Market Structure & Instruments

This document explains what trades on Proof and how the instruments fit together. It is conceptual: it covers the order book model, the instrument types, and the relationships that link them. It is not an integration guide — there are no endpoints, payloads, or code here.

Proof has two layers. The first is a set of ordinary perpetual futures on assets, each on its own order book. The second is what makes Proof distinct: conditional markets, which let you trade an asset conditioned on whether a real-world event happens, alongside a market on the event's probability itself. Everything below builds up to that second layer.

The order book (CLOB)

Every market on Proof is a central limit order book (CLOB). Buyers and sellers post limit orders — a price and a size — and the engine matches them by price-time priority: better prices match first, and among equal prices, the order that arrived first matches first.

A few properties matter for understanding how the market behaves:

• Matching is at the resting price. When an incoming order crosses a resting order, the trade happens at the resting (maker) price; the incoming (taker) side pays it. The maker sets the price; the taker accepts it.
• Time-in-force controls what happens to the unfilled remainder. A good-til-canceled order rests on the book until filled or canceled. An immediate-or-cancel order fills what it can right now and drops the rest — it never rests. A fill-or-kill order is checked up front against available liquidity and rejected entirely if it cannot fill in full.
• Tick and lot gates. Each market defines a smallest price increment (the tick) and a smallest size increment (the lot). Limit prices must be a whole multiple of the tick, and sizes a whole multiple of the lot; orders that violate either are rejected. A pure market order carries no price, so it skips the tick check but still respects the lot.
• Quote refresh is favored over fresh taker flow within a block. When orders are processed together, cancels and amendments are applied before new limit orders, which are applied before market orders. The practical effect: a market maker who updates a quote by amending or canceling-and-replacing is processed ahead of taker flow arriving in the same batch. This is a deliberate speed bump that protects resting liquidity from being picked off by fast incoming orders.
• One account, one collateral pool. All of an account's positions across every market share a single collateral balance. This cross-margin model is what makes hedged positions in a conditional-market family capital-efficient; it is covered in depth in the margin documentation and summarized later here.

These mechanics are identical across every instrument type below. What differs between instrument types is how a position is valued, whether it carries funding, and how it settles.

The underlying perpetual

The base instrument is a standard perpetual future — call it asset-PERP. It tracks an underlying asset (for example, a major cryptocurrency) with no expiry. Key facts:

• Mark price. A perpetual is valued against a mark price derived from an external oracle feed. The mark drives margin, unrealized P&L, and liquidation. (Proof maintains a second, separate reference — a book-tracking average — that is used only for funding; the two can diverge by design.)
• Funding. Like any perpetual, asset-PERP carries funding: a periodic payment between longs and shorts that keeps the contract's price tethered to the underlying. When the contract trades above the oracle, longs pay shorts; when it trades below, shorts pay longs. Funding is zero-sum at the index level and is settled to cash. Funding applies to perpetuals only — it never applies to any of the conditional instruments below.
• Leverage. The perpetual carries configurable leverage via its initial-margin (IM) and maintenance-margin (MM) requirements. IM is the collateral to open a position; MM is the minimum to keep it open, and falling below MM triggers liquidation.

We will denote the underlying perpetual's fair price the forward, written F. This single value is the anchor for an entire conditional-market family.

Conditional markets: the family

A conditional market is created over one pre-existing underlying perpetual and one binary event — a real-world question that will resolve YES or NO (for example, an event such as a scheduled economic decision, or a price-threshold event of the form "the asset closes above a level K"). Creating one conditional market produces a linked set of order books, created together, that we call a family.

A family has five linked books:

1. The underlying perpetual — asset-PERP, described above. It already exists and is referenced by the family, not recreated.
2. The conditional perpetual (YES) — the asset if the event resolves YES.
3. The conditional perpetual (NO) — the asset if the event resolves NO.
4. The prediction binary (YES) — a $0–$1 instrument whose price is the implied probability the event resolves YES.
5. The prediction binary (NO) — the mirror, for the NO outcome.

The four event-driven books (the two conditional perpetuals plus the two prediction binaries) are collectively the conditional legs. The underlying perpetual is not a conditional leg; it is the shared anchor.

