How Trading Works

This is the practical, plain-language tour of trading on Proof: positions and orders, margin and leverage, funding, liquidation, how a conditional market resolves, where prices come from, and one rule of thumb that keeps you out of trouble around resolution.

It is a conceptual reference — behavior in plain terms, not wire formats, APIs, or code. Other docs go deeper on each topic; this one is the orientation. Every worked number below is illustrative.

A quick note on instruments, since they come up throughout. A conditional market lets you trade an underlying asset conditioned on whether a real-world event happens. Each one is a family of five linked order books built from one underlying perpetual and one binary event: the underlying perp, a conditional perpetual (YES) and conditional perpetual (NO), and a prediction binary (YES) and prediction binary (NO). The four event-driven books are the conditional legs. You trade each book on its own. The rest of this doc assumes you know that much; the overview and settlement docs cover the family in full.

Placing a view

Every market on Proof is a central limit order book with price-time priority — the same model as a standard exchange. You express a view by buying (going long) or selling (going short), and you do it with one of two order styles.

• A limit order names the worst price you will accept. If part or all of it can't fill right now, the remainder rests on the book as a standing bid or offer at your price, waiting for someone to trade against it. While it rests, you are a maker: you provide liquidity, and anyone who fills you is a taker who pays your resting price. Orders always match at the resting (maker) price.
• A market order takes whatever liquidity is available now, walking the book from the best price outward until it is filled. You are always the taker. A market order has no price of its own, so it fills at the prices already resting against it.

Long vs short is just the direction of your position. Long means you profit if the price rises; short means you profit if it falls. On a perpetual there is no expiry forcing you out — you hold until you close, get liquidated, or (for a conditional leg) the event resolves.

A few mechanics worth knowing up front, because they shape how fills behave:

• What happens to the unfilled part of an order depends on its time-in-force. A standard resting order leaves its remainder on the book. Other styles drop the remainder instead of resting it, and a fill-or-kill style is checked up front and rejected outright if it can't fully fill against current liquidity — it never partially fills.
• Tick and lot gates. Each market has a minimum price increment (tick) and a minimum size increment (lot). Orders that aren't clean multiples are rejected. A market order skips the price-tick check (it has no price) but still must respect the lot size.
• Refreshing quotes is privileged over taking. Within a single block, maker quote updates — cancels and amends — are processed before new resting orders, which are processed before market orders. The practical upshot for anyone quoting: refresh your quotes with cancel/amend rather than place-then-cancel, and you get a small structural protection against being picked off by faster taking flow.

There is one account-wide consequence of placing an order that is easy to miss: a resting order reserves margin. It continuously holds aside the initial margin for the position it would create if filled, exactly as if you already held that position. That reservation is released the moment you cancel — which, as you'll see under liquidation, can matter.

Margin and leverage in plain terms

Proof is cross-margin on a single collateral pool. You have one account, one collateral balance, and every position you hold draws on that same pool. There are no isolated per-position buckets — a gain on one position is available to support another, within the rules below.

Four numbers describe your account:

• Balance — your settled cash. Deposits, withdrawals, realized profit and loss, fees, and funding all move it.
• Unrealized PnL — the mark-to-market gain or loss on your open positions. For a long, (mark − entry) × size; for a short, (entry − mark) × size.
• Equity — balance plus unrealized PnL. This is what actually backs your risk.
• Free margin — the equity left after covering what your positions and resting orders already require. This is what you can deploy or withdraw.

Two requirements govern any position:

• Initial margin (IM) is what you post to open it (and to rest an order). It is the larger, up-front requirement: you can't put on risk your free margin doesn't cover.
• Maintenance margin (MM) is the smaller, ongoing requirement to keep it open. If your equity falls below MM, the position becomes eligible for liquidation.

Both are a fraction of notional (notional = mark price × size). A market sets an IM rate and an MM rate, with MM always at or below IM. The gap is your buffer: you open at IM and have room to lose down to MM before liquidation.

A worked feel for it. If a market's IM rate is 10% and its MM rate is 5%, opening one unit of a $50,000 instrument needs $5,000 of free margin (IM), and the position survives until the equity backing it falls to $2,500 (MM). Those rates are illustrative.

Leverage is just the inverse of the IM rate — 10% IM is 10×, 5% is 20×. Lower margin means higher leverage and a bigger equity swing per unit of price move. You can choose to be more conservative than a market's default by setting a tighter personal IM level; this only ever raises your initial requirement (lowers your leverage) and never loosens anything or changes the maintenance requirement.

Two Proof-specific points:

• Conditional perpetuals carry leverage too. A conditional perp — a perp on the underlying that pays out only if its event resolves its way, and otherwise voids — has its own positive IM and MM rates and is margined like a perp in the branch where it fires. It is a margined position, not a full-collateral instrument. Leverage on conditional markets is set conservatively by default.
• Prediction binaries carry no separate position margin. A binary's maximum loss — it settles to $0 if it loses — is already fully reflected in your equity under each outcome, so charging margin on top would double-count. They are accounted for through equity directly.

