icey

[icey @sosicey]

Hot take: Polymarket misprices crypto markets because the crowd doesn't read options.

I'm testing it with a bot. 7 days, $10k paper, every number public — even if I lose.

If the model wins, 10 people get a month of access 🧵

[icey @sosicey]

@grok @grok

[icey @sosicey]

Code is public. No wallet, no keys.

https://t.co/1PHekHiaSU

If you see how to adapt this so it actually makes money, not one lucky day, I want to read it.

[icey @sosicey]

Open-sourcing Polybot.

One week, paper only. The test was: does a Deribit option probability beat the Polymarket price?

It doesn't.

[icey @sosicey]

$10,000 paper account. Right now it's −$95.

Oct 15 paid. The days after gave it back. The "buy every outcome under $1" sets made $0.39.

No real orders. That part stays off.

[icey @sosicey]

The score I said I'd trust. Brier, lower is better. Through Oct 19.

All resolved markets: model 0.0446, Polymarket 0.0428.

Markets the bot actually traded: 0.218 vs 0.206.

The market won both.

[icey @sosicey]

@weedincgame @solana

[icey @sosicey]

(158) Day 2 of the week. Same log on Oct 22, win or lose.

If the model is still behind on Brier, no real money.

Not advice. A paper gain is not a track record.

[icey @sosicey]

Hot take: Polymarket misprices crypto markets because the crowd doesn't read options.

I'm testing it with a bot. 7 days, $10k paper, every number public — even if I lose.

If the model wins, 10 people get a month of access 🧵

[icey @sosicey]

(208) Risk did not do what I wrote down.

A restart wiped the limits. They live in memory. First cluster was about 6R per asset. The cap is 3R. 1R = $50.

BTC and ETH were the same bet. One move made the result.

[icey @sosicey]

(191) Same test, only markets I traded.

12 resolved markets. 4,782 snapshots.

Model 0.197 Polymarket 0.187

Still behind. The profit was one squeeze through the strikes, not a better forecast.

[icey @sosicey]

(228) The test I committed to. Brier, lower is better.

Only moments with a spread of 5¢ or less. An empty book cannot count.

33,301 snapshots. 1,366 resolved markets.

Model 0.0432 Polymarket 0.0421

The market is ahead. No edge.

[icey @sosicey]

(254) Five Oct 15 tickets were marked about −$330. They settled +$323.

BTC closed $82–84k. ETH closed $2,400–$2,500. That afternoon both were under the band.

Oct 16 directional: −$12. Full sets: +$0.39.

10 of 12 settled bets won. A streak is not an edge.

[icey @sosicey]

(148) How the +$255 breaks down.

Settled: +$311 Still open, marked at the mid: −$56 Equity: $10,255

Yesterday at 14:26 MSK the book was marked −$349.

[icey @sosicey]

1/ (173) 2 days in. Paper account: +$255 on $10,000.

The score I said I'd trust, Brier, says the model is still worse than Polymarket.

Numbers below. Paper only. No real orders.

[icey @sosicey]

🎁 10 people get a month of access to the bot on real money — only if the week is a win.

To enter:

Follow @sosicey
Rt
Leave like
Reply to the 1st tweet: which Polymarket market would you give the bot?
Rys Oct 21. NFA. No guaranteed results

[icey @sosicey]

✅ Win: model consistently beats the market on Brier, PnL > 0, drawdown within limits → small real money.

❌ Fail: model no better than the market → no real money, and I post a full breakdown of why.

Started today. Results Oct 21, right here. Follow to see it.

[icey @sosicey]

Only moments with spread ≤0.05 count — so empty books can't hand the model a fake win.

Also tracking: • calibration: "70%" should hit ~70% of the time • PnL after fees + slippage • drawdown in R • full-set profit

[icey @sosicey]

How I'll know if it works.

Main metric: Brier score — how close a probability forecast is to what actually happened. Lower = better.

Model vs Polymarket price, at the exact same moments.

If the model isn't more accurate than the market, there's no edge. I'll say so.

[icey @sosicey]

Because in the final hour you're not a trader, you're exit liquidity.

That's where bots watching Binance in real time live. Speed isn't my edge.

My bet: prediction markets are inefficient wherever the crowd doesn't understand options.

Agree? Or is Polymarket efficient?

[icey @sosicey]

Risk management > the model:

• 1 trade = 0.5% of bankroll (1R) • −7R in a week → size cut in half • max 3R per asset • max 1% of market volume • no directional trades in the last hour before expiry

Why the last one? 👇

[icey @sosicey]

Strategy #3: full sets. If all outcomes sum to < $1, buy every one — one of them pays $1, guaranteed.

Under the hood: • tick every 30s • snapshot of every order book every 5 min (~0.7 GB/day) for backtests • wallet scanner for consistently profitable crypto traders