Trading & Strategy in Theory

This doc is about how to think when you trade on Proof — how to express a view, how to build a position out of the family's linked books, how to reason about fair value and where edge could come from, and how to approach making markets here. It is a conceptual companion to the mechanics docs, not a trading playbook and not an SDK. There are no calibrated parameters, no ready-to-run formulas, and no recipes. The point is to give you a correct mental model so you can do your own work on top of it.

A few conventions carry through. A conditional market lets you trade an underlying asset conditioned on whether a real-world event happens. One event over one underlying creates a family of five linked order books: the underlying perpetual (the underlying perp), a conditional perpetual (YES) and a conditional perpetual (NO) — perps that pay out only if their branch wins and otherwise void (profit-and-loss zero, margin returned) — and a prediction binary (YES) and prediction binary (NO), $0–$1 instruments whose price is the market-implied probability of the outcome. We write the underlying's forward price F, the probability the event resolves YES p, the conditional-perp prices CY and CN, and the binary prices BY and BN. Two relationships recur: the forward identity F = p·CY + (1 − p)·CN, and the binary box BY + BN ≈ $1. Both are no-arbitrage relationships between fair prices, not rules the engine enforces — they hold because traders trade them.

Every number below is illustrative.

Expressing a directional or conditional view

Start with the basic question: what view do you actually have? Proof gives you instruments that separate views that are tangled together elsewhere, and the first skill is matching the instrument to the view.

A plain directional view on the asset. If you think an asset goes up (or down) and you have no particular event in mind, the underlying perpetual is the right tool. It is a standard linear perp: linear PnL in the asset price, margined and liquidated against an oracle-based mark, and it carries funding that tethers it to the underlying. Nothing exotic. Use it when your view is "the asset moves," full stop.

A view on whether an event happens. If your view is purely about the event — "this scheduled decision goes a certain way," "this threshold is breached" — and you have no view on how much the asset reacts, the prediction binary is the clean instrument. Its price is the market's implied probability, it pays $1 if it wins and $0 if it loses, it carries no funding, and a filled binary carries no separate position margin because its maximum loss is already fully reflected in your equity. A binary is the one instrument designed to be held through resolution: it expresses P(event) and nothing else.

A view on the asset, but only because of an event. This is the view Proof is built for and the one nothing else expresses cleanly. Suppose you believe an asset rallies if a particular catalyst fires, but you do not want exposure to the asset for any other reason, and you do not just want to bet on whether the catalyst fires. That is a conditional view, and the conditional perpetual is the instrument: a perp on the asset that only fires in its branch. If you are long the YES conditional perp and the event resolves YES, you hold the asset position at the conditional settlement price; if it resolves NO, the position voids — PnL zero, margin returned. You have isolated the asset's exposure to the one branch you care about, with no theta bleed and nothing to roll.

It is worth being precise about what each tool does and does not give you, because the instinct from other venues will mislead you:

• A spot or perp position gives you the asset's move from every driver, not just your catalyst.
• A prediction-market contract elsewhere pays a fixed amount on the event and gives you zero exposure to the size of the asset's reaction.
• An option gives you directional, time-decaying exposure regardless of why the asset moved, and it compresses on the calendar.
• A conditional perpetual isolates the asset's reaction to one event: you have exposure only in the branch you named, and only to the asset, with no expiry and no decay.

A worked feel. Suppose an asset's underlying perp trades at F = $100, and you believe that conditioned on a YES outcome the asset is really worth more than the market thinks. The YES conditional perp trades at, say, CY = $104. If you are right and YES resolves, that leg settles to the asset's then-current underlying price — your edge is the gap between where you bought conditional exposure and where the asset actually prints in the YES world. If NO resolves, you void: you are out only the fees you paid to enter and exit, but not a directional loss on the asset. You have bought "the asset, in the YES universe," and you only pay for it if that universe arrives.

One more degree of freedom: conditional perps carry leverage. They have their own initial- and maintenance-margin rates, both positive, and they are margined like a perp in the branch where they fire. Conditional instruments are not full-collateral instruments — they are levered positions, with leverage set conservatively, and you size them as you would any margined perp — see the risk section below.

