What a prediction market is, how the DeepBook version differs from event-market platforms, and why a strike expressed as a price tick changes how settlement works.
Predict Bay ·
DeepBook prediction markets are binary contracts that live entirely on Sui. Each one pays a fixed amount if an asset settles on one side of a strike price at a stated expiry, and nothing if it does not. They are built on DeepBook Predict, an on-chain protocol that sits alongside DeepBook — Sui's native central limit order book — and positions are minted, held and redeemed on-chain. No off-chain matching engine holds the book, and no operator decides the outcome.
ROUND CLOSES IN
--:--
LAST PRICE
—
STRIKE
—
NO POSITIONS YET THIS ROUND — BE THE FIRST
Practice mode · no wallet, no deposit, free to start
That last point is what separates a DeepBook prediction market from the event-market platforms most people picture when they hear the term. There is no rules text to interpret and no resolution committee to appeal to. A market is a strike, an expiry and an oracle feed; at expiry the protocol reads the price and pays. This page covers what a prediction market is at the mechanical level, where the price you pay actually comes from, why the strike is expressed as a price tick rather than a dollar figure, how settlement resolves, and exactly what Predict Bay runs today.
A prediction market is a market in a contract whose payout is determined by a future fact rather than by an issuer's promise. The simplest form is binary: the contract pays one unit of the settlement asset if a stated condition holds at a stated time, and zero if it does not. Because the maximum payout is fixed and known in advance, the contract's price carries information. It is the market's working estimate of how likely the condition is.
Suppose a contract pays 1.00 of the settlement asset if BTC finishes above its strike at expiry, and it is trading at 0.62. Buying costs 0.62 to win 1.00 — roughly a 61% return if the condition holds, and a complete loss of the 0.62 if it does not. The market is quoting about a 62% chance. Take the other side and you are selling that same probability, which is to say buying the complement at 0.38. Price and probability are one number seen from two directions, which is why prediction-market quotes are often written as percentages with no unit at all.
This bounded structure is what makes prediction markets legible compared with leveraged directional products. There is no liquidation price, no funding rate quietly accruing against the position, and the worst case is fixed at entry. What you give up is convexity: being spectacularly right pays exactly what being barely right pays. That trade-off, and where binary options and event contracts genuinely diverge, is covered in binary options versus prediction markets.
Most platforms sold as prediction markets are event markets. A question is written in prose — an election result, a policy decision, a sports outcome — and a resolution process decides what that prose meant once the event has happened. The prose is the product. Well-written rules text anticipates ambiguity, names the source of truth and specifies what happens if the source goes silent. Badly written rules text becomes a dispute, and disputes are the risk surface that event markets have to engineer around: oracle committees, challenge windows, bonded reporters, escalation to a token vote.
A DeepBook prediction market takes a much narrower subject and removes the interpretation step entirely. The condition is arithmetic: is the oracle price at expiry above or below this level? There is no wording to parse, so there is nothing to dispute and no committee to be captured. The cost of that certainty is scope. You can only build a market on something an oracle already publishes, which in practice means asset prices.
| Dimension | Event-market platform | DeepBook prediction market |
|---|---|---|
| Subject of the contract | An arbitrary future event stated in words | A price level on an underlying asset at a fixed expiry |
| How the question is defined | Rules text, interpreted by humans after the fact | A numeric strike expressed as a price tick, plus an expiry |
| Who resolves it | An oracle committee, a reporter with a bond, or the operator, usually with a dispute window | The protocol, by reading its on-chain oracle price at expiry |
| When the contract ends | When the event happens and the resolution process concludes, which can lag the event itself | At a timestamp fixed when the contract is created |
| Where the market state lives | Commonly an off-chain matching engine, with settlement posted on-chain | On-chain: positions are minted, held and redeemed as chain state, and no off-chain matching engine holds the book |
| What can go wrong | Ambiguous wording, contested resolution, operator discretion | Oracle availability and the granularity of the strike grid |
Neither model dominates the other. Event markets can price things no feed will ever publish, and that breadth is their reason to exist. Price markets give up subject breadth in exchange for a resolution rule a machine can evaluate with no discretion at any step. If you want the protocol-level detail on the DeepBook side, what DeepBook Predict is goes through the object model.
