Two non-crypto assets, on a venue built for crypto price rounds. Adding them was not a configuration change: oil has no spot price feed to point at, and both markets close for a third of every week. Here is what that actually required.
Predict Bay ·
Predict Bay now lists gold and oil alongside BTC, ETH and SUI. Same instrument as the crypto markets — a strike set at the open, a choice of above or below, resolution against an on-chain price feed at expiry — on two underlyings that have had a canonical published price for considerably longer than crypto has existed.
ROUND CLOSES IN
--:--
LAST PRICE
—
STRIKE
—
NO POSITIONS YET THIS ROUND — BE THE FIRST
Practice mode · no wallet, no deposit, free to start
That is the easy sentence. The reason this post exists is that neither asset behaved like the crypto ones once we started wiring them up, and both differences are the kind that fail quietly rather than loudly. One of them would have broken the oil market roughly a month after launch. The other would have kept both markets open through every weekend, taking trades on a price that could not move.
Gold resolves against Metal.XAU/USD — Pyth's spot-gold feed, quoted in US dollars per troy ounce. It behaves the way a price feed is supposed to: one feed, one underlying, published continuously while the market is open, no expiry.
There is one trap worth naming, because it is easy to hit and produces a market that looks correct. Pyth also publishes feeds for tokenised gold — XAUT and XAUM among them. Those are Crypto asset-class feeds, and they track the price of a wrapper token, which is a claim on gold rather than the metal. Their price can and does diverge from spot on token-specific supply, redemption mechanics and venue liquidity. A market labelled "gold" that settles against a tokenised-gold feed is settling a different question from the one the label asks. Predict Bay's gold market reads the Metal feed for exactly this reason.
Ask Pyth for oil and you do not get a feed. You get a list of dated futures contracts — WTI expiring in September, in October, in November — each with its own feed ID and its own price. There is no Commodities.OIL/USD, because there is no such thing as the spot price of crude in the way there is a spot price of gold. What people mean by "the oil price" is a convention: the price of the nearest-dated futures contract, the front month.
So an oil market has to choose a contract. We take the earliest non-expired one, deliberately: the front month is both the most actively traded and the most volatile of the ladder, and it is the number quoted when anyone says oil is up. A further-dated contract would be more stable to point at and would drift away from the price a trader is actually watching, by a carry premium that has nothing to do with the question being asked.
That choice creates the failure mode. Contracts expire. When one does, Pyth does not delete the feed — it rewrites the feed's description with a DEPRECATED FEED - prefix and stops publishing new prices to it. A venue holding a hardcoded feed ID therefore keeps a market that looks entirely healthy: the asset is listed, the chart renders, the round opens. The price simply stops changing. Every round then settles at its own strike, and it does so silently, because nothing errored.
Predict Bay resolves the front month at runtime instead of storing one. The contract list is read from Pyth on a schedule, contracts whose description carries the deprecation prefix are discarded, contracts already past their date are discarded, and the earliest survivor becomes the oil feed. The roll happens on its own, monthly, with nobody editing a constant — which is the only version of this that stays correct, because the alternative is a recurring calendar task that works until the one month somebody misses it.
Metal.XAU/USD — spot gold, USD per troy ounceCrypto never shuts, and a venue built only for crypto absorbs that assumption everywhere without ever writing it down. Gold and oil break it. Both feeds carry a CME/COMEX schedule, and it has real holes in it:
A three-minute up/down round is only a market while the price can move. Run one against a closed feed and the strike and the settlement price are the same number, so the round resolves flat every time and both sides lose the fee. Over a single weekend that is hundreds of rounds per commodity, on a screen that looks completely normal — the chart is drawn, the countdown runs, the price is a real price that was true on Friday afternoon.
Predict Bay does not open commodity rounds while the underlying exchange is shut. The check happens where the rounds are created, which is stronger than detecting staleness at settlement: by settlement the trader has already staked. Rounds opened before the close still settle normally. Crypto is unaffected and is never skipped.
A short-duration price market needs one thing from its underlying: a continuously published, canonical price that both sides agree to be bound by. It does not care whether that price comes from a blockchain. Gold and oil satisfy the requirement completely, and they bring a property crypto does not — they are the assets most people have an unprompted opinion about, because their prices show up in petrol stations and headlines rather than only in portfolios.
They are also not an unusual thing to list. Short-duration commodity contracts exist elsewhere in the category, including intraday oil markets on mainstream retail platforms, and gold settlement against a Pyth feed is an established pattern rather than something invented here. What differs between venues is not whether the market can exist but how carefully the two problems above are handled, which is most of what this post has been about.
The instrument is the same one described in binary options vs prediction markets: a fixed payout on one side of a strike, fully collateralised at the ticket price, with no margin, no funding and no liquidation. The resolution path is the same one in oracle resolution and on-chain settlement — a price read at one timestamp, with no proposer and no dispute window. Adding two commodities widened the list of underlyings and nothing else about how a round works.