Predict Bay vs Polymarket: Two Different Products

Predict Bay and Polymarket are both prediction markets, and they are not competing for the same trade. Polymarket is an event-market venue on Polygon: a central limit order book over binary outcome shares covering elections, sports, macro releases and other real-world questions, resolved by an optimistic oracle with a human dispute path. Predict Bay is a short-duration crypto price venue: hourly and daily rounds on BTC, ETH, SOL, SUI, DOGE and XRP, where a strike is set at the open, you pick above or below, and the round resolves against an on-chain price feed. Pick by the shape of the question first. "Who wins the election" and "where is BTC in an hour" are different instruments with different resolution problems, and no single venue design is good at both.

The second difference is maturity, and it is worth stating before anything else. Polymarket is live with real money. On Predict Bay, the Sui path targets a testnet deployment of DeepBook Predict and is switched off for real funds, and EVM/Polygon real-money trading is currently paused — practice mode is what is actually tradeable on Predict Bay today. That is the same thing the DeepBook Predict reference page says, for the same reason: a comparison you cannot verify is not worth reading.

Two different products, not two versions of one

Polymarket is the venue most people mean when they say "prediction market", and it defined the shape of the event-market category. Its core asset is an outcome share priced between 0 and 1 USDC that redeems for exactly 1 USDC if its outcome is the one that happens. That design is deliberately general: any question that can be phrased as a set of mutually exclusive outcomes with a written resolution rule can be listed, whether it settles next week or next year. The hard part of that generality is resolution, because most real-world questions have no canonical number to read at expiry.

Predict Bay narrows the question type on purpose. Every market Predict Bay lists is a short-duration price market on a major crypto asset, which means the thing being asked about already has a continuously published price. That removes the adjudication problem entirely and replaces it with an oracle problem, which is a smaller and more mechanical one. The trade-off is real and runs both ways: Predict Bay cannot list "will the central bank cut rates in September", and a general event venue cannot settle an hourly BTC round in seconds without a price oracle doing the work anyway. If you want the mechanics of the binary payoff itself, binary options vs prediction markets covers the instrument shape in more detail.

Side by side

Two rows in that table carry most of the weight — resolution, and where the book lives. They are the choices everything else follows from, so the rest of this post is mostly about them.

How each one decides who won

An optimistic oracle is a social mechanism wearing a smart contract. Someone proposes the outcome and posts a bond. A challenge window opens. If nobody disputes within the window, the proposal stands and settles. If someone does dispute, the question escalates to a vote of the oracle's token holders, and whichever side is wrong forfeits its bond. The design is elegant because it costs almost nothing in the common case — most questions are not close, nobody bothers to dispute, and settlement is effectively free. It is also the only mechanism that can settle a question no data feed publishes, which is precisely why an event venue needs one.

Its costs are the ones you would expect. Settlement is not instant, because the challenge window has to run. Ambiguity is adjudicated by people, so the exact wording of the resolution criteria matters enormously and edge cases become governance events. And the final backstop is a vote, which is a different trust assumption from reading a number.

A price-feed oracle inverts every one of those properties. DeepBook Predict settles against its on-chain oracle price at expiry: the contract reads the feed, the payoff is determined by arithmetic, and the holder redeems. Nobody proposes anything, there is no window, and there is no vote. The cost is that it only works when a canonical price exists, and it moves the entire trust question onto the feed — its update cadence, how it aggregates across the publishers contributing to it, and how the contract handles a stale or missing update at the settlement instant. That is not a smaller problem, it is a different one, and oracle resolution and on-chain settlement works through it properly.

The practical consequence for a trader is simple. An hourly round can only exist on the second design. A challenge window measured in hours cannot resolve a market measured in hours.

Where the order book actually lives

Neither venue prices its binaries with an automated market maker, which is the right call for this instrument — a bounded 0-to-1 payoff has a natural quoting structure that a constant-function curve fits badly. The difference is where the book is kept. Polymarket runs a hybrid design: orders are matched off-chain for speed and cost, and the resulting trades settle on-chain non-custodially, so the operator sequences the book but cannot take your collateral. That is a well-understood engineering compromise, and on a chain with block times and fees like Polygon's it is a reasonable one.

DeepBook takes the other branch. The book itself is on-chain state on Sui — resting orders, cancellations and fills are transactions against a shared object, not rows in an operator's database. This is tractable because of how Sui executes: objects are owned or shared explicitly, transactions touching only owned objects can skip consensus entirely, and transactions touching disjoint objects execute in parallel rather than queueing behind one global lock. Traffic against a shared order-book object still has to be ordered by consensus — a shared object is a sequencing point, not a free lunch — but the rest of the chain does not queue behind it. An on-chain CLOB is a bad idea on a chain that serialises everything and a workable one on a chain that does not. DeepBook Predict sits on that stack and adds the options layer: you mint a binary position, hold it on-chain until expiry, and redeem it afterwards. On the testnet deployment the settlement asset is dUSDC. DeepBook, CLOB liquidity and prediction markets goes further into what that means for depth and spreads.

