Oracles and disputed resolutions

A smart contract can hold funds and pay them out, but it has no way to find out who won an election or what a statistical agency published. That information has to be delivered to it by something outside, and that something is the oracle. It is the single most consequential component of an on-chain prediction market, because whatever the oracle says, the contract pays — there is no appeal to a human administrator afterwards.

This makes the oracle the counterparty risk of the design. On a regulated exchange, if you believe a determination is wrong you can complain to the exchange and, failing that, to its regulator. On an on-chain market the mechanism is the whole of the process; there is no venue with discretion to overrule it, by design.

The dominant design is optimistic, meaning it assumes the first answer offered is correct and only does expensive work when someone objects. In outline: after the resolution date, a proposer submits an outcome and posts a bond. A challenge window opens. If nobody disputes within it, the proposal is accepted and the market settles. If somebody does dispute — also posting a bond — the question escalates to a vote, and the loser’s bond compensates the winner.

Polymarket resolves through UMA’s optimistic oracle, with escalated disputes decided by a vote of UMA token holders. The economics are the interesting part: proposing a false outcome costs you your bond if anyone is watching, and disputing a true one costs you yours. The design is not trying to make lying impossible, only more expensive than it is worth.

Bond sizes and window lengths are parameters, set per market and changed over time, so they are not stated here — but they are the two numbers that determine how much the mechanism actually deters. A bond small relative to the money in a market deters less; a window short relative to how long it takes anybody to notice deters less. Both are visible on the market itself, and both are worth a glance before trading something you expect to be contentious.

The word suggests a factual disagreement, and occasionally it is one. Far more often it is a textual argument: the event happened, everybody agrees it happened, and the question is whether the market’s wording captures it. Did a partial measure count as the measure named? Did an announcement on the last day fall inside the deadline? Does the named source, which published a figure and then revised it, mean the first figure or the second?

These arguments arise because prose is worse at edge cases than the world is at producing them. A market drafted before an event rarely anticipates the specific way the event turns out to be strange. When it does not, the oracle is being asked to interpret rather than to report, and interpretation is where honest people diverge.

It follows that the way to reduce your exposure to disputes is not to pick markets with reliable oracles. It is to pick markets whose text is specific enough that interpretation is not required — and to be more careful with the ones whose text is loose, however obvious the underlying event seems.

A market under dispute does not trade at the price implied by its likely outcome. If a contract is very probably resolving YES but the proposal has been challenged, it will trade below where certainty would put it — sometimes well below — because the price now carries two things it did not carry before: the chance the dispute goes the other way, and the cost of capital committed for an unknown further period.

That gap is tradeable in both directions and is not the same kind of bet as the original market. Buying a disputed contract at 80¢ is a position on a governance process, not on the underlying event. Some traders specialise in exactly that; it is worth being clear with yourself about which of the two things you are doing.

The other visible signature is volume drying up. Participants who understand the mechanism tend to stand back until it clears, which widens the spread, which makes the price less informative at precisely the moment people are most curious about it.

Read the resolution criteria and try to break them. Take the two or three most plausible messy outcomes and see whether the text says what happens. If it does not, the market has an interpretation risk that is not in its price for you.

Look at whether the market has a resolution history. A market that has already been proposed and disputed once tells you something about how legible its wording is. So does a venue’s record on similar questions, which is public and readable in a way that most platform claims are not.

And separate the two risks in your own head before committing anything. “Will the event happen” and “will the mechanism agree that it happened” are different questions with different answers, and a position is exposed to both.