🎯 Polymarket Multi-Outcome Arbitrage Scanner

No noise, just alpha. Concrete arithmetic arbitrage on Polymarket's multi-outcome events.

A signal engine that scans every Polymarket event with mutually-exclusive outcomes — election winners, championship finishers, "who will be the X" markets — and detects when the YES prices fail to sum to their mathematical expected value. That's free arbitrage. Every output is a complete basket trade with the math worked out, fees subtracted, and per-leg liquidity verified against live order books.

---

✨ Key Features

• Real arbitrage, not vibes — every signal is rigorous arithmetic. Σ YES across N mutually-exclusive outcomes must equal a known constant (1.0 for winner-take-all, K for top-K). Deviations are quantifiable edge.
• Two arb directions — buy_yes_basket (when Σ YES < expected) and buy_no_basket (when Σ YES > expected). The scanner picks the profitable side automatically.
• Conservative event classifier — three structures qualify (winner-take-all, top-K) and the scanner explicitly rejects cumulative "by date" events, price ladders, and prop bets. No false signal from non-exclusive outcomes.
• Fee-adjusted returns — every opportunity reports gross_return_pct, fees_pct (Polymarket's 4% round-trip), and the headline net_return_pct you actually capture.
• Real per-leg liquidity tests — walks each child market's CLOB order book at a configurable position size; flags all_legs_fillable: true/false/null per opportunity.
• Confidence-scored signals — every result tagged with a 0–100 signal_score weighing return, fillability, leg count, and time-to-resolution.
• Public-read only — no API keys required. Runs on Apify free tier.
• Async-first pipeline — concurrent event fetch + bounded-concurrency CLOB probes; full scan in 30–90 seconds.

---

📊 Sample Output

Each opportunity is a complete basket trade with the math worked out:

{
  "id": "lebanon-parliamentary-election-winner-buy_yes_basket",
  "event_title": "Lebanon Parliamentary Election Winner",
  "event_url": "https://polymarket.com/event/lebanon-parliamentary-election-winner",
  "resolution_date": "2026-05-31",
  "event_type": "winner_take_all",
  "expected_sum_yes": 1.0,
  "arb_type": "buy_yes_basket",
  "leg_count": 21,
  "sum_yes_price": 0.789,
  "deviation_from_one": -0.211,
  "fees_pct": 4.0,
  "gross_return_pct": 26.74,
  "net_return_pct": 22.74,
  "legs": [
    {
      "market_id": "0x7fc8...",
      "question": "Will Lebanese Forces win the most seats in the 2026 Lebanese parliamentary election?",
      "outcome_label": "Lebanese Forces (LF)",
      "side": "YES",
      "price": 0.06,
      "market_url": "https://polymarket.com/event/will-lebanese-forces-win-the-most-seats...",
      "liquidity_usd": 19892,
      "fillable": true
    }
  ],
  "liquidity": {
    "tested_usd_per_leg": 100,
    "all_legs_fillable": true,
    "fillable_leg_count": 21,
    "total_leg_count": 21
  },
  "signal_score": 85,
  "signal_label": "Pure arbitrage",
  "notes": null
}

A ScanSummary record is also written to the default key-value store under OUTPUT_SUMMARY:

{
  "scanned_at": "2026-04-29T18:00:00Z",
  "total_events_scanned": 1000,
  "eligible_events": 38,
  "pure_arb_count": 7,
  "ev_positive_count": 11,
  "avg_net_return_pct": 18.4,
  "best_opportunity_id": "lebanon-parliamentary-election-winner-buy_yes_basket"
}

Signal labels

Label	Score	Meaning
Pure arbitrage	≥ 80 + net return > 0	Risk-free spread after fees
Strong EV+ signal	60–79	High-conviction edge
EV+ with edge	40–59	Mild edge — position sizing matters
Marginal	< 40	Excluded by default

---

🧮 The Math

For a Polymarket event with N mutually-exclusive outcomes (e.g. "who wins the 2026 MLS Cup?", "which team finishes top 4 in Serie A?"), exactly K outcomes resolve YES. So under perfect pricing:

Σ P(outcome_i = YES) = K

• K = 1 for winner-take-all events ("X election winner", "Y nominee", "Z championship")
• K = N_top for top-K events ("Top 4 EPL finish", "Reach the final" with K=2, "Eurovision Top 5")

When Σ YES drifts from K, there's a basket trade with locked-in profit:

Buy-YES-basket arb (Σ YES < K):

• Cost: Σ YES dollars to buy 1 share of every YES outcome
• Payout: K dollars (exactly K of them resolve YES)
• Gross return: (K - Σ YES) / Σ YES

Buy-NO-basket arb (Σ YES > K):

• Cost: N - Σ YES dollars (since NO_i = 1 - YES_i)
• Payout: N - K dollars (the legs that DON'T resolve YES)
• Gross return: (N - K - (N - Σ YES)) / (N - Σ YES) = (Σ YES - K) / (N - Σ YES)

The scanner subtracts a conservative 4% round-trip fee (Polymarket's 2% taker × open + close) to compute net_return_pct. Every leg is liquidity-tested against the live order book at the configured position size before being marked fillable.

