How Polymarket Arbitrage Bots Find Gaps—and Why Trades Fail

A Polymarket arbitrage bot looks for combinations of positions whose executable cost is lower than their combined settlement payout. Finding that discrepancy is not the same as capturing it: insufficient liquidity, partial fills, trading costs and mismatched settlement rules can leave the trader with an exposed position instead.

The useful distinction is between scanning, stake calculation and execution. This guide focuses on those boundaries; for the underlying odds and staking concepts, start with Arbitrage Betting Explained. All prices and trades below are illustrative examples, not live opportunities.

What does a Polymarket arbitrage bot actually do?

“Arbitrage bot” can describe three different functions. A tool may perform one without performing the others, so evaluate its actual workflow rather than assuming that an opportunity alert includes a completed hedge.

• Scanning identifies potentially related outcomes and price discrepancies, ideally using current executable quotes and available size.
• Stake calculation determines how much to buy or stake on each outcome so that the intended settlement payouts align.
• Execution submits orders, checks actual fills and manages any remaining exposure when the intended combination is incomplete.

For opposite shares in one binary market, the scanner can test whether buying both sides costs less than their combined payout. Across bookmakers and prediction markets, it must also compare outcome definitions and translate odds into compatible payout calculations. Neither test proves that both legs remain available when orders reach their destinations.

Why are displayed prices not enough to identify executable arbitrage?

According to the Polymarket Help Center, displayed prices generally represent the midpoint between the best bid and ask. When the spread exceeds $0.10, the displayed price uses the last traded price instead. These displays describe the market; they do not guarantee a purchase at that price.

For example, a YES bid of $0.46 and ask of $0.50 produce a displayed midpoint of $0.48. If the NO ask is $0.51, buying both sides costs $1.01 per pair before costs, not an apparent bargain based on the YES midpoint. A purchase calculation needs the offers available to buyers.

Size matters too. The best ask might cover only part of the intended order; the remainder may require higher prices. A useful scanner therefore distinguishes the top-of-book gap from the gap available at the trader’s full size. Polymarket says prices respond to real-time supply and demand, so a previously valid calculation also needs refreshing before execution.

How should the bot calculate shares and stakes?

Consider an illustrative basketball market with exactly complementary YES and NO shares. Assume its settlement terms pay $1 to the winning share and $0 to the losing share. An equal number of YES and NO shares would then pay $1 per pair, whichever side wins.

Suppose executable asks are $0.48 for YES and $0.49 for NO, with enough depth to buy 100 shares of each. The YES leg costs $48, the NO leg costs $49, and the combined outlay is $97. The settlement payout is $100, leaving a theoretical $3 surplus before costs.

For this equal-share example, net settlement surplus equals q × (1 − pYES − pNO) − C, where q is the number of shares bought on each side, pYES and pNO are actual average purchase prices in dollars, and C is total additional costs in dollars. The calculation assumes both legs fill and the stated settlement payout applies.

A $194 purchasing budget would support 200 pairs at those unchanged prices before reserving costs, but only if both books offered that depth. An illustrative $1 total cost allowance reduces the 100-pair surplus from $3 to $2. Across different payout structures, equal stakes may not equalize returns; use the arbitrage calculator to split stakes, then check the available size and applicable costs separately.

What happens when only one leg fills?

Partial fills turn a balanced calculation into a directional position. In the same example, suppose all 100 YES shares fill at $0.48, but only 40 NO shares fill at $0.49 before the remaining offer disappears. The trader has spent $67.60 and holds only 40 complete pairs, plus 60 unhedged YES shares.

• If YES wins, the position pays $100 and produces $32.40 before costs.
• If NO wins, the position pays $40 and loses $27.60 before costs.
• The 40 paired shares provide a balanced payout, but the remaining 60 YES shares determine the outcome-dependent exposure.

An execution system needs to reconcile filled quantities, not merely report that both orders were submitted. Its contingency could be to seek the missing hedge within a defined price limit or reduce the exposed position. Neither action guarantees the original surplus. For the separate problem of designing exits around depth and failed fills, see the Polymarket stop-loss bot guide.

How can settlement rules invalidate an apparent hedge?

Two markets can refer to the same game without being complementary contracts. A basketball selection described as “winner” is not enough information to establish matching settlement. The bot—or the trader reviewing its result—must compare the actual market rules rather than matching names alone.

• Event identity: do both selections refer to the same fixture and occurrence?
• Game scope: do both contracts include the same periods and treatment of overtime?
• Outcome boundary: are the winning conditions exhaustive and mutually exclusive, including any draw or push treatment?
• Exceptional outcomes: what does each venue specify for postponement, cancellation, abandonment or a void?
• Resolution basis: do the contracts use compatible result definitions and resolution criteria?

