Ranked by how often they’re asked, tagged by who asks them — from your very first night to the quant corner. Can’t find yours? The box at the bottom answers back.
No. You never pick winners. Prices come from the live Polymarket markets for each leg, and your margin sits on top. You are the casino owner, not the player — the table’s math works whether or not you watch the games.
You’re the other side of the gamble. The bettor buys a small chance at a big payout; you sell it — at a price above its mathematical fair value. Any single night can lose (that’s variance, and it’s real), but the average is priced in your favor. The math page works this out with real tickets.
Any amount — the shape matters more than the size. The recommended shape: no ticket risks more than ~5% of your bankroll, all open tickets together stay under ~30%, and any single team under ~12%. On $500 that’s a $25 ticket cap, $150 open, $60 per team — a worst-case night is bounded at 30%, survivable by construction.
There’s no universal number — margin is a market. Short parlays clear around a few hundred basis points; long ones carry 10%+ because margin compounds per leg. The practical answer: use follow-the-market mode, start near 75–90% of the winning median, and let your fill rate teach you. Filling everything? You’re too cheap. Filling nothing? Too proud.
When a bettor builds a parlay, Polymarket broadcasts a Request For Quote to every connected market maker. Each maker has 400 milliseconds to answer with a signed price. The bettor gets the best one. If it’s yours, you collect their stake instantly and the ticket is live.
No. When a ticket fills, its worst-case payout is locked from your wallet as collateral up front. The engine cannot quote money the wallet doesn’t hold — solvency is enforced in code, not by a setting you could get wrong.
It keeps quoting inside your rules, 24/7. Your caps bound every ticket and the whole book; anything unusual — error storms, accounting mismatches, markets moving strangely — halts the desk automatically and tells you why. And you never have to start live: watch mode shows real flow with zero risk.
You arm the desk before the evening slate. Requests stream in by the thousand; your engine skips most (out of band, wrong sport, too big) and quotes the ones inside your rules. Maybe a handful fill. You watch your exposure bars move, games end, tickets die or pay, and by morning settlement has swept the results back to your wallet.
You’re losing the auction — other makers are pricing under you. Open the pricing console and compare the MARKET column (what winning quotes clear at) to YOU. If you’re consistently above it, cut your undercut percentage or your manual bps. Filling is a choice about price, not luck.
Every skip has a recorded reason: outside your leg band, a team or game you turned off, fair value outside your price band, stale or missing market data, correlated legs, no room under a cap, or simply past the 400ms deadline. The Floor shows the reason on every skipped card — the engine never skips silently.
The product of live CLOB midpoints for each leg, fetched with sub-second freshness requirements and hard staleness gates (a quote never prices off a mid older than ~2s). Leg identifiers are mapped from RFQ position ids to CLOB token ids via the markets catalog. No de-vigging or consensus blending yet — mids are the source, and anything unresolvable is a skip, not a guess.
We don’t — we refuse it. Legs sharing a game or a team violate the independence assumption behind product-of-mids pricing (naive pricing undersells those combos by 30%+), so the engine declines them outright. Declining is free; a copula you can’t calibrate on thin data is not.
The public trade tape: 24-hour winning-fill margins per leg count, medians computed with our own fills excluded (echo-chamber guard), re-calibrated every few minutes. Your undercut percentage prices against that median; thin buckets fall back to a compounded per-leg curve; your floor applies always. The console shows the engine’s live calibration, not a web-side estimate.
In your own wallet on Polygon. Stakes land in it the moment a ticket fills; collateral locks on-chain while games play; when tickets resolve, winnings sweep back automatically. There’s no balance sitting with us.
One switch. The engine stops quoting within about a second; quotes already sent expire on their own within seconds; open tickets simply settle when the games end. Stopping never strands anything.
Whatever your caps and price allow — a tight cautious book might fill 3–5 a day; a wider one dozens. Fills aren’t the goal; positive-EV fills are. One well-priced fill beats ten at fair value.
When something’s moving that prices haven’t caught — a scratched starter, weather, lineup news — or when your exposure to that team’s cluster is already at your comfort line. Turning a team off blocks every parlay containing them, any game, instantly.
Skew when you want to ride the market at an offset — ’always +2¢ on this outcome.’ Pin when you believe a number the market hasn’t priced yet and want it fixed. The board shows every pin’s age and drift against the live market so a stale opinion can’t hide.
On-chain settlement costs — small. The desk’s first real winning ticket paid roughly $0.41 in total fees on a $65.73 payout. Pregame’s own economics run on a builder code attached to fills, not a cut of your bankroll.
The submission window is 400ms wall-clock from broadcast. Requests are evaluated concurrently at socket arrival against cached market data; only the money tail (reserve → sign → send → record) serializes. Typical decision-to-send lands well inside the window — the desk’s first fill was answered in 155ms.
All money is integer micro-dollars (e6 fixed point), no floats anywhere in the money path. For a BUY-notional request: size = floor(notional·10⁶ / price); collateral locked = ceil((10⁶ − price) · size / 10⁶). Every division carries an explicit rounding mode and rounding dust always favors the maker. Partial sizing below the requested notional is allowed and used to fit your caps.
Layered, all fail-closed: a mid-movement gate (if a leg moved hard in the trailing window, someone knows something — skip), staleness gates, a favorites floor and fair-value band (longshot relative error is lethal), buy-only flow, per-team concentration caps, and calibration that excludes our own fills. Undercut pricing against the winning median also tempers winner’s curse: you win auctions on price you chose, not price you misjudged.