The worst-case check across outcomes

A perp-only account is simple: equity must stay above the sum of each position's requirement, all marked at one price. Conditional markets complicate this, because a conditional position's value depends on how its event resolves — a gain on a YES conditional perp is real if the event resolves YES and vanishes if it resolves NO. You can't be allowed to borrow against a gain that can disappear.

Proof handles this with a scenario-margin check. Instead of valuing your account at a single point, the engine enumerates every combination of YES/NO outcomes across the conditional markets you touch and requires that you stay solvent — equity at or above requirement — in every one of them. The binding constraint is the worst outcome. No credit travels between outcomes: you can never lean on a profit that some branch erases, and the equity figure shown for your account is the equity in its worst scenario. (The cross-margin doc covers this in depth.)

The headline payoff of this model is that fully hedged, offsetting positions net to roughly zero margin. A position that nets to no directional exposure in every outcome charges almost no IM or MM — which is what makes building strategies out of conditional legs capital-efficient. A bare directional leg pays full margin; a properly offset one pays almost none.

Funding in plain terms

Funding applies to perpetuals only. The entire conditional family — both conditional perps and both prediction binaries — is excluded and never accrues funding. (The reason: a conditional leg's price is tethered to its event and its eventual settlement, not to a continuously-running underlying, so the expiry-replacement that funding provides doesn't apply.)

For a perpetual, funding is the force that replaces expiry. A standard future converges to spot at expiry; a perpetual never expires, so on its own its book could sit persistently above or below the underlying. Funding fixes that with a periodic payment that flows directly between longs and shorts — never to or from the exchange — sized to how far the perp's traded price has drifted from the underlying.

• When the perp trades above the underlying, longs pay shorts. Holding the crowded side costs a little each period; the side pushing the price back down gets paid.
• When the perp trades below, shorts pay longs. The incentive reverses.

The payment is zero-sum at the market level — every dollar a long pays, a short receives — and the exchange takes no cut. The further the perp drifts, the larger the premium and the larger the payment, so the pressure scales with the size of the dislocation.

The premium that drives funding is measured between two prices: the oracle (the external, manipulation-resistant reference for the underlying — the price funding pulls the perp toward) and the funding mark (the perp's own recent traded price, a moving average of where the book has actually been filling — the price funding measures from). The premium is the gap between them as a fraction of the oracle. It is the raw gap, not a blend of the two and not halved before use; positive when the book sits above the oracle (longs pay), negative below (shorts pay). Each market caps the premium symmetrically so a single runaway period can't blow up, and before any trade has happened the funding mark just equals the oracle, so the premium is zero.

Funding settles on a fixed cadence and lands in your free cash balance — it is never folded into your entry price or position size, so it shows up as a clean cash flow and doesn't silently move your liquidation level around. The amount you pay or receive scales with your position's notional. If your side's premium is owed against you, your balance is debited; if it's owed to you, you're credited, and the receiving side is always paid in full.

Liquidation: what triggers it and what happens

The trigger is the maintenance-margin check — the same worst-case predicate that gates your orders and withdrawals. When your equity falls below your MM, your account is undercollateralized and eligible for liquidation. For an account holding conditional markets, MM is evaluated across every YES/NO outcome, and breaching MM in any single outcome — even one you think is unlikely — makes you eligible. The trigger is the worst branch, not the average; a position that looks healthy on today's marks can still be liquidatable if one resolution outcome would leave it underwater.

Three things about what happens matter most:

Cancelling resting orders can save you — and the engine tries it first. Your requirement includes the IM reserved by your resting orders, not just your open positions. So the first thing liquidation does is cancel all your resting orders and re-check. Freeing that reserved margin lowers your requirement, and that alone can bring you back above MM — in which case no positions are closed. If you're near the edge, cancelling resting orders yourself is a real lever: it reduces your requirement, not just your exposure.

Closing happens at the reference (oracle) mark — the book is not swept. This is the most commonly misunderstood point. Liquidation does not fire an aggressive order into the book, eat the depth, and pay the slippage. The position is closed directly at the oracle-based reference mark, and the resulting profit or loss is realized against that price. The reasons: no slippage spiral (a forced sale into a thin book would deepen the loss and could cascade), much less exposure to spoofing-induced liquidations (an oracle-based mark is far harder to move with a single manipulated print or a momentary book wick than a book-based mark would be), and no "liquidation flow" sitting in the book for others to pick off at a discount — there is none.

It runs at end of block, as a sweep. Each block, after regular transactions, the engine walks accounts in a deterministic order and liquidates any that fail the maintenance check. If a mid-block fill pushes you under MM you may see a margin warning, but that warning is observability only — it closes nothing. The actual liquidation is the end-of-block sweep, which gives you the rest of the block to add collateral or reduce risk before it runs.

The engine also supports partial liquidation — closing your largest position first, re-checking after each, and stopping the moment you're solvent — configured per market; other markets close a triggered account in full. Either way, the close prices are reference marks, not book executions.