These are checks, not claims about any venue’s specific rules. If one leg can be refunded while the other remains active, the original payout calculation may no longer hold. Related totals or handicaps can also have multiple settlement paths. A combinatorial scanner must evaluate the complete payoff map, not just add attractive-looking prices.

What does research show about the execution window?

The study Arbitrage Analysis in Polymarket NBA Markets reconstructed market states from more than 75 million order-book snapshots across 173 games. It reported only seven executable single-market in-game episodes, with a median duration of 3.6 seconds. Those findings describe the analyzed games, not a forecast for every sport or future market.

The same study reported 290 active combinatorial episodes, concentrated in the final minutes of live games. It also reported that 76.9% of combinatorial opportunities were constrained to an average executable size of 14.8 shares. More detected opportunities therefore did not mean deep liquidity or unrestricted trading capacity.

Execution risk is not unique to prediction markets. The NBER paper on foreign-exchange arbitrage describes negative-spread and three-currency opportunities appearing and disappearing within seconds, with no guarantee that all transaction orders fill within a fraction of a second. That is evidence from foreign exchange, not a Polymarket latency estimate, but it illustrates why detecting a gap and completing every leg are separate tasks.

What should a reliable execution workflow check?

1. Validate the payoff map. Identify the event, outcome definitions and settlement conditions before treating positions as hedges.
2. Refresh executable quotes. Use buy-side offers for purchases and available bids for planned sales, rather than displayed midpoint prices.
3. Calculate the size-dependent result. Account for the depth needed on every leg, applicable fees and a cost reserve.
4. Define a rejection point. Specify the maximum combined cost and the minimum completed surplus required before submitting orders.
5. Plan for incomplete execution. Decide how much unmatched exposure is acceptable and what happens if the remaining hedge becomes too expensive.
6. Reconcile the result. Compare actual filled quantities and average prices, cancel unwanted outstanding orders, and recalculate the remaining exposure.

This checklist is a design standard, not a claim that any scanner implements it. Pendle’s risk documentation, discussing a different arbitrage product, explicitly warns about unhedged legs and execution costs exceeding the spread. Its venue-specific assumptions should not be imported into Polymarket calculations, but the distinction between displayed opportunity and completed hedge remains relevant.

Where does OddsFantasy’s arbitrage finder fit?

The OddsFantasy Polymarket trading terminal lists live and pre-match sports prediction-market events and supports filtering and sorting by date, liquidity and volume. Market views show charts, the order book, available liquidity and the current spread. Those views help traders inspect the execution conditions behind a candidate discrepancy.

The OddsFantasy arbitrage finder lists opportunities across bookmakers and Polymarket, while its arbitrage calculator supports splitting stakes across outcomes. Position these as discovery and calculation tools. A listed opportunity is not a completed trade, and these verified capabilities do not establish atomic multi-leg execution or a bot that places bookmaker bets.

Which order tools help manage a candidate trade?

• Market orders are supported in the OddsFantasy trade modal and can carry optional take-profit and stop-loss settings. Those settings are exit instructions, not proof that an arbitrage hedge is complete.
• Limit orders enter only at the user’s chosen odds and remain open until matched, cancelled or expired. Price control does not mean the complementary order will fill.
• Quote orders place both sides of a market to target spread profit, calculating quote levels and the stake split from capital and a target profit percentage. This is a quoting workflow, not evidence of guaranteed arbitrage capture.

The OddsFantasy trading documentation explains those order types. Positions are coloured red when exposed, yellow when an open limit order may complete the hedge, and green when fully hedged. OddsFantasy can also merge opposite positions on the same market to lock in the spread and redeem resolved positions from the Positions page.

Open positions can be closed with a market close, a limit close at chosen odds or a scheduled time close, such as before kickoff. Scheduled time orders appear on the Orders page, where their execution time can be edited or the order cancelled. A scheduled close is an exit instruction, not simultaneous execution of all arbitrage legs.

The trading terminal is free to use; paid plans add advanced automation, the positive EV and arbitrage finders, and the AI research assistant. OddsFantasy is non-custodial: users connect their own wallet and funds stay in it. Neither the pricing model nor wallet custody removes execution or settlement risk.

What should you verify before acting on an alert?

• Does the alert use executable prices, and is there enough depth for every intended leg?
• Do the contracts cover every settlement outcome under compatible rules?
• Does the expected surplus remain positive after the actual stake split and all applicable costs?
• Who submits each order, and what happens if a leg is rejected, delayed or partially filled?
• Can you distinguish a discovered opportunity, an open order and a fully hedged position?

Judge a Polymarket arbitrage bot by those answers, not just the number of alerts it produces. Discovery tells you where to investigate; calculation tells you the intended position; execution determines what you actually own.