Worked example · 2025-03

Polymarket UMA governance attack — Ukraine mineral deal market — 2025-03

Loss: ~$7M in market volume resolved against the underlying real-world outcome. No funds drained; aggrieved YES-side bettors lost their stakes to a coordinated UMA-tokenholder vote that overrode the market's stated resolution criteria.
OAK Techniques observed: OAK-T9.006.001 — DVM Vote Capture by Economically-Interested Holder. The canonical worked-example anchor for the sub-Technique. Parent Technique: OAK-T9.006 (Subjective-Oracle Resolution Manipulation).
Attribution: confirmed (Polymarket itself publicly described this as an "unprecedented governance attack"); inferred-strong at the operator level — a single UMA whale cast ~5M UMA tokens via 3 wallets representing ~25% of the resolution vote.
Key teaching point: The Ukraine mineral deal market (March 2025) is the canonical first case of UMA DVM vote capture — the earliest documented instance where a Polymarket market resolved against its stated rules through coordinated UMA tokenholder voting. A single whale cast ~5M UMA tokens (~25% of the resolution vote) to drive a YES outcome on a market where no underlying real-world event had occurred. The structural disequilibrium — UMA's market cap (~$95M) smaller than a single market's adjudicated volume — means the economic cost of corrupting the oracle is a fraction of the prize. This is the foundational anchor for T9.006.001 and the first time Polymarket itself publicly acknowledged its oracle layer had been corrupted.

Timeline (UTC unless noted)

When	Event	OAK ref
2025-03 (market created)	Polymarket lists "Ukraine agrees to Trump mineral deal before April?" market; resolution rules require "official information from the US and Ukrainian governments" confirming a deal	T9.006.001 (standing DVM vote-capture surface)
2025-03-24	YES probability surges from ~9% to 100% during the resolution window; investigation later reveals a single UMA whale cast ~5M UMA tokens across 3 wallets (~25% of total resolution vote)	T9.006.001 (DVM vote capture by economically-interested holder)
2025-03-24/25	Market resolves YES despite no mineral deal being announced or signed — resolution directly contradicts the market's stated rules	T9.006.001 (oracle-resolution divergence from real-world outcome)
2025-03-25 onward	Polymarket publicly characterizes the resolution as "incorrect" and describes the event as an "unprecedented governance attack"; citing UMA's decentralized structure, refuses refunds	(platform admission — oracle layer corruption acknowledged)
2025-03 onward	Industry coverage (CoinDesk, The Block) documents the vote-capture mechanism and the structural disequilibrium (UMA market cap ~$95M < disputed market volume)	(public documentation — structural risk characterized)
Continuing	First documented case of a Polymarket market resolved against its stated rules through coordinated UMA tokenholder voting; canonical anchor for T9.006.001	(canonical reference — DVM vote capture)

Summary

The Polymarket market "Ukraine agrees to Trump mineral deal before April?" had stated resolution rules requiring "official information from the US and Ukrainian governments" confirming a deal. As of March 25, 2025, no such deal had been announced or signed. Despite this, the market resolved YES on March 24-25 after the YES probability surged from ~9% to 100% in the resolution window.

Investigation by Polymarket and CoinDesk identified a single UMA whale who cast approximately 5 million UMA tokens across 3 wallets (~25% of the total resolution vote), driving the YES outcome. Polymarket publicly conceded that the resolution was "incorrect" and described the event as an "unprecedented governance attack" — but, citing the decentralised structure of UMA's optimistic oracle, refused refunds.

This is the first documented case of a Polymarket market resolved against its stated rules through coordinated UMA tokenholder voting. It is also the first time Polymarket itself publicly acknowledged that its oracle layer had been corrupted.

Why this is structurally novel

Spot-AMM oracle manipulation (Mango Markets, BonqDAO, Vee Finance — all canonical OAK-T9.001 cases) is a liquidity attack: the attacker borrows or accumulates capital, moves a thin pool, and books the price reading. Mitigations are mechanical — TWAP, multi-venue feeds, deviation halts (OAK-M09).

Polymarket-class subjective-oracle manipulation has no mechanical mitigation. The "oracle" is a token-weighted vote over a natural-language claim. Four properties make it qualitatively different:

The oracle adjudicates language, not numbers. "Agrees to a deal" has no TWAP. Spec-ambiguity is the attack surface; the spec itself is the vulnerability.
Oracle market cap < adjudicated volume. UMA's market cap was ~$95M during this period; subsequent disputed markets (e.g., the Zelenskyy-suit market at $237M) had volume larger than the oracle's market cap. The economic cost of corrupting the vote is a fraction of the prize. This is a stable disequilibrium that gets worse as Polymarket scales.
Voter–bettor identity overlap. UMA tokenholders are not a separate pool from market participants; nothing prevents a whale from voting on a market they hold a position in. Classical oracle defenses assume voter neutrality, which is structurally absent here.
The resolution source itself is attackable off-chain. When a market resolves on a real-world claim ("a deal was announced"), the resolution input — news headlines, official statements — can be manipulated upstream of the oracle.

OAK's existing T9.001 (price-feed oracle manipulation) does not cover this attack class. v0.x should add T9.006 as a first-class Technique with sub-patterns .001 (DVM vote capture), .002 (resolution-spec ambiguity exploitation), .003 (off-chain resolution-source coercion), .004 (platform-override of oracle outcome).

What defenders observed

Pre-event: the market's resolution criteria were ambiguous in a way that voters could exploit ("official information" — published where? authoritative as of when?). Spec-tightening would have raised the cost of dispute but not eliminated the vote-capture vector.
At-event: the YES surge from 9% to 100% in the resolution window is observable on-chain in the Polymarket CLOB. Voter–position correlation analysis (the 3 wallets that cast 5M UMA also held large YES positions) was performed retrospectively by Polymarket and CoinDesk.
Post-event: Polymarket's public characterisation as a "governance attack" is the strongest available admission that the oracle layer is not mechanically defensible against economically-rational voters. The platform's refusal to refund — citing UMA's decentralised structure — exposes the trust gap that subsequent markets (Zelenskyy suit, US-Iran ceasefire, UFO declassification) only widened.

What this example tells contributors

The Ukraine mineral deal market is the canonical first of a now-documented cohort of Polymarket-UMA resolution-manipulation cases. Subsequent cases (Zelenskyy suit July 2025, UFO declassification December 2025, US-Iran ceasefire April 2026, Iran-strike journalist coercion March 2026) belong to the same Technique class.

Contributors writing the v0.x T9.006 entry should anchor on this incident as the structural-novelty proof and use the Zelenskyy and Iran cases as cohort evidence. The journalist-coercion case (March 2026) is the most unsettling sub-pattern (.003) because it extends the attack vector into physical coercion of off-chain humans.