules.
- Primary regulatory and court texts that were blocked during the research.
- The reported US withdrawal backlog: its duration and effect on customers.
- Open litigation and arbitration, including the marketing suits and a motion to compel arbitration.
- Legacy wallet-control questions, which were not strong enough to carry evidentiary weight.
11Final CHI assessment
Polymarket’s economic bargain with customers is better than the gambling alternatives it increasingly resembles: market-rate fees, free makers, no limiting of winners and self-custody of funds. On the evidence, CHI does not describe its pricing as hostile.
The concern is governance. Polymarket chooses what can be bet on, writes the words that decide who wins, holds an override in its published code, and decides case by case whether to make anyone whole. Some of the markets it lists carry unusual insider risk on a venue without identity checks, and its marketing has been reported, though not found, to show wins that never happened.
Why not a stronger finding? Because the economics and custody came out in Polymarket’s favour, none of the three findings has been adjudicated, and resolution control is partly inherent in any exchange. Why not a weaker one? Because the control rests on published code and dated company statements, and the compensation record is plainly discretionary.
12Methodology and limitations
How this page was built, and how far it can be relied on.
Three layers of testing
Twelve research streams, an independent falsification pass against a written preliminary thesis, and a final adversarial verification of the drafted dossier. Where they differ, the verified version controls. The preliminary disposition was one step higher.
What was checked
Polymarket’s and UMA’s public contract code, Polymarket-published audits, UMA governance documents, dated mirrors of Polymarket’s documentation, and press and filing summaries graded by strength.
How evidence was weighed
Code shows what contracts can do, not how often. Charges and lawsuits are allegations. Press findings are reported conduct. Polymarket did not take part in this research and has not reviewed it.
Limitations
- Polymarket’s own websites, regulators’ and courts’ sites, and most news sites were blocked in the research environment. No primary regulatory, court or congressional text was read verbatim.
- No on-chain state was read, so deployed contract versions, admin key holders, live fee caps and override use are unverified.
- The current international Terms of Use and the Polymarket US Rulebook were not read; clause details rest on an earlier-version copy and an unattributed compilation.
- Documentation was read from third-party mirrors whose fidelity to the live site was assumed, not proven.
- The marketing finding rests on reporting of the WSJ investigation, not on the articles themselves.
- Polymarket has not been asked to respond. Under CHI principles it should have the opportunity to explain.
13Sources
Every claim on this page traces to one of these.
Polymarket fee documentation and changelog, read from a dated third-party mirror (snapshots 26 March and 1 April 2026, to 25 August 2026). Etherdrake/PolymarketDocumentationDocs mirror
Polymarket client fee tests, commit bc3f830 (27 February 2026); The Block’s January 2026 fee figure. Polymarket/clob-client-v2Primary code
Polymarket US fee filings by QCX LLC under CFTC Regulation 40.6 (July and September 2026), via summaries; primary PDFs not retrieved. CFTC filing (Jul) · CFTC filing (Sep)Summary
Resolution architecture: Polymarket’s UMA adapter and exchange contracts, published audits, and UMA governance documents. uma-ctf-adapter · contract-security · UMIP-189Primary code
Custody and collateral: exchange V2 and wallet contracts and their audits. ctf-exchange-v2 · proxy-factoriesPrimary code
Compensation record: Barron Trump / $DJT (June 2024); Ukraine minerals (March 2025); Zelenskyy suit (July 2025). The Block · CryptoSlate · DecryptPress
Rule-interpretation cases: Venezuela invasion market (January 2026); Venezuela election market (August 2024). Yahoo Finance · Bitcoin.comPress
Charged insider cases (allegations). DefenseScoop · TechCrunch · BloombergPress
Integrity reforms and referrals. Bloomberg · Invezz · crypto.newsPress
Influencer campaign: reporting of the WSJ investigation and the NACA suit (June 2026). SBC Americas · FKKS · Bloomberg LawPress
Trust & Safety tools launched 1 October 2026. Casino.orgPress
Polymarket US fraud and withdrawal reporting (WSJ, September 2026; allegation). Crypto TimesPress
Comparators: Kalshi fees; Massachusetts sportsbook limiting; Betfair commission. Kalshi · iGaming Business · Racing PostMixed
Terms of Use: an earlier-version copy of the international terms, and an unattributed compilation describing the 2026 version. Weak sources; the current text was not read.Weak
Regulatory history: 2022 CFTC settled order (no admission or denial); 2025 probe closure as reported; Nevada preliminary injunction (May 2026). SBC Americas · Nevada AGSummary
The complete research dossier, source ledger, timeline, pattern test and verification memo are preserved by CHI and are not reproduced here. Polymarket and related marks belong to their owners; the name is used only to identify the company, and no endorsement or affiliation is implied. This page is research, not legal or financial advice.
Reading this beside the rest of the index
Same question, same method. On Polymarket, the answer sits in governance, not price.
This page carries a disposition and no score. Read it alongside the methodology and the other Finance assessments, including the regulated prediction-market comparator.