Conditional-Market Margin

A conditional market is a family of linked order books built over one underlying perpetual and one binary event: the underlying perp, conditional perpetuals tied to the event's outcomes, and prediction binaries for those outcomes. This document explains the durable principles of how margin works across that family — how a fully hedged set of legs nets to almost no margin, how margin behaves as an event approaches resolution, and what a broken hedge costs.

It builds on the general cross-margin model — a single collateral pool, initial versus maintenance margin, and a worst-outcome solvency check — and focuses on what is specific to conditional markets. It is a conceptual and mechanism-design reference: it describes behavior in plain terms, not wire formats or APIs, and the principles below hold even as the engine's internals evolve.

A few facts to fix up front, because they drive everything below:

• Conditional perpetuals carry leverage. Each conditional leg has its own initial- and maintenance-margin rates, set when the market is created, both strictly positive. They are not full-collateral instruments.
• A losing conditional perpetual is voided, not zeroed. When the event resolves the other way, the losing conditional perp settles with profit-and-loss of exactly zero and its margin is returned. It is not a loss to $0.
• Prediction binaries carry no separate position margin. Their entire possible loss is already reflected in your equity, so charging margin on top would double-count.
• Funding never applies to the conditional family. Only the underlying perpetual (and other standalone perps) accrue funding.

How margin works across a conditional-market family

The defining feature of a conditional perpetual is that it lives a double life. It trades today, on its own order book, at its own conditional price — the market's estimate of the asset's price given the event. But it settles at resolution to the underlying's price, and only in its branch — if the other branch wins, it simply voids. So an open conditional position is a branch-contingent claim: worth something if its event fires, worth nothing (margin back) if it doesn't.

Because of this, the engine margins the whole family through a worst-outcome (scenario) solvency check. It considers the possible resolutions of the events your account touches and, for each, computes your equity and your margin requirement in that outcome:

• Your equity in that outcome starts from your cash balance, then adds the value each position would have if the world resolved that way. A conditional perp that fires contributes its gain or loss versus entry; a conditional perp that voids contributes exactly zero; a prediction binary contributes its real settled value relative to what you paid.
• Your requirement in that outcome charges margin only on the legs that actually fire. A voided leg requires nothing — it has no risk left.

Your account is solvent only if equity meets the requirement in every outcome — not on average, not netted across branches, but in each branch on its own. The exchange treats your account's published equity as the worst of these outcomes, because that is the one that binds. A gain that exists only in one branch can never be leaned on to cover a loss in another, because the two never happen together.

This is the key difference from naive margining. Marking a conditional perp's unrealized gain into equity and letting you borrow against it would let you over-leverage against a profit that vanishes the moment the event resolves the other way. The worst-outcome check forecloses that: a gain that only exists in the YES branch does nothing for you in the NO branch, where that leg has voided to zero.

Because the same worst-outcome check gates opening orders, withdrawals, and liquidation, all three use the same scenario solvency structure, each with its own margin rate. Opening reserves at the initial rate; liquidation protects you at the lower maintenance rate, leaving a buffer between the two.

What you're charged on each kind of leg

Leg	Position margin	Valued in an outcome at
Underlying perpetual	Standard perp IM/MM on notional	The oracle mark (fires in every outcome)
Conditional perp	IM/MM on notional, only in the outcome where it fires	Its settlement value if it fires; zero (void) otherwise
Prediction binary	None	Its settled value: it wins or it loses

Resting (unfilled) orders also reserve initial margin continuously, in the outcome where they would fire, and that reserve is released the moment the order fills or is cancelled. One useful corollary: cancelling resting orders frees the initial margin they were holding, which can pull an account back from the edge of liquidation. When an account is checked for liquidation, the engine cancels its resting orders and re-checks before closing anything — sometimes that alone is enough.