Synthetic positions: hedges, the box, and conditional bets

Because the five books are linked by the forward identity and the binary box, you can combine them to manufacture exposures that no single book gives you. Three constructions matter.

Pairing a conditional perp with the underlying — a single-branch hedge

A common move is to take a conditional-perp position and offset it with the underlying perp to look delta-flat. Before resolution this looks hedged: a long YES conditional perp plus a short underlying perp of the same size has roughly zero net exposure to the asset's price.

The key fact is the void branch. The hedge holds in the branch where the conditional perp fires. In the other branch the conditional perp voids — it ceases to exist, PnL zero, margin returned — and your underlying-perp leg is then naked, fully exposed to the asset's price. A long-YES-conditional plus short-underlying position is hedged in the YES world and a bare short in the NO world.

This is a defining feature of how conditional families behave, and the risk engine prices exactly that branch (see the next section and the margin doc). The discipline follows directly: a single conditional perp paired only with the underlying is a single-branch hedge, not a both-branch hedge. Either you size that paired position against your post-void exposure deliberately, taking a directional stance in one branch, or you complete the box.

The box — a both-branch hedge

To be hedged in both branches you need both conditional perps. A box holds the YES conditional perp, the NO conditional perp, and an offsetting position in the underlying — sized so that whichever way the event resolves, the firing conditional leg and the underlying leg leave you flat. Because one and only one branch fires, you are covered in every outcome: in the YES world the YES leg fires and pairs with the underlying; in the NO world the NO leg fires and pairs with the underlying; the non-firing leg voids harmlessly each time.

The box is the structurally complete position, and it is the foundation of Proof's capital efficiency. Under the scenario-margin engine, a fully matched box's payoff is locked in at entry in every branch, so the engine recognizes that there is essentially no residual risk and charges close to zero additional margin on it — versus the full per-leg margin you would pay if you held the legs without the offset. A position that nets to roughly zero risk should cost roughly zero margin, and on Proof it does. (The mechanics of how the engine sees this — net-delta grouping across same-underlying legs, scenario by scenario — live in the cross-margin doc.)

Two things keep the box precise:

• A box built from the binaries depends on resolution. The "both binaries" version of the box relies on BY + BN ≈ $1, which holds when the event resolves YES or NO. An unresolvable event pays $0 on both binaries, so any box whose payoff depends on the binaries summing to a dollar is keyed to the event resolving one way or the other. Where an unresolvable outcome is genuinely plausible, build the box from the conditional perps rather than leaning on the binaries.
• A box is a box while it is matched. Resolution transforms it: the winning conditional perp cash-settles and the losing one voids, so the moment the event resolves you hold whatever the surviving legs net to rather than a box. Capital efficiency from boxing is a pre-resolution property.

Conditional bets and the cash-out path

If your view is genuinely conditional and you do not want to neutralize it, you hold a conditional perp outright. The thing to internalize is that a conditional position's value is universe-internal until resolution. An unrealized gain on a YES conditional perp is a claim that exists in the YES world — it is not cash you can withdraw, because it disappears if NO wins.

This shapes how you take profit before resolution. Closing a conditional perp early does not pay cash; it crystallizes your conditional gain into a firing-branch prediction-binary position, which you then monetize on the binary book at roughly the implied probability (see Settlement, Resolution & Void for the full mechanic). The practical consequence for strategy: early exit on a conditional position runs through the binary book. The deeper the binary side of the family, the more readily you monetize a conditional gain before resolution. Plan exits around where the liquidity actually is.

Thinking about fair value and edge

This section is a framework, not a recipe. It tells you what determines fair value so you can reason about where you might have an edge; it deliberately does not hand you a turnkey pricing model, because a ready-made formula would just be a way to get picked off (and pricing models are proprietary besides). Treat every relationship below as structural.

The three things that price a family. A conditional family is pinned down by three quantities:

• F, the underlying forward — the asset's price across both worlds.
• p, the probability the event resolves YES — which is exactly what the YES prediction binary's price is telling you.
• Δ = CY − CN, the impact — how far the asset's expected price differs between the YES world and the NO world.

