OAK — OnChain Attack Knowledge

Worked example · 2024-20

Polymarket Spec-Ambiguity Resolution Dispute Cohort — 2024–2025

Loss
Disputed-market volume varies by market; the Zelenskyy-suit market (~$237M) is the largest, but smaller markets with spec-ambiguity disputes in the single-digit millions cumulatively represent material bettor-side losses.
OAK Techniques observed
OAK-T9.006.002 (Resolution-Spec Ambiguity Exploitation — primary; natural-language market resolution criteria with unmachine-checkable predicates enabled divergent UMA voter interpretation). OAK-T9.006 (Subjective-Oracle Resolution Manipulation — parent technique; the Polymarket/UMA oracle layer was the structural attack surface).
Attribution
unattributed — attribution is to the market-design layer (Polymarket's natural-language resolution specifications) and the oracle-design layer (UMA's voter discretion) rather than to an individual operator. Individual UMA voters acting on spec-ambiguity are voting consistent with one reasonable interpretation of an ambiguous spec.
Key teaching point
T9.006.002 is not limited to the Zelenskyy-suit market. Any Polymarket market whose resolution criterion includes an unmachine-checkable predicate — "credible reporting," "official," "genuine," "agrees to," "confirmed by," "widely recognised as" — has spec-ambiguity as a standing surface. The Zelenskyy-suit market made the structural disequilibrium visible at scale ($237M); smaller markets with the same spec-language pattern are exposed to the same attack class at proportionally smaller volumes. The mitigation is at the market-spec design layer: markets resolving on language are exposed to T9.006.002 by construction; markets resolving on cryptographically-verifiable inputs are not.

Summary

The Zelenskyy-suit market (July 2025, ~$237M) is the canonical T9.006.002 anchor because its scale, press coverage, and the multi-outlet media consensus (Reuters, AP, AFP all reporting Zelenskyy in a suit) vs. UMA resolution (NO, citing "lack of credible reporting consensus") made the spec-ambiguity surface unambiguously visible. But the Zelenskyy-suit market was not the first Polymarket market with an ambiguous resolution spec, nor the first to produce a disputed resolution.

The broader 2024–2025 Polymarket spec-ambiguity cohort includes markets whose resolution criteria contain predicates that are structurally unmachine-checkable:

  • Fashion / wardrobe predicates: "Wearing a suit," "wearing a tie," "wearing a specific garment." These predicates depend on photographic interpretation; there is no canonical authority on what constitutes "wearing a suit" (jacket only? jacket + tie? specific colour?).
  • Political-status predicates: "Agrees to a deal," "announces," "confirms," "officially states." These predicates depend on interpreting political speech — a statement that one reasonable observer considers an announcement may be considered a non-announcement by another.
  • Subjective-adjudication predicates: "Credible reporting consensus," "widely recognised as," "generally accepted that." These predicates are meta-subjective — the resolution spec asks voters to evaluate the subjective assessment of an unspecified cohort of observers.
  • Interpretive-framing predicates: "Genuine," "official," "verified," "legitimate." These predicates embed an evaluative judgment into the resolution criterion.

The common structural feature: each predicate embeds a natural-language interpretation step that cannot be reduced to a machine-checkable on-chain state transition. The UMA DVM is asked to resolve a question that has no deterministic answer. Voters resolve consistent with their interpretation of the spec; bettors whose positions align with the resolution benefit.

The mitigation is at the market-design layer, not at the oracle or voter layer. Markets resolving on cryptographically-verifiable inputs (block-height events, on-chain state, signed oracle attestations with explicit source enumeration) are not exposed to T9.006.002. Markets resolving on natural-language predicates are exposed by construction.

Timeline (UTC)

When Event OAK ref
2020–2024 Polymarket launches and scales; natural-language market resolution specifications are the product default; spec-ambiguity as a standing surface from market-creation time T9.006.002 surface creation
2024–2025 Smaller Polymarket markets with ambiguous specs produce disputed resolutions; bettor-side complaints cite spec-ambiguity as the divergence driver T9.006.002 pre-Zelenskyy cohort
2025-06-24 Zelenskyy photographed in a suit at NATO; Reuters, AP, AFP publish confirming coverage (real-world event)
2025-07-01 UMA voters resolve Zelenskyy-suit market NO citing "lack of credible reporting consensus" despite multi-outlet media consensus T9.006.002 canonical anchor
2025–2026 Post-Zelenskyy market-spec review; Polymarket and other prediction-market platforms face pressure to tighten natural-language resolution specifications (market-design response)

Realised extraction

Varies by market. Zelenskyy-suit at ~$237M is the largest; smaller markets in the single-digit millions cumulatively represent material bettor-side losses. No refund mechanism exists (UMA resolution is binding per Polymarket's terms).

Public references

  • Cross-reference: T9.006.002 (Resolution-Spec Ambiguity Exploitation) at techniques/T9.006.002-resolution-spec-ambiguity-exploitation.md.
  • examples/2025-07-polymarket-zelenskyy-suit.md — Zelenskyy-suit market, July 2025 (canonical T9.006.002 anchor, ~$237M).
  • Decrypt, CoinDesk ("This Isn't Decentralized"), The Defiant — Zelenskyy-suit market coverage.
  • Polymarket market-listing guidelines — resolution-spec language standards (pre- and post-Zelenskyy).
  • UMA DVM documentation — voter discretion in subjective-oracle resolution.

Techniques demonstrated (2)