If a position is closed at a loss larger than your collateral can cover, the leftover deficit is absorbed not by your counterparties' fills but by an ordered backstop waterfall built to keep the exchange solvent — protocol-level reserves first, then a per-pool insurance fund, then a capped socialized loss, and finally auto-deleveraging that closes profitable counterparties at the bankrupt trader's bankruptcy price. Pools are isolated, so a blowup in one can't drain another's reserves. The liquidation doc covers it in full.

Settlement and resolution in plain terms

A perpetual has no settlement event — you realize your PnL when you close it, full stop. Settlement is a conditional-market concept: it is what happens to the four conditional legs when their event finally resolves. An event has exactly three dispositions — YES (it happened), NO (it didn't), or Void (it could not be resolved) — and resolution disposes of all five books in the family at once, while the underlying perpetual keeps trading normally afterward as the standalone perp it always was.

The few properties that change how you trade around resolution:

• The winning conditional perpetual cash-settles at the underlying's mark, and the losing one voids — and a voided position has zero profit-and-loss with its margin returned. It is not a loss to $0. This is the defining, most-misunderstood property of conditional markets.
• The two prediction binaries pay $1 / $0 through any YES or NO outcome, which is why their prices sum to about $1 while trading (the binary box). A Void is the exception: both conditional perps void and both binaries pay $0, so that relationship breaks. Void is never automatic — the safe assumption whenever you hold conditional legs is that any of the three outcomes can land.
• A conditional perp's unrealized gain is branch-contingent — only real in the branch where the event fires — so until resolution it isn't withdrawable cash. You don't have to wait, though: you can close a conditional leg on its own book any time it's trading, and there's a built-in early-exit path that lets you monetize that branch-contingent gain through the prediction binaries.

See Settlement, Resolution & Void for the full per-leg payoff, the early cash-out path, and exactly how each outcome settles.

Where prices come from

Two ideas govern every price the engine cares about.

First, there is an external oracle, not the order book, behind anything risk-bearing. Every underlying perpetual is anchored to a push-based oracle: signed price updates are pushed into the exchange (it doesn't poll an endpoint or read its own book), each carries a strictly-increasing publish time so a stale price can never overwrite a fresher one, and oracle updates are processed at the highest priority in a block, ahead of any trading — so you never get filled or liquidated against a price the block hasn't seen yet. This oracle is the single source of truth for what an underlying is worth. It also decides event resolution for price-question events: for a threshold event ("did the asset close above K?"), the exchange reads the oracle at resolution and computes YES or NO directly, and re-checks any asserted outcome against the oracle, rejecting it on a mismatch.

Second, the engine maintains two distinct marks per market, on purpose, and they can diverge:

• The risk mark drives margin, equity, unrealized PnL, liquidation, and conditional-market settlement. For a perpetual it is the oracle price — not the last trade, not the book mid. That's the deliberate anti-manipulation choice: a single spoofed print or a momentary book wick is far harder to push into your margin or trigger your liquidation when the engine isn't valuing your position off the book.
• The funding mark drives the funding premium and nothing else. It is book-based: the order book's own moving average of recent trades, by design, because funding's whole job is to measure how far the traded price has drifted from the oracle.

The reason for two marks is that the two jobs have opposite requirements: risk valuation wants a price that's hard to manipulate and stable under hostile book conditions (the oracle), while funding wants a price that reflects where the market is actually trading (the book). Forcing both onto one number would either blind funding to real drift or expose margin to book manipulation.

The conditional legs have no oracle of their own. There's no external feed for "the asset's price if the event resolves YES" — an attempt to push an oracle update straight to a conditional leg is rejected. A conditional perpetual is margined and settled against the underlying perpetual's oracle, because that's the price its winning branch settles to; its own book price (its conditional forward) is a display/marking reference, not its risk mark. A prediction binary has a known $1/$0 payoff by construction, so it needs no feed to be valued for risk. The only oracle any conditional market consults is the underlying's.

The relationships between conditional prices — for example, that the underlying forward equals the probability-weighted blend of the two conditional prices, or that the two binaries sum to about $1 — are no-arbitrage relationships between fair prices, not rules the engine enforces. The margin engine never assumes the books are coherent; it simply requires you to survive each outcome.

The flatten-before-resolution rule of thumb

If you take one operational habit away from this doc, make it this: be solvent in the post-resolution world — flatten or fully box your conditional inventory before an event resolves. The trap is that a conditional perp plus an offsetting perp looks delta-flat today but only hedges in the branch where the conditional fires; in the other branch the conditional voids and disappears, leaving your perp leg naked with full directional risk. The scenario-margin engine already prices that void branch into your requirement — which is why such a position can be flagged for liquidation while it looks flat on today's marks.

So don't carry a one-branch hedge through a resolution: either flatten your conditional exposure before the event, or hold a fully matched box that nets to no exposure in every outcome, including Void. See Settlement, Resolution & Void for the full mechanic — how each leg settles and why being matched across all branches is the property that protects you.