Use case · Arbitrage

Exploit mispricings. Lock in the spread.

Polymarket sometimes lets YES and NO trade for less than $1.00 together. Buy both, the market pays you $1.00 at resolution - minus fees and slippage. Technically: this is sum-to-one arbitrage, executed against the CLOB matching engine with sub-500ms latency. We size positions against orderbook depth so you don't move the price against yourself.

$40M+
Aggregate arb profit on Polymarket, Apr 2024-Apr 2025
src: public on-chain analysis
210 /mo
"polymarket arbitrage bot" US search volume
src: DataForSEO May 2026
<500 ms
Median signal-to-fill latency on CLOB
src: Predikted production runs
0.98
Typical YES+NO floor we'll fire on (2¢ edge)
src: our default config
How it works

Three flavors of arbitrage we ship.

Same machinery, different signals. Pick the one that matches your capital and risk profile.

01 · SUM-TO-ONE

YES + NO < $1.00 on the same market. When the orderbook sells YES at 49¢ and NO at 49¢, buy both - guaranteed $1.00 payout at resolution. Edge: typically 1-3¢ per dollar, holding period: days to weeks.

02 · CROSS-MARKET

Same outcome priced differently across two Polymarket markets. Less common but higher edge when it appears. Requires market-mapping logic to spot.

03 · CROSS-PLATFORM

Polymarket vs Kalshi on the same event. Two regulated venues, two prices, one truth. See the dedicated page →

Config

What the strategy looks like in code.

Configurable thresholds. Auditable runs. If a trade fires, the matching log line is in the dashboard.

scanner.yaml config
# Sum-to-one arbitrage scanner - example client config
strategy: "sum-to-one"
markets:
  categories: ["politics", "crypto", "economics"]
  min_volume_usdc_24h: 25_000     # skip illiquid books
  min_days_to_resolution: 1

trigger:
  min_edge_cents: 1.5             # YES.ask + NO.ask <= 0.985
  min_orderbook_depth_usdc: 500   # both legs combined

execution:
  legs_in_parallel: true            # YES and NO fire simultaneously
  slippage_bps: 50
  partial_fill_policy: "unwind"     # if only one leg fills, exit

risk:
  max_position_usdc: 5_000
  max_concurrent: 15
  daily_drawdown_stop_usdc: 800
Honest framing

Things to know before you wire funds.

i
Edges shrink as more bots run.

The 2¢ floor of 2024 was 1¢ in late 2025. We size strategies for what's actually available today, not what was profitable two years ago.

i
Partial fills can flip a winning arb negative.

If your YES leg fills at 49¢ but NO bounces to 53¢ before your second leg lands, you're long YES with no hedge. Our default policy unwinds the partial fill - but that costs ~30bps. Bake it into the math.

!
Resolution risk is real.

Polymarket resolution disputes are rare but they happen. A 2¢ edge held for 3 weeks pays roughly the same as a 1-day USDC yield. If you're not getting paid for the time risk, don't trade.

×
Gas is a fixed cost. Small positions die.

Polygon gas is cheap (~$0.005/trade) but you still pay it on both legs. Below ~$200 position size, gas eats your edge. We enforce minimum sizes in code.

Starting points

Markets we've seen arb most often.

Generalizations, not promises. Edges rotate.

Politics +1.6¢ avg edge

Long-tail elections (parliamentary, governorships)

Long-tail elections (parliamentary, governorships). Liquidity asymmetry between YES/NO sides creates floors.

BTC 15-min markets +0.8¢ avg edge

Fast-resolving binary markets on BTC price thresholds

Fast-resolving binary markets on BTC price thresholds. Tighter but more frequent.

Sports tail markets +2.1¢ avg edge

Niche outcomes with thin books

Niche outcomes with thin books. Watch fees - Polymarket's 2% on certain sports markets eats arbs.

Budget bracket

Where this typically lands.

Solo strategy
$8k-$18k · 3-4 weeks

Sum-to-one scanner with risk caps. Good first arb bot.

  • One strategy (sum-to-one OR cross-market)
  • Polygon-only execution
  • Discord/Telegram alerts
  • Backtest + 2-week paper
  • 30-day warranty
Get started

Want this running on your capital?

Send us your capital range and target edge. We'll come back with a strategy spec, backtest results from public data, and a quote.