Outcome shares versus binary options

The payoff profiles rhyme — both pay a fixed amount on one side and nothing on the other — but the objects differ. A conditional-token outcome share is minted as a complete set: deposit 1 USDC of collateral, receive one YES and one NO, sell whichever side you do not want, and redemption burns the winning side against the collateral pool. A DeepBook Predict position is specified instead by an expiry and a strike expressed as a price tick, with sentinel tick values encoding one-sided contracts and a bounded pair of ticks encoding a range. That representation is why one primitive expresses "above 70,000", "below 70,000" and "between 68,000 and 72,000" without three different market types.

Underlyings

BTC, ETH, SOL, SUI, DOGE, XRP

Round length

Hourly and daily

Strike

Set at the round open

Your choice

Above or below

Resolution source

On-chain price feed for the underlying

Sui settlement asset (testnet)

dUSDC

Tradeable on Predict Bay today

Practice mode

How a round runs

1. The strike is set at the open

   When a round opens, the strike is fixed from the reference price at that moment. Everyone in the round trades against the same number, and it does not move afterwards.

2. You take a side

   Above or below. The mode switch on a market page decides where that goes — with EVM off it renders two options, Practice and Sui, and Practice is the one that accepts a trade today. Selecting Sui shows the coming-soon panel in place of the trading controls.

3. The round runs

   Price moves, and so does the fair value of each side. A one-hour round spends most of its life with the outcome genuinely uncertain, which is what makes short-duration markets worth pricing at all.

4. Expiry reads the oracle

   At expiry the settlement price is read from the feed and compared with the strike. No proposal, no challenge window, no vote — the comparison is arithmetic.

5. The winning side redeems

   On the Sui path — the testnet integration, which is not open for trading on Predict Bay today — the holder redeems the position on-chain for the payout, in dUSDC on that deployment. In practice mode, which is what you can actually trade now, the equivalent credit lands in your practice balance.

Accounts, custody and gas

Both products have worked on the same problem — that asking a new user to write down twelve words is where most of them leave — and solved it differently. Polymarket creates an embedded wallet from an email login, with an external wallet as the alternative for people who already have one. You still end up holding USDC on Polygon, so funding generally means an on-ramp or a bridge.

Predict Bay signs you in with Google through zkLogin, which is a Sui-native primitive rather than a wallet vendor. The short version: your browser holds a short-lived ephemeral key, and a zero-knowledge proof establishes that a valid OAuth credential from your provider, combined with a salt that is never published on-chain, corresponds to your Sui address — without putting the credential or your identity into a transaction. The network verifies that proof together with the ephemeral signature, so what authorises a transaction is a login plus a proof rather than a permanent key you had to write down and keep. There is no seed phrase to lose because there is no seed phrase. Transactions can also be sponsored, meaning you can transact without holding SUI for gas — a meaningful difference from any design whose first step is acquiring the chain's native token. Connecting a Sui wallet directly is supported for anyone who prefers that. Prediction markets without a seed phrase covers the mechanism and its trade-offs, including what the ephemeral key's expiry and the salt actually mean for you, and how it works walks through the flow in the product.

What is live, and what is not

That box is not a disclaimer bolted on at the end; it is the honest half of this comparison. If you want to place a real-money trade on an event outcome right now, Polymarket does that and Predict Bay does not. What Predict Bay offers today is the full market structure with simulated balances — same rounds, same strikes, same on-chain resolution, no capital at risk — plus a settlement path being built in the open rather than announced as finished. You can see the structure on the BTC market page, and the Sui tab and its coming-soon panel on the Sui tab on the BTC market.

Publishing which switch is off, and where it lives, is a deliberate position rather than an oversight. Most venues in this category describe roadmap as product. We would rather cost ourselves a conversion and keep a reader who can trust the next thing we publish. The FAQ answers the same questions in the same terms, and what is DeepBook Predict covers the protocol on its own.

On the token: $PBAY is a concept at this stage. Intended utility can reasonably be described in conditional terms — the fee and reward mechanics most trading venues eventually adopt — but nothing about supply, timing or distribution has been decided or announced, and no page here will imply otherwise. The token page holds whatever is actually known, and the $PBAY explainer says the same thing at more length.

Which venue fits your question

• You want to trade a real-world event outcome with real money. Polymarket. Predict Bay does not list event markets, and practice balances are not a substitute for a live position.
• You want to express a view on where a crypto asset lands in the next hour or day. That is Predict Bay's entire market set, and the resolution path — a price feed read at expiry — is what makes those durations possible at all. On Predict Bay that means practice mode today; the real-money paths are covered below.
• You want to learn the instrument before risking anything. Practice mode mirrors the real market structure exactly — the same rounds, the same strike-at-open, the same on-chain feed at expiry. It is not a demo running on invented prices.
• You care where the order book physically lives. An off-chain matcher with on-chain settlement and a fully on-chain CLOB are genuinely different trust models. Neither is strictly better; pick the one whose failure mode you would rather have.
• You do not want to manage a seed phrase or hold a gas token. zkLogin plus sponsored transactions removes both requirements without handing custody to a wallet vendor.
• You need real-money crypto price markets today. Neither the Sui path nor the Polygon path is open for that on Predict Bay right now. Stated plainly so you do not discover it three clicks in.

The useful summary is that these are complements more often than substitutes. An event venue and a short-duration price venue address different halves of the same instinct — putting a number on what you think will happen — and they arrive at different architectures because the questions they answer resolve differently. If what you actually wanted was a comparison of the market types rather than the brands, prediction markets explained starts from first principles.