A central limit order book is the market structure most traders already know from equities and futures. Participants post limit orders at discrete prices; those orders rest until they are matched, cancelled or expire; incoming orders match against the best available price with time priority breaking ties at the same level. DeepBook implements this natively on Sui, which means the resting orders, the matching and the fills are all chain state rather than rows in an operator's database. That is the stack DeepBook Predict is built alongside, and it is why a binary position can be minted, held and redeemed with the same on-chain guarantees a spot trade gets.
A binary option cannot simply be left to rest on a book and forgotten, because its fair value is a probability that moves continuously with spot, with time to expiry and with implied volatility, and then collapses to exactly zero or exactly one at a known moment. Something has to be modelling it the whole way. DeepBook Predict does that on-chain: at mint time the protocol reads the oracle's forward price for the relevant expiry along with the published volatility surface, and prices the contract as the risk-neutral probability that settlement lands on your side of the strike. There is no off-chain counterparty agreeing a number with you, and no venue-controlled figure anywhere in the path.
The dominant alternative in DeFi is an automated market maker, where a pool quotes a price from a formula over its reserves. AMMs are excellent when the two assets are fungible, long-lived and roughly symmetric. Binary options are none of those things. A contract on BTC above a given strike expiring in one hour is a distinct instrument from the same strike expiring the next hour, and it becomes worthless or fully valued on a schedule. A reserve-ratio curve has no input for time to expiry, so it has no way to know that. A model that prices from the forward and the volatility surface does, because expiry is one of its inputs.
What size costs you is therefore not a matter of walking down a book. The quoted price responds to how much exposure the protocol already carries on that side and how heavily the collateral behind it is utilised, so a large position moves the price against itself through those terms rather than by consuming resting depth. The all-in cost — fees, any financing on a leveraged position, congestion terms — is computed and enforced on-chain as the mint executes. That is why the sane way to protect yourself is a maximum-cost guard on the transaction rather than a slippage tolerance measured in ticks: you are bounding what you will pay, not guessing how deep a book is.
The hard part of liquidity in any options market is that supply splits across a grid. Every combination of strike and expiry is technically a separate instrument, and a design that treats them as unrelated ends up with hundreds of thin, disconnected markets. Pricing the whole surface from one model is the structural answer: every strike on an expiry is quoted from the same forward and the same volatility surface, so an additional strike is not a new market that has to attract its own liquidity — it is another point on a curve that is already being published. Shared infrastructure does the rest. On the deployment Predict Bay integrates against, a trader funds every position from a single on-chain account object rather than keeping a separate balance per instrument. That is the structural argument for building prediction markets on an existing CLOB stack instead of shipping a bespoke venue, and it is unpacked further in how the DeepBook CLOB supplies prediction-market liquidity.
Order books do not have a continuous price axis. They have a grid, and the spacing of that grid is the tick size. A strike of "$68,400" is really the index of a level on that grid, and the price it denotes is that index scaled by the tick size. DeepBook Predict follows the same convention: a position is defined by an expiry and a strike expressed as a price tick.
That choice is not cosmetic. Settlement has to compare two numbers and decide which is larger, and every validator has to reach the identical answer. Move, the language Sui contracts are written in, has no floating-point type at all — on-chain arithmetic is integer arithmetic, and prices are carried as fixed-point integers with a fixed scaling factor. Discretising the price axis and comparing integers makes the settlement predicate deterministic by construction. A decimal price rounded slightly differently by two implementations is a nuisance in a spot market; in a binary contract it is the difference between the full payout and zero.
The tick encoding also gives the protocol a clean way to express contract shapes. Sentinel tick values represent one-sided contracts — "above X" and "below X" — where one boundary is conceptually infinite and needs no real level on the grid. A bounded pair of ticks expresses a range contract: it pays if the settlement price lands between the two levels. The same position type covers all three, with the boundaries doing the work.
There is a second grid worth knowing about, and almost nobody bothers to write it down. The tick size is the finest increment the market quotes on, but the strikes you are permitted to mint against sit on a coarser admission grid — a whole multiple of the tick size. Open interest therefore concentrates on a manageable set of strikes instead of smearing across every quotable increment, which is what keeps a surface coherent rather than fragmenting it back into hundreds of unrelated instruments. A mint aimed at a strike that is not on the admission grid is rejected by the chain outright, not silently rounded to the nearest admitted level.
The lifecycle of a position is on-chain from end to end. Nothing in it depends on an operator being online, solvent or honest at the moment it matters.
Mint
You submit a transaction that creates a position against a specific expiry and strike tick, paying the premium in the settlement asset. The price is computed on-chain as the transaction executes, and the position exists as chain state from that moment.
Hold
The position sits in your on-chain account. Its value moves with the underlying and with time to expiry, and the deployment exposes a redemption path before expiry as well as after it — a position does not have to be carried to the end.
Expire
At the expiry timestamp the contract stops being a probability and becomes a fact. Nothing the price does afterwards affects it.
Resolve
The protocol reads its on-chain oracle price for the underlying at expiry and compares it against the position's strike tick. The comparison is integer arithmetic and admits no discretion.
Redeem
The holder redeems the position for the payout — the full amount if the condition held, nothing if it did not.
Pricing and settlement both read from feeds published on Sui rather than from any internal number the venue controls. DeepBook Predict reads spot, forward and volatility-surface data, alongside Pyth price feeds for the underlying asset. The extra feeds matter more than they look. A binary contract's fair value is not a function of spot alone; it depends on where the market expects the asset to be at expiry (the forward) and on how widely it might be distributed around that point (implied volatility). Two contracts on the same asset at the same strike, expiring an hour apart, are worth different amounts for exactly that reason, and a protocol that priced from spot alone would be systematically wrong on both.
The volatility surface is published as a compact set of parameters describing the whole curve rather than as a grid of individual quotes. That is what makes on-chain pricing tractable: the contract does not look up a stored price for your specific strike, it evaluates the curve at your strike. Every admitted strike on that expiry is priced from the same publication — which is also why the surface, rather than any individual market, is the thing liquidity is really being provided to.
The practical consequence for a trader is that the resolution rule is knowable in advance and checkable afterwards. You can read the feed the market settles against, watch it in real time, and verify the settlement transaction. There is no step where a human decides what the price "really" was. The mechanics of oracle reads, staleness handling and what happens when a feed is unavailable are covered in oracle resolution and on-chain settlement.
Every market page carries a mode switch. With EVM trading withdrawn it renders two options: Practice for free and Sui for DeepBook Predict. Sui is the segment that leads to the protocol rather than to an internal order book — when the integration is switched on, the app builds the mint transaction, submits it, and reads positions and balances back from the chain. Today that tab renders the coming-soon panel described above instead of the trading controls, so the path you can take end to end right now is Practice: the same market, the same strike, the same resolution feed and the same expiry, with simulated funds.
The live markets are short-duration price markets on BTC, ETH, SOL, SUI, DOGE and XRP, running as hourly and daily rounds. A strike is set at the open, you pick whether the asset finishes above or below it, and the round resolves against an on-chain price feed. That structure is deliberate: practice mode is not a toy with invented prices. The strike, the feed and the expiry are the ones the real market uses — only the funds are simulated — so the habits you build carry over rather than having to be unlearned. You can see it on the BTC market page.
Two things about the account model are worth calling out, because they change what onboarding costs. Sign-in is Google via zkLogin: your Sui address is derived from the OAuth credential together with a salt, and transactions are signed by a short-lived ephemeral key accompanied by a zero-knowledge proof that the credential matches the address. There is no seed phrase to write down and no browser extension to install, and the identity provider is not handed the ability to move your funds. The details are in prediction markets without a seed phrase. Second, transactions can be sponsored, so holding SUI for gas is not a prerequisite to placing a trade. Connecting a Sui wallet directly is supported for anyone who would rather hold their own keys.
The questions below are the ones that separate a venue with real settlement guarantees from one with a good landing page. They apply to DeepBook prediction markets and to everything else in the category.
Predict Bay publishes the answer to that last question rather than burying it: the DeepBook Predict reference page states the current status of the integration in the first screen, and how it works walks through the trading flow end to end. Trading prediction markets carries risk and you can lose the full premium you commit to a position — that is the design, not a failure mode.