Each book in the family is an ordinary CLOB with its own bids, asks, and price. They are linked because their fair values are tied together by no-arbitrage relationships, and because they settle against the same event and the same underlying. Each book matches independently; the relationships that connect them are economic, and they are covered below.

Conditional perpetuals (YES / NO)

A conditional perpetual is a perpetual on the underlying that only fires if the event resolves its way. It trades on its own order book at its own price — the conditional forward — and behaves like a perpetual while the event is open. What makes it conditional is what happens at resolution:

• If its branch wins, it cash-settles to the underlying perpetual's mark, exactly as if you had held the underlying.
• If its branch loses (the other outcome wins, or the event turns out to be unresolvable), the position is voided: profit-and-loss is zero and your margin is returned.

This voiding behavior is the single most important property of conditional markets to understand. A losing conditional perpetual is not a loss. It does not settle to $0. It simply disappears: nothing changes hands, and the collateral you posted to hold it is released. Concretely, if you are long the YES conditional perpetual and the event resolves NO, you do not lose your position's value; you get your margin back and walk away flat on that leg.

Two consequences follow immediately:

• Conditional perpetuals carry leverage. Like the underlying, each has configurable initial- and maintenance-margin requirements (both strictly positive, with MM ≤ IM), and is margined as a perpetual in the branch where it fires.
• A conditional perpetual hedged against the underlying is a hedge in the firing branch. Suppose you are long a YES conditional perpetual and short the underlying perpetual to neutralize price risk. That holds while the event is open and if YES fires. If the event resolves NO, the YES conditional perpetual voids and the underlying short stands on its own — so the practice is be solvent in the post-resolution world, and flatten or fully box your position before the event resolves (see Settlement, Resolution & Void for the full mechanic).

We write the YES conditional perpetual's fair price as CY and the NO conditional perpetual's as CN. Each is fair at its conditional expectation — CY is the expected price of the underlying given a YES outcome, CN given a NO outcome — not at the underlying forward F. (Why this must be true, rather than both simply quoting F, is explained in the no-arbitrage section.)

Prediction binaries (YES / NO)

A prediction binary is a simple instrument priced in dollars between $0 and $1. Its price is the market-implied probability of its outcome. At resolution it pays:

• $1 if its outcome wins, $0 if its outcome loses.

So if you buy the YES prediction binary at a price of $0.30 (illustrative), the market is implying a 30% probability of YES; you pay $0.30 now and receive $1.00 if YES wins, or $0.00 if it loses. (Read its price as the market-implied probability, not an objective one.)

We write the YES prediction binary's price as BY and the NO binary's as BN. Because a fair YES probability and a fair NO probability sum to one, BY + BN should sit at about $1 — the binary box relationship, discussed below.

Prediction binaries have a few defining traits:

• No funding. Like the rest of the family, binaries never carry funding.
• No separate position margin. A binary's maximum loss is fully known and effectively pre-paid the moment you trade it (a long can lose at most what it paid; a short can lose at most one dollar minus what it received). Because that worst case is already reflected in your equity, the engine does not reserve additional position margin for a held binary. (A resting binary order does reserve its potential loss until it fills or cancels.)
• A static price grid. Binaries trade on a fixed, fine grid (sub-cent price ticks and small, hundredth-of-a-share lots) that does not inherit the underlying perpetual's grid. This keeps the $0–$1 probability space granular regardless of how the underlying is priced.

The unresolvable outcome (Void)

Beyond YES and NO, an event can be declared unresolvable — a Void. This is the operator path for events that genuinely cannot be settled. On a Void:

• Both conditional perpetuals void — P&L zero, margin returned, on both sides.
• Both prediction binaries pay $0.

That last point is worth stating plainly: on a Void, the binary box does not pay out a dollar split between the two sides. Both binaries go to $0. A holder of either directly-traded binary forgoes what they paid. The binary box identity (BY + BN ≈ $1) holds through a YES or NO resolution and resolves to $0 through a Void. For events where a Void is operationally implausible (a meeting that always reaches a decision, an oracle that always prints a level), the box is effectively reliable; read it as conditional on a YES/NO resolution.

How the family fits together

Put the pieces side by side. One family is built from:

• one underlying perpetual (the anchor, priced at forward F),
• two conditional perpetuals (CY and CN — the asset's price conditioned on each outcome), and
• two prediction binaries (BY and BN — the probability of each outcome).

