How to scalp prediction markets (without losing to the spread)

Binary Up-or-Down windows on Polymarket look tradable — but the spread eats you alive unless you approach them right. Here is what actually works.

Every scalper who tries Polymarket for the first time thinks the same thing: "Those Up-or-Down markets for BTC / ETH / SOL look tradable — I'll just scalp them." Then they discover the spread and give up in a week. Here's the version that survives contact with reality.

What you're actually trading

Every 5 minutes and every 15 minutes, Polymarket lists a market for each of ~6 crypto tickers: "Will BTC close above open at 12:00 UTC?" YES + NO shares. Both settle to 0¢ or 100¢. Spreads at market open are often 2-5¢ wide — half your edge is gone before you type in the size.

Where the edge is

Post-window opening imbalance. First 30 seconds of a new window, the book is usually thin and mispriced against fair (rough spot momentum + funding-rate-like drift). Small edge, executable if you're fast.
Late-window overreaction. Last 60 seconds, spreads WIDEN as market-makers pull. Big overreactions happen on news. If you're disciplined about only fading extreme moves (10c+ from midpoint), there's edge here.
Volatility contraction (the "squeeze"). When realized vol drops below 0.15% (std-dev of last 12 closes / spot), the market almost always breaks the range afterward. Trade the break, not the squeeze itself.

What the terminal gives you

The scalp pit at app.polyterminus.com/app/admin/scalp (currently founder-only, wider access soon) is built exactly for these three edges:

Live WebSocket top-of-book for all 6 tickers × Up/Down = 12 tokens. No polling, no stale prices at window open.
One-tap $1/$5/$10 FOK so you're not fighting UI latency when the spread compresses.
Technical gauges: RSI-14 (overbought / oversold), momentum score (-3..+3), realized vol %, EMA cross, volume spike, support/resistance from last 12 candles. All server-side computed from Binance klines, cached 15s.
SQUEEZE badge when σ < 0.15% — literally the "wait for the breakout" signal.
Trade flow bar: UP vs DOWN new-buy pressure this window, so you can see crowd bias.
Daily risk gate so a tilt-driven revenge-trade session can't drain the account.

The rules that keep you alive

Never chase. If the spread is wider than your edge estimate, skip the window.
Size for 40 windows a day. $500/day risk cap ÷ 40 windows = $12.50 per trade. Anything bigger is discretionary, not scalping.
Stop after two losers. The gauges are input, not verdict — if the signal is wrong twice in a row, the regime probably changed.
Track P&L honestly. The terminal ledger separates realized/unrealized for a reason. Don't just look at portfolio value.

The uncomfortable truth

Scalping any binary market well is hard. Polymarket has less liquidity than a real derivatives exchange, which means slippage matters more, and the crowd is smaller which means your alpha decays fast when other people notice the same signal. Treat it as a skill discipline, not a get-rich path. If you want to try — the tooling is here.