How conditional legs are priced and marked

Two prices coexist for every conditional perpetual, and keeping them straight resolves almost all confusion:

• Its own conditional price is what the leg trades at on its book — the conditional forward, the market's estimate of the asset's price given the event. This is the price you buy or sell at, and the realizable value of your position today.
• The underlying's oracle price is what the leg settles against at resolution, because the winning conditional perp cash-settles to the underlying's mark.

These two prices differ by the basis — the premium the conditional carries over (or under) the underlying because the event is expected to move the asset. If a rate decision is expected to push an asset up, the "asset-if-the-decision-cuts" leg trades above the underlying. That gap is a fair, priced premium, not a mispricing; it shrinks toward zero as the event becomes a near-certainty, because a sure thing has nothing left to be conditional on.

The conditional legs settle against the underlying perpetual's oracle — the manipulation-resistant anchor the risk engine relies on. There is no separate external feed for "the asset's price if the event resolves YES." Each conditional leg's own book provides the realizable exit price for display and trading, while settlement and risk anchor to the underlying's oracle.

A prediction binary needs no price feed to be risk-marked: it pays a fixed amount if it wins and zero if it loses, both known by construction. That is exactly why a filled binary carries no separate position margin — its full possible loss is already inside your equity. A resting binary order still reserves its potential loss, because it isn't in your equity until it fills.

Strike-style events price above (or below) the threshold by construction

For a threshold event — "does the asset close above K?" — the conditional legs behave in a way that surprises people new to them. If the YES branch fires, the asset is, by definition, above K. So the YES conditional perp legitimately trades well above the underlying's current price whenever the underlying is below K: it is pricing a world that cannot co-occur with today's spot. A NO leg mirrors this below the threshold. This is a fair-value property of how the leg trades: the structurally-high (or low) price of a strike leg reflects the world it prices, and is treated as a real, correctly-priced premium.

Offsetting legs and the near-zero-margin hedged box

The reason to build conditional markets out of perpetuals — rather than as standalone, fully-collateralized bets — is capital efficiency through offsetting. Within any single outcome, positions that share an underlying offset each other before margin is charged. A set of positions that is delta-flat in an outcome charges essentially nothing in that outcome.

The clean case is the hedged box: long the YES conditional perp, long the NO conditional perp, and short the underlying perpetual, in equal size. Walk the two outcomes:

• If YES: the YES conditional fires and the NO conditional voids. You're left with the firing YES leg (long) and the underlying short — equal and opposite on the same underlying. Net-flat, so margin is ≈ $0.
• If NO: the NO conditional fires and the YES conditional voids. You're left with the firing NO leg (long) and the underlying short — again net-flat. Margin ≈ $0.

In every outcome the box is delta-neutral, so it charges roughly zero initial and maintenance margin, versus the full per-leg margin you'd pay holding any one of these positions alone. That is the capital-efficiency thesis: a correctly assembled box ties up almost no collateral because it carries almost no residual risk.

Why is the box so cheap? Because its payoff is locked at entry in every branch. In the YES branch the firing conditional settles at the same underlying price the perp marks against, so the underlying's price cancels out and you're left with a fixed number — the difference between your entry prices. The NO branch likewise reduces to a fixed entry-price difference. Both numbers are known the instant you build the box; nothing about the eventual outcome changes them. With no residual price exposure, the correct margin is just enough to cover the worse of the two locked branches — which, for a well-priced box, is near zero.

A few important points:

• The whole box matters, not a partial hedge. A two-leg hedge — say a YES conditional plus a short underlying, without the NO leg — is only net-flat in the YES branch. In the NO branch the conditional voids and the underlying short stands as a directional position, carrying full perp margin in that outcome. The engine charges that branch accordingly. Only the complete three-leg box is ≈ $0 in every branch.
• Netting requires the same underlying. Offsetting works because both conditional perps and the underlying perp settle against the same asset. A "hedge" assembled across unrelated underlyings doesn't net.
• Prediction binaries net through equity, not delta. A conditional perp hedged with the opposite-side prediction binary offsets economically — the binary's settled value moves opposite the conditional — and is margined a touch more conservatively than a pure perp-based box.
• Markets are treated as independent. When you hold conditional positions across several events, the engine considers each combination of outcomes on its own terms. This is conservative by design: it holds margin against every combination rather than assuming events move together.

Margin behavior approaching resolution

As an event nears resolution, two things change and they pull in opposite directions.

The basis shrinks. The premium a conditional carries over the underlying is, in effect, the expected move scaled by how uncertain the event still is. As the market converges on an outcome, that uncertainty collapses and the conditional price converges toward the underlying. The leg you hold looks more and more like a plain position on the underlying.

The jump at resolution does not. The moment the event resolves, your conditional perp stops marking against its own book and cash-settles against the underlying's price at that instant. If the resolution is a surprise — the event fires while the underlying hasn't yet moved to reflect it — your settlement can differ from the conditional price you'd been marked at seconds earlier. Approaching resolution, the visible basis can be small while the realized settlement gap is not.

The margin consequence is direct: a complete box stays ≈ $0 risk through resolution because both branches are locked, while a partial hedge does not — in the branch where its conditional leg voids, the surviving underlying leg stands alone as a directional position carrying full perp margin, and the post-resolution risk sweep evaluates your account in that state. So the safe posture into a resolution is to hold either nothing or a complete box (see Settlement, Resolution & Void for the full mechanic).

Settlement is atomic, and what matters for margin is which legs survive it: the winning conditional perp cash-settles to the underlying, a losing or voided conditional perp returns its margin at zero profit-and-loss, and an unresolvable event voids both branches. That change in which positions still exist is precisely why the post-resolution picture, not the pre-resolution one, is what your account must be solvent against. Settlement, Resolution & Void is the canonical reference for how each leg resolves.

One more point about realizing conditional profit early. Before resolution, a conditional perp's unrealized gain is a branch-contingent claim, not withdrawable cash — closing it early converts that claim into a prediction-binary position whose cash value today is roughly the gain scaled by the event probability (see Settlement, Resolution & Void for the full mechanic). The margin engine is stricter than that realizable value by design: it credits a conditional gain as zero in the branch that binds and charges a conditional loss in full, never margining against the probability-weighted figure. You are always at least as well-collateralized as the cash you could actually extract.

Worked example — the hedged box

All numbers are illustrative. Take an underlying trading around $65, an event whose YES branch is expected to push the asset up, with a YES conditional leg at ~$68 and a NO leg at ~$64. Build the complete box: long 100 of the YES conditional at ~$68, long 100 of the NO conditional at ~$64, and short 100 of the underlying perp at ~$65.

• YES outcome: the YES conditional fires (long), the NO conditional voids, the underlying short stands. Firing-leg long + underlying short on the same asset → net-flat → margin ≈ $0. The locked payoff is the difference between your underlying-short entry and your YES-conditional entry, fixed at the moment you built the box.
• NO outcome: the NO conditional fires (long), the YES conditional voids, the underlying short stands. Again net-flat → margin ≈ $0, with a different but equally-fixed locked payoff.

In both outcomes the box is delta-neutral and its payoff is already determined, so the engine charges essentially zero initial and maintenance margin. Holding the same three legs separately would tie up the full per-leg margin on each. This is the capital efficiency that makes conditional market-making viable: a correctly assembled box reserves almost nothing because it risks almost nothing, and it stays that way straight through resolution because both branches are locked.

A partial hedge margins very differently. Long the YES conditional and short the underlying, with no NO leg, is net-flat only in the YES branch; in the NO branch the conditional voids and the underlying short stands alone as a directional position. Because solvency must hold in every outcome, that branch binds, and the account is margined as a directional short — which is why a partial hedge is the case to avoid carrying into a resolution (see Settlement, Resolution & Void for the full positioning rule).