Concentrated in one market type - usually the 5-minute BTC Up/Down markets
Consistent position sizes - not random gut-feel bet sizing
Positions on both sides of the same market - the dead giveaway for arb bots
The three accounts
@first - directional momentum bot
19,923 predictions since October 2025. All positions I could see were in "Bitcoin Up or Down - 5 minute" markets, buying a single side (Up or Down, not both). Entry prices: 33,ドル 44,ドル 69,ドル 76,ドル 81,ドル 89¢.
This isn't arbitrage. It's a bot reading BTC price momentum and betting on the 5-minute direction. The 33¢ entry that returned 202% was almost certainly triggered on a strongly trending candle where the market was underpricing the obvious direction.
Total P&L: +1,484ドル on ~20k trades. That's thin per-trade, but it's been running for 8 months. Directional edge is slower to compete away than mechanical arb.
@second - merge/redeem arb, fully exited
16,454 predictions since April 2026. Current positions: zero. Portfolio value: 0ドル. Biggest win: 606ドル.
This is what a successful merge/redeem arb bot looks like after it's done. The strategy:
1. Scan binary markets where P(YES) + P(NO) < 1ドル.00
2. Buy both YES and NO simultaneously
3. Merge tokens into a complete set
4. Redeem the complete set for 1ドル.00
5. Pocket the difference (minus fees)
When YES+NO = 0ドル.95, you make 0ドル.05 per share bought. With large position sizes and fast execution, this adds up. bowbow16 ran 16k trades in ~2 months, extracted its edge, and is now sitting idle - probably because the spread has tightened as more bots entered.
@third - live intra-market arb, caught mid-trade
Joined June 2026 (brand new). 4,170 predictions. Current open positions:
Market: Bitcoin Up or Down - June 29, 2:45AM-2:50AM ET
Down 43.4¢ 625.1 shares → 309ドル.42 (+38ドル.11)
Up 59.1¢ 595.9 shares → 300ドル.95 (-51ドル.00)
Both sides. Same market. Entry was when Down + Up summed below 1ドル.00 - but prices moved after entry, so combined they're now 102.5¢. The bot is slightly underwater waiting for resolution. This is exactly how the strategy looks mid-execution.
What's interesting: it entered both sides within the same 5-minute window, holding until the market resolves, then collects 1ドル.00 per complete set regardless of which direction BTC moved. The directional outcome doesn't matter. Only the entry price matters.
What this tells you about the current state of Polymarket arb
The arb is real but the windows are narrow. third is already underwater on a position from the spread moving 2-3 cents between signal and fill. At the scale these bots operate, execution latency is the main variable.
Pure mechanical arb has a lifespan. second ran 16k trades then stopped. The edge gets competed away as more bots scan the same markets. Directional strategies (first, still active after 8 months) seem to have more staying power because they require actual market judgment.
The 5-minute BTC markets are the arb hotspot. Every bot I found is concentrated there. They resolve in 5 minutes, have consistent liquidity, and the binary structure makes merge/redeem clean. The tradeoff is that every other bot is also there.
The bot I built
Based on studying these strategies, I built a Polymarket bot in Python that implements 5 strategies:
Strategy 1 - Intra-market merge/redeem (what wolf9478 is doing)
Buy YES+NO when sum < 1ドル.00, merge, redeem. Risk-free on paper, execution-speed-dependent in practice.
Strategy 2 - Combinatorial arb
Markets with 3+ outcomes. All outcome prices should sum to 1ドル.00. When they don't, same approach.
Strategy 3 - Cross-platform arb
Compare Polymarket prediction prices to Binance spot. If BTC has a 70% chance of being above $X and the market is pricing it at 45%, buy the underpriced side and wait for convergence.
Strategy 4 - Endgame arb
Near-resolution markets where one outcome is at 93%+ probability. Buy it, wait a few hours, collect ~7% return on near-certain outcome.
Strategy 5 - Momentum/mean-reversion
What eknih appears to be doing. Track YES/NO prices as a time series, apply z-score and RSI, enter when overextended. Three timeframes: 5m scalp, 15m swing, 1h position.
The bot runs all strategies in parallel, then uses a composite scorer to rank signals:
score = (
profit_score * 0.30 +
confidence * 0.25 +
strategy_prio * 0.20 +
urgency * 0.15 +
risk_reward * 0.10
)
Top-ranked signal executes first. Risk manager prevents duplicate positions in the same market.
Lessons from watching these bots
If you're building something similar, here's what the data suggests:
Speed beats strategy for merge/redeem. The arb exists briefly. If your order takes 2 seconds to fill, someone with a 200ms latency already closed it.
Don't rely on a single strategy. second's pure arb approach ran out of steam. first's directional approach is still grinding months later.
The 5m BTC market is crowded. Profitable but competitive. If you're starting fresh, look at less-trafficked markets where the arb windows stay open longer.
Track your Sharpe, not just P&L. A high prediction count with modest P&L (like first) might be negative expected value after fees if the win rate isn't high enough.
Repo: github.com/casatrick/polymarket-arbitrage-bot
Not financial advice. Polymarket can be illiquid. You can lose money. These accounts might not even be bots - but 19,000 predictions in 8 months is hard to do by hand.