The underlying tells you what the asset is worth unconditionally. The two conditional perpetuals tell you what it would be worth in each of the two possible worlds. The two prediction binaries tell you how likely each of those worlds is. Together they decompose a single asset's value into a probability and a pair of outcome-conditional prices — which is precisely the structure a trader needs to take a view on "what happens to this asset if that event goes one way."

A worked feel for the numbers: suppose an asset's forward is F = $100, an event is priced at p ≈ 0.40 (the YES binary trades near $0.40), and the market thinks the asset is worth about $108 if YES happens and about $94.67 if NO happens. Those are illustrative figures, but notice they hang together: the probability-weighted blend 0.40 × $108 + 0.60 × $94.67 ≈ $100 recovers the forward. That consistency is what arbitrage drives the books toward, and it is the subject of the next section.

No-arbitrage relationships between the legs

The five books are linked by relationships that hold between their fair prices. The matching engine matches each book independently; these relationships are economic, held in line by traders and arbitrageurs. Treat them as economic relationships.

Define three quantities:

• F — the underlying forward.
• p — the probability the event resolves YES. In a coherent family, p equals the YES prediction binary's price: p ≈ BY.
• Δ (the impact) — the conditional spread, Δ = CY − CN: how much the underlying's expected price differs between the two outcomes.

The relationships:

1. The forward identity. The underlying equals the probability-weighted blend of the two conditional prices:

F = p·CY + (1 − p)·CN

This is just the law of total expectation: the asset's price today is the average of what it is worth in each world, weighted by how likely each world is. It links the underlying directly to the conditional legs.

2. The binary box. The two prediction binaries sum to a dollar:

BY + BN ≈ $1

because exactly one outcome wins and pays $1. (As noted, this breaks through a Void.)

3. Implied probability. Rearranging the forward identity backs the probability out of the two conditional perpetuals:

p = (F − CN) / (CY − CN)

This gives you a second, independent read on the event's probability — one derived from the conditional perpetual books — that you can cross-check against the prediction binary's price BY. When the two disagree, there is an inconsistency to trade.

Why the conditional legs are fair at their conditional mean, not at F. It might seem natural to quote both conditional perpetuals at the underlying forward F. That is not a stable equilibrium. If both were quoted at F, then whichever leg's true conditional value is above F is underpriced and the other is overpriced — and a trader who knows (or simply believes) the event is informative about the asset can lift the cheap leg for positive expected value, with no offsetting risk on the other side. The selection-proof prices are CY = E[asset | YES] and CN = E[asset | NO]: each leg priced at its own conditional expectation, with the spread Δ between them absorbing the event's informativeness. Equivalently: CY = F + (1 − p)·Δ and CN = F − p·Δ, which you can verify reproduce the forward identity.

What drives the impact Δ. The spread between the two conditional prices is driven by how informative the event is about the underlying — structurally, by the correlation between the event (or its conditioning asset) and the underlying, scaled by the underlying's volatility. An event that barely moves the asset produces a thin spread; an event that strongly determines the asset's path produces a wide one. For threshold events of the form "the asset closes above K," the conditional prices follow truncated-distribution logic — in particular, the YES conditional sits above the strike K by construction, because if "above K" is the firing outcome, then by definition the settled price is above K. (The exact relationship between correlation, volatility, and Δ is a pricing model, not a fixed formula; the qualitative driver — correlation times volatility — is the part worth internalizing.)

What the books pin down. Between them, the family's prices pin down the event's marginal probability (p) and the two outcome-conditional means (CY and CN — the first moments). This is precisely the structure a trader needs to express a conditional view: a probability and a pair of outcome-conditional prices.

Cross-margin and capital efficiency

Because every position shares one collateral pool, the family's legs offset each other inside the margin engine. The binding check is a scenario test: the engine enumerates every YES/NO combination across the account's active conditional markets and requires the account to remain solvent in the worst scenario. No credit travels between scenarios — you can never lean on a gain that vanishes in some branch — which keeps the check robust.

