---
title: "UMA (Optimistic Oracle & the DVM)"
url: "https://caper.network/wiki/daos/infrastructure/uma"
updated: 2026-08-31
license: CC-BY-4.0
license_url: "https://creativecommons.org/licenses/by/4.0/"
---

# UMA (Optimistic Oracle & the DVM)

|  |  |
| --- | --- |
| **What it is** | An [optimistic oracle](https://docs.uma.xyz/protocol-overview/how-does-umas-oracle-work) — a protocol that reports arbitrary facts on-chain by assuming a proposed answer is true unless someone disputes it, with a token-holder vote as the court of last resort |
| **Category** | Oracle · dispute resolution · optimistic governance |
| **Token** | UMA (ERC-20) — staked in the **DVM** to vote on disputes and earn emissions |
| **Governance archetype** | The **token-vote-as-oracle**: instead of governing a treasury, UMA holders vote to _adjudicate disputed truth_ — a Schelling-point court of last resort |
| **Built by** | [Risk Labs](https://risklabs.foundation/) (UMA — "Universal Market Access"); Optimistic Oracle live since 2020 |
| **Primary sources** | [uma.xyz](https://uma.xyz/) · [docs.uma.xyz](https://docs.uma.xyz/) |

**UMA** is best known as an oracle, but its structure makes it one of the most instructive governance designs in the industry: the oracle _is_ a DAO. Where a price-feed oracle like [Chainlink](https://chain.link/) pushes data from trusted nodes, UMA answers questions the opposite way — it assumes a proposed answer is correct and only convenes a vote when someone objects. That escalation vote, the [Data Verification Mechanism (DVM)](https://docs.uma.xyz/protocol-overview/how-does-umas-oracle-work), is a token-holder governance process whose sole job is to decide _what is true_. It is the clearest live example of a distinct DAO archetype: not governance _over_ a protocol, but governance _as_ the protocol's source of truth — a decentralised court of last resort.

## The Optimistic Oracle: assert first, dispute if wrong

UMA's [Optimistic Oracle](https://docs.uma.xyz/protocol-overview/how-does-umas-oracle-work) reports data through economic incentives rather than continuous consensus. A **proposer** asserts an answer and posts a refundable bond; anyone may act as a **disputer** and challenge it within a _liveness_ (challenge) window by posting their own bond. If no one disputes before the window closes, the answer finalises and the proposer reclaims their bond plus a reward — hence _optimistic_. Each proposal can be disputed only once, and a dispute is the only thing that triggers a vote. Per UMA's own figures, roughly **99.8% of requests resolve with no vote at all**: the bond-and-challenge game settles almost everything, so the expensive human-judgement layer is invoked only on the genuinely contested tail. This is the same "assert, bond, wait, escalate" pattern behind [Zodiac](/wiki/dao-governance/tooling/frameworks/zodiac)'s [Reality.eth](https://reality.eth.limo/) module — UMA is the alternative oracle underneath it.

## The DVM: governance as a court of last resort

When a proposal _is_ disputed, the question escalates to the **DVM**, a [Schelling-point](/wiki/economics/schelling-points) mechanism where UMA stakers vote for the answer they expect other honest voters to converge on — the one most defensible against on-chain reality. Voting runs a **commit-reveal** cycle documented in UMA's [voter guide](https://docs.uma.xyz/using-uma/voting-walkthrough): a **24-hour commit phase** of encrypted votes, then a **24-hour reveal phase**. Hiding votes until the reveal stops lazy voters from copying the crowd and stops dishonest voters from coordinating on a wrong result. A question resolves once the modal answer carries at least **65% of all staked UMA** (the **SPAT**), and only if at least **5,000,000 UMA** took part at all (the **GAT**). Both are governance parameters, frozen at a round's first reveal, and the 65% figure is recent: it has applied only since [20 April 2025](https://eth.blockscout.com/tx/0xb4435c2aff6e61f73f722d342e417ff2eaff5b2fcba25b5bcd570aed6cd77f35), before which the requirement was **50%**. The gap matters. A month earlier, a disputed Polymarket question [settled Yes on 54.60% of revealed stake](/wiki/economics/schelling-points), a margin the current threshold would not have allowed.

The DVM 2.0 staking design ties the vote to real skin in the game. Staked UMA earns a prorated share of protocol emissions at **0.18 UMA per second**; each vote that lands against the resolved majority — or a vote a staker simply skips — is penalised **0.1% of their staked balance**, and those slashed tokens are redistributed to the voters who got it right. Unstaking triggers a **7-day cooldown** during which tokens neither earn nor vote. That cooldown is the linchpin of UMA's security argument: to corrupt the oracle an attacker must hold enough stake to swing the 65% threshold, but a captured oracle would crater the UMA price over the week it takes to exit — so the _cost of corruption_ is designed to exceed the _profit from corruption_ for any economically significant dispute.