---

🛡 What This Scanner Will NOT Flag

The scanner uses a conservative event classifier that rejects structures where Σ YES has no constraint. These look superficially like arb but are not — buying their baskets has no guaranteed payout.

Rejected: cumulative "by date" events

"Will Drake release Iceman by April / May / June?" — releasing in April implies released by May too. Outcomes are nested, not exclusive.

Rejected: price/threshold ladders

"Will Solana reach $110 / dip to $70 / reach $100 in April?" — price can hit multiple thresholds. Outcomes are independent ranges.

Rejected: prop bets and multi-condition events

"Player X: Points O/U 17.5 + Rebounds O/U 4.5 + Assists O/U 3.5" — independent stat lines. "Will Trump publicly insult someone on April 26 / 28 / 29 / 30?" — independent daily events. "What demands will Trump agree to in April?" — multiple demands can co-resolve YES.

If you're seeing a Polymarket event you'd expect to be arb-eligible but it's not in the output, it likely fell into one of these categories. Check OUTPUT_SUMMARY.eligible_events to see how many of the day's events qualified.

---

🚀 How to Use

Method 1: No-Code (Apify Console)

1. Open the actor on Apify
2. Adjust min_signal_score and min_net_return_pct for your strategy
3. Click Start — opportunities land in the dataset, summary in OUTPUT_SUMMARY

Method 2: API Integration

curl -X POST "https://api.apify.com/v2/acts/seralifatih~pm-arbitrage/runs?token=YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "min_net_return_pct": 0,
    "min_signal_score": 70,
    "min_event_liquidity_usd": 10000,
    "liquidity_test_amount_usd": 250,
    "output_limit": 25
  }'

Fetch results:

$curl "https://api.apify.com/v2/datasets/DATASET_ID/items?token=YOUR_API_TOKEN&format=json"

Scheduled runs

Schedule the actor every 5–15 minutes during high-volatility windows (election nights, championship games, FOMC days, Eurovision, primary debates) to catch fresh basket mispricings before market makers correct them.

---

⚙️ Input Configuration

{
  "min_net_return_pct": -1.0,
  "min_signal_score": 30,
  "min_event_liquidity_usd": 5000,
  "min_liquidity_per_leg_usd": 200,
  "min_legs_per_event": 3,
  "liquidity_test_amount_usd": 100,
  "max_days_to_resolution": 365,
  "max_events_to_scan": 1000,
  "output_limit": 50
}

Field	Type	Default	Description
min_net_return_pct	number	-1.0	Minimum net return after fees. 0 = pure arbitrage only.
min_signal_score	integer	30	Minimum 0–100 signal score to include.
min_event_liquidity_usd	integer	5000	Skip events below this cumulative liquidity.
min_liquidity_per_leg_usd	integer	200	Each leg must have at least this liquidity (filters dead placeholders).
min_legs_per_event	integer	3	Skip events with fewer mutually-exclusive legs than this.
liquidity_test_amount_usd	integer	100	Per-leg position size for the depth test.
max_days_to_resolution	integer	365	Skip events resolving beyond this horizon.
max_events_to_scan	integer	1000	Cap on Polymarket events fetched. Higher = wider coverage, slower.
output_limit	integer	50	Cap on opportunities returned (≤ 200).

---

🧰 Tech Stack

• Python 3.11 — async-first
• httpx — concurrent Gamma + CLOB requests with bounded concurrency
• pydantic v2 — strict output schema
• Apify SDK for Python

Estimated run time: 30–90 seconds for the default config at 512 MB memory.

---

⚠️ Disclaimer

This tool is for informational and analytical purposes only. Prediction market trading involves financial risk. Always validate signals independently before committing capital. Not financial advice.

Important caveat — implicit "Other" probability: The arithmetic identity Σ YES = K assumes the listed outcomes cover all real-world possibilities. For winner-take-all events (e.g. "Alaska Governor 2026") with many candidates, Polymarket may not list every potential winner — the missing probability mass on unlisted outcomes is what often keeps Σ YES below 1.0 even on well-priced markets. Top-K events (Eurovision Top 5, EPL Top 4, Serie A Top 4) are typically more reliable because the outcome space is rigorously defined: exactly K teams finish top-K, no "other" possible. Single-winner events with >10% deviation deserve manual verification of outcome-space completeness before trading. The scanner reports the raw mathematical signal — you must judge whether the listed candidates exhaust the field.

---

🔗 You might also like

• Crypto Arbitrage Scanner — CEX spot arbitrage across 8+ exchanges
• CEX Funding Rate Arbitrage — Perpetual funding rate basis trades