The payoff of this design is hedged positions cost almost nothing to carry. A fully matched box — long the YES conditional perpetual, long the NO conditional perpetual, and short the underlying — has its payoff locked at entry in every branch, so it requires close to zero net initial and maintenance margin, rather than full margin on each leg separately. This near-zero cost for a fully hedged position is the core capital-efficiency thesis of conditional markets, and it is what makes running both sides of a family economical. (Each conditional market is scored independently in this enumeration, so the check never relies on correlation credit between events.)

One non-obvious consequence: because the requirement includes margin reserved for your resting orders, canceling resting orders can reduce your margin requirement and, in a tight spot, pull an account back from the edge of liquidation.

Settlement, void, and the "flatten before resolution" rule

When the event resolves, the entire family settles atomically — all of its books in a single step — to one of three outcomes: YES, NO, or Void. The winning conditional perpetual cash-settles to the underlying; the losing or voided one voids with P&L zero and margin returned; the binaries pay $1 / $0 (and $0 on a Void). See Settlement, Resolution & Void for the full per-leg mechanic and how each outcome is determined.

One positioning fact belongs here because it shapes how families are traded: a conditional-perpetual-plus-underlying hedge only holds in the branch where the conditional perpetual fires — in the branch where it voids, the underlying leg is left standing on its own, and the post-resolution picture is what the risk check sees. The standard practice is therefore to be solvent in the post-resolution world: flatten your position, or fully box it across both branches, before the event resolves. See Settlement, Resolution & Void for the full treatment, including how a box carries through resolution cleanly and how the risk sweep evaluates the post-settlement state.

Market parameters (conceptual)

Each market carries a set of configuration values that govern how it trades and how it is risked. Conceptually, the ones worth knowing:

• Tick size — the smallest allowed price increment. Limit prices must be whole multiples of it.
• Lot size — the smallest allowed size increment. Order sizes must be whole multiples of it.
• Size scale — a published decimal precision for sizes, so that displayed and traded sizes are unambiguous. Prediction binaries use a fixed, fine grid (sub-cent price ticks, small hundredth-of-a-share lots) independent of the underlying's grid; conditional perpetuals and the underlying use the underlying market's grid.
• Initial and maintenance margin (IM / MM) — the leverage settings. IM is required to open; MM is the floor below which the position is liquidated. Conditional perpetuals carry their own IM and MM (both positive, MM ≤ IM); the engine's binding gate is the worst-case scenario check, not a per-leg sum.
• Fees — a per-fill taker fee and a per-fill maker fee, applied to fill notional on every market type. These are flat, low single-digit-basis-point rates, with the taker rate higher than the maker rate. The fee mechanism also supports volume-tiered schedules.
• Funding cadence — how often funding is exchanged. Set for the underlying perpetual; disabled for the entire conditional family (conditional perpetuals and prediction binaries never accrue funding).
• Event metadata — for a conditional market, the event question, its resolution deadline, and a grace window. At the deadline the conditional legs stop accepting new orders and enter a pre-resolution state; structured oracle-resolvable events then settle automatically, and otherwise the event is voided after its resolution window.

These parameters are set per market at creation. Some — like the size scale and the event metadata — are immutable once the market exists; others, like fee schedules, can be adjusted. The values themselves vary by market and by asset, and are not fixed across the exchange.

Summary

• Every market is a central limit order book with price-time priority, tick/lot gates, and a single shared collateral pool per account.
• The base instrument is the underlying perpetual (forward F), with an oracle-based mark and funding.
• A conditional market is a family of five linked books over one underlying and one event: the underlying perpetual, two conditional perpetuals (CY, CN), and two prediction binaries (BY, BN).
• A conditional perpetual fires and settles to the underlying if its branch wins, and voids — P&L zero, margin returned — if its branch loses. It carries leverage. It is not a loss when it voids.
• A prediction binary's price is the implied probability of its outcome; it pays $1 if it wins, $0 if it loses, $0 on a Void.
• The legs are tied by no-arbitrage relationships — F = p·CY + (1 − p)·CN, BY + BN ≈ $1, Δ = CY − CN, p = (F − CN)/(CY − CN) — that arbitrage enforces, not the engine.
• The impact Δ is driven by correlation and volatility; conditional legs are fair at their conditional means, not at F.
• A fully hedged box costs near-zero margin (the capital-efficiency thesis); a one-sided hedge is positioned for a single branch, so flatten or fully box before the event resolves.