## oSnap and the Optimistic Governor: optimistic execution of DAO votes

UMA productised its oracle for governance through the **Optimistic Governor**, a [Zodiac](/wiki/dao-governance/tooling/frameworks/zodiac) module, and the [oSnap](https://docs.uma.xyz/resources/osnap) front-end built with [Snapshot](/wiki/dao-governance/tooling/voting/snapshot). The idea mirrors Zodiac's Reality Module but swaps in UMA's oracle: a DAO attaches transaction payloads to a Snapshot proposal, and once the off-chain vote passes, anyone can _assert_ on-chain that it passed, backed by a bond. Unless a disputer challenges within the liveness window — escalating to the DVM — the transactions become executable directly from the DAO's [Safe](/wiki/dao-governance/tooling/treasury/safe) treasury, with **no executing multisig in the loop**. It is the same trust-minimising move as UMA's core oracle, applied to a treasury: replace a trusted human signer with a bonded assertion anyone can contest.

_Freshness (2025–26):_ UMA [deprecated oSnap support on 15 December 2025](https://docs.uma.xyz/resources/osnap) — after that date the module can no longer execute transactions from a DAO's Safe treasury. The optimistic-execution _pattern_ persists (Zodiac's Reality Module still offers it via Reality.eth), but UMA's own governance-execution product has wound down, leaving the **Optimistic Oracle and the DVM** as its enduring contribution.

## How Caper approaches this

UMA exists because most on-chain governance has to reach _outside_ the chain — "did this Snapshot vote pass?", "what was this asset's price?" — and someone must assert that answer and be disputable. [Caper](/wiki/foundations/what-is-a-caper) sidesteps that layer for its own governance entirely. A caper's proposals are **typed and self-contained** — `PAYOUT`, `INVEST`, `DIVEST`, `UPGRADE` and `METADATA` — and are executed permissionlessly against the contract's own on-chain state (the `execute_proposal_*` methods on `CaperMain`, in `contracts/logic/src/lib.rs`, are public with no privileged caller). There is no off-chain claim to assert and therefore no optimistic-oracle-and-dispute court to run.

The deeper contrast is in _who_ adjudicates. UMA turns disputed truth into a vote weighted by a separate pool of _staked, tradeable_ UMA. Caper weights governance by `w = (t·v) / (V·T)` (`compute_vote_weight`, `contracts/common/src/lib.rs`), where `v` is a member's holding of a **soulbound, non-transferable proof-of-vote token** — `DIVISIBILITY_MAXIMUM` with metadata describing it as a soulbound stake token that drives exit redemption, its depositor role locked to `global_caller(component)` so it can never be moved between wallets, minted one per ballot cast and 0.01 per XRD of gross trade value (the first two read off the vote-token declaration in `contracts/core/src/caper_dao.rs`). That same weight sets a member's treasury share on `exit()` in `contracts/logic/src/lib.rs`, which reuses the identical helper and then burns the tokens. The decisive governance factor is therefore _earned by the account that transacts_, by voting or by trading, and can never be acquired from another holder — but this is not "bagless": holdings `t` are a multiplicative term, so stake still counts. The honest concession is that UMA answers arbitrary real-world questions Caper does not try to — the comparison is only about governing the DAO itself, where Caper keeps the surface minimal and the influence untransferable.

## References

- UMA, [How does UMA's oracle work?](https://docs.uma.xyz/protocol-overview/how-does-umas-oracle-work) — Optimistic Oracle, the DVM, the 65% resolution threshold, and the cost-of-corruption security model.
- UMA, [Voting walkthrough](https://docs.uma.xyz/using-uma/voting-walkthrough) — commit-reveal timing, the 0.18 UMA/s emission rate, the 0.1% slash, and the 7-day unstaking cooldown.
- UMA, [oSnap](https://docs.uma.xyz/resources/osnap) — the Optimistic Governor / Snapshot execution flow and the 15 December 2025 deprecation notice.
- Related: [Zodiac & the Reality Module](/wiki/dao-governance/tooling/frameworks/zodiac) · [The DAO tooling stack](/wiki/dao-governance/tooling/dao-tooling-overview) · [EigenLayer](/wiki/daos/staking/eigenlayer) (intersubjective token-vote adjudication) · [Caper: on-chain execution](/wiki/governance/execution).