By: Igor (@justigor)

This op-ed was originally published in the predictionindex.xyz inaugural Annual Report '25-26

In November 2025, Monad launched its mainnet. Within 24 hours, a $53 million prediction market on Polymarket asked a simple numerical question: did Monad's fully diluted valuation exceed $4 billion on the day after launch?

The token price cleared the threshold on Upbit, the exchange with twice the trading volume of Coinbase during the relevant window. CoinGecko's average FDV showed it had exceeded $4 billion. Coinbase's own API candlestick data showed prices above the threshold. Three rounds of UMA arbitration later, with the resolution source changed retroactively and the definition of "most liquid price source" vague enough to be weaponised, the market resolved "No."

This is not an isolated incident. Resolution disputes have become a recurring pattern across prediction markets' highest-profile and highest-stakes moments. The underlying questions are often numerical and verifiable. The data exists. Yet subjective interpretation is baked into the resolution layer regardless, and the mechanism cannot converge on what is observable.

The contradiction at the centre

This report makes a strong case that prediction markets are becoming information infrastructure. Markets that once behaved like episodic wagers now accumulate history, absorb information in real time, and produce prices that increasingly function as shared models of reality. We agree with this thesis. But it contains a quiet assumption: that the information system can settle its own claims credibly.

State confirmation, the process of agreeing on what actually happened, is how any system converts an open question into a settled fact. In permissioned systems, institutions handle this work: courts, exchanges, regulators. Participants grant certain rights in exchange for efficiency, and the mechanism works because of laws, history, and reputation. For regulated, permissioned venues, this is a reasonable design.

But for markets that claim to be permissionless, this architecture introduces a contradiction. The trade surface is decentralised. The settlement layer is not. The system inherits a single point of institutional dependency at the moment that matters most: when money changes hands.

Why existing approaches break for trustless markets

Optimistic verification, Schelling-style voting, escalation games: genuine attempts at trustless resolution, but when stakes rise and outcomes become contested, finality defaults to token-weighted governance rather than economic incentive aligned with facts. Infrastructure-led oracles aggregating API data are excellent for price feeds, but as a resolution layer for permissionless markets they require infinite horizontal scaling and create an economic dependency at odds with the trust model the market claims to offer.

Consider: Ethereum does not outsource its consensus rewards to a third-party oracle. The mechanism that secures the network is native to the network. If Ethereum's block rewards were determined by Chainlink rather than by Ethereum itself, we would rightly question the system's integrity. The same logic applies to prediction market settlement.

The oracle should be a market

If existing approaches are institutions, APIs, or governance games bolted onto prediction markets, the alternative is to make the oracle itself a market.

A market-led confirmation surface means financially incentivised participants converge on a Schelling point representative of common knowledge, not through voting weight or governance authority, but through capital. Disagreement becomes price discovery rather than a coordination game vulnerable to whale capture. As capital is exchanged rather than used to signify voting weight, the oligarchic tendencies of stake-based systems are reduced. The mechanism's incentives become auditable at every layer.

When outcomes are further mapped onto numerical ranges rather than binary yes/no buckets, the surface area for subjective distortion shrinks. The question "did X exceed $4 billion?" invites semantic disputes over sources and definitions. A continuous numerical market simply asks "where did X land?" and lets the confirmation mechanism converge on a point. Ambiguity is reduced structurally, not adjudicated away.

This is not a theoretical distinction. It could be argued that the last two to three percent of an event-contract's probability is purely resolution risk priced in. That spread, the gap between what the market believes and what it is willing to pay given settlement uncertainty, is measurable. It shrinks when the settlement mechanism earns trust over time and expands when it does not. Resolution credibility is not abstract. It lives in the price.

When that confirmation surface is mathematically linked to its prediction market counterpart, it can strengthen the market's manipulation resistance and deliver compounding trust over time. The benefits of AI or API-based truth extraction, including speed and finality, can then operate as a layer above the resolution mechanism rather than being the mechanism itself.

The design space is open

Just as DEXs and CEXs serve different trust models and should co-exist, truly decentralised prediction markets can and should co-exist with their permissioned counterparts. But decentralisation is a design commitment, not a label. Each component of the market mechanism, including settlement, needs to be designed for that environment.

If the market is trustless but the oracle is not, the system has not eliminated trust. It has relocated it.

Oracle design is institutional design. It determines how reality enters a digital system, how errors propagate, and how costly it is to distort shared facts. When that surface is credibly neutral and economically legible, participants can reason about it. When it is opaque, they cannot.

Igor leads research at @functionspaceHQ an open-source project exploring market-led resolution and novel economic instruments for prediction markets.