From these, the conditional-perp fair prices follow mechanically: CY = F + (1 − p)·Δ and CN = F − p·Δ. Equivalently, each conditional perp is fair at its conditional expectation — CY = E[asset | YES], CN = E[asset | NO] — not at the forward F. This last point is the single most important fair-value idea on Proof. A naive quote would put both conditional legs at F, the all-worlds price. But a conditional perp pays out only inside one world, and inside that world the asset's expected price is shifted by Δ. Quote both legs at F and you have left a free option lying on the book: a sharper trader lifts whichever leg is mispriced toward its true conditional mean and you are the one who paid for it. Fair value for a conditional leg is its branch-conditional expectation, period.

Where each input comes from, and which one is "yours." The two pieces of information the family's books genuinely pin down are the event marginal (p, from the binary) and the two conditional means (the first moments, from the conditional perps). They do not pin down the full joint distribution or higher moments.

• p is sourceable. Many events Proof conditions on have an external probability you can observe or estimate independently. Because p is observable, warehousing a view on p is, at bottom, a directional bet on the event — and if your "conditional edge" disagrees with the rest of the world mostly through p, you are really just taking a naked directional position dressed up as relative value. Be honest with yourself about that. The clean instrument for a p view is the binary, not the conditional perp.
• Δ is the Proof-native quantity. Δ is the reaction — how much the asset's expected level moves between the two worlds — and it is not quoted anywhere else, because no other venue prices conditional-on-event asset levels. Structurally, Δ is driven by how informative the event is about the asset: the more the event and the asset co-move (correlation ρ) and the more volatile the asset is over the horizon (volatility σ), the larger the impact. For a threshold event — "the asset closes above K" — the conditional means are governed by truncated-distribution math: the YES-conditional price must sit above the strike by construction, because if "above K" is the world you are in, the asset is above K by definition.

That structural statement — impact grows with correlation and with volatility — is as far as this doc goes. The proportionality constant, the way ρ and σ are estimated, the regime handling, and everything else that turns "Δ scales with ρ·σ" into a tradeable number are exactly the parts that constitute edge, and they are proprietary. The framework tells you which direction Δ should move when correlation or volatility changes; turning that into a price is your work.

Cross-checking and where edge lives. Two consistency checks fall out of the identities and are worth running constantly. First, the implied probability backed out of the conditional perps, p = (F − CN) / (CY − CN), should agree with the binary's price; a persistent disagreement is either an opportunity or a sign your model is wrong. Second, the forward identity F = p·CY + (1 − p)·CN should hold across the family. When these break, something is mispriced — but be disciplined about decomposing what. A gap can be a p disagreement (a disguised directional bet), a Δ disagreement (a genuine view on the reaction's size), or a stale leg. The edge worth having on Proof is on Δ — the reaction — because that is the quantity the venue is uniquely positioned to price and that informed flow elsewhere is not already trading. Edge that turns out to live entirely in p is edge you could have expressed more cheaply, and more honestly, with a binary.

Adverse selection, and why making markets here is hard

If you intend to provide liquidity rather than take it, the central fact to absorb is that the dominant cost of market-making here is adverse selection, not fees. On this venue fees are a small flat taker and maker charge per fill, the same across market kinds. The maker side is a charge rather than a rebate, so you are paid by the spread you capture, and that spread has to cover the losses you take to traders who know more than you at the instant they hit you. That buffer is the whole game.

Why adverse selection bites particularly hard on Proof:

• You quote against an oracle you cannot move. Perp marks are driven by an external, push-based oracle. A market maker who posts a tight two-sided quote and does nothing else will systematically be lifted on the side the oracle is about to move toward — picked off by flow that is reacting to information faster than a passive quote can. A quote that is not informed by some view on where the underlying is going is a quote that donates to informed takers. The defensive posture is to quote a base spread wide enough to survive being wrong, and to skew or pull on the side where you have reason to believe flow is toxic — never to sit blindly near the touch.

• The conditional legs reward an informed quote. Because each conditional perp must be fair at its branch-conditional mean rather than at F, quoting it correctly requires a live view on Δ and p simultaneously. Quote a conditional leg lazily — at F, or with a stale Δ — and you are handing a free option to anyone with a better estimate of the reaction. The no-arbitrage identities that tie the five books together hold because traders trade them, and as a maker you are often the trader doing so. That means you quote all the legs from one coherent (F, p, Δ) view so that no single leg can be lifted in isolation against the others. The identities pay you defensively when you quote the whole family coherently, and erode your edge when you quote leg-by-leg.

• Quote management is part of the edge, not an afterthought. The matching engine processes maker quote-refreshes (cancels and amends) ahead of incoming taker flow within a block, which gives a disciplined maker a real chance to pull stale quotes before they are hit. Use it: refresh by amending or cancelling, not by leaving stale orders resting while you place new ones. And remember that resting orders reserve margin continuously; a deep one-sided ladder pins capital whether or not it ever fills. Liquidity provision is a balance between depth (which earns spread but accumulates adverse-selected inventory) and width (which protects you but earns less), and the right balance depends on how informed the flow against you is.

The summary for a prospective market maker: your P&L is spread captured minus adverse selection minus fees, and on this venue the middle term dominates. Tight, alpha-blind quoting loses money even at low fees. Profitable making here looks like a defensible base spread, a coherent view across the whole family, fast quote management, and inventory discipline — not a thin two-sided quote left to run.

Risk discipline: leverage and resolution timing

Two risks on Proof are specific enough that they deserve their own discipline, on top of ordinary leverage management.

Leverage. Everything that can be levered is levered against your single cross-margin pool: the underlying perp, and the conditional perps too. The binding solvency check is the scenario-margin engine, which requires your account to be solvent in the worst outcome across every conditional market you touch — not on average, and with no credit travelling between branches. Two consequences for sizing. First, leverage is real on conditional legs; the conservative default does not make them collateralized, it just sets a sane starting point, and you can size yourself into trouble exactly as on any perp. Second, your true exposure is the worst-branch exposure. A position that looks flat on a net basis can be badly underwater in one branch — most commonly the void branch, where a conditional-perp "hedge" disappears and leaves the underlying leg naked. Size to survive the worst branch, because that is the branch the engine — and reality — will hold you to. (One useful and sometimes-overlooked lever: cancelling resting orders frees the margin they reserve and can by itself pull an account back from the edge, because the requirement includes reserved order margin, not just position margin.)

Resolution timing — flatten or fully box before resolution. This is the single most load-bearing operational discipline on Proof, and the rule is simple and absolute: do not carry an unbalanced conditional position into resolution. The account's risk picture changes discontinuously at the instant the event settles — a position that was hedged a moment before can be naked a moment after, because the leg that was doing the hedging has voided away — and you must be solvent in that post-resolution world before the event resolves (see Settlement, Resolution & Void for the full mechanic). The strategic implication is what matters here: either flatten the position entirely beforehand, or hold a fully matched box that stays solvent in every outcome, so there is no naked leg to be caught by. Treat the run-up to resolution as a flatten-or-complete deadline — that is standard operating discipline on a conditional venue.

A note on backstops. Below the level of any individual account, Proof has a layered loss-absorption design — a protocol vault, per-pool insurance, a capped socialized-loss step, and auto-deleveraging as a final layer — that exists so that a single blowup does not become everyone's problem, and pools are isolated so a failure in one cannot drain another's insurance. This is the venue's risk architecture; your own posture is to manage risk so that you are never the account the waterfall has to clean up after, and never the profitable counterparty that auto-deleveraging reaches for. Being flat, or being properly boxed, is what keeps you out of every one of those paths.

---

The through-line across all of this: Proof separates views that are bundled together elsewhere — direction, event probability, and the asset's reaction to the event — and gives you a distinct instrument for each, plus a margin engine that rewards you for holding genuinely offsetting combinations. The edge worth pursuing is on the reaction, Δ, the one quantity the venue prices that others do not. The risks worth respecting are the void branch and the resolution instant, where positions that look hedged quietly stop being hedged. Get the instrument-to-view match right, quote the whole family coherently if you make markets, and never carry an unbalanced conditional position into the moment it resolves.