Marcus Vance

Ex-prop desk, Chicago. I pipe orderflow, news and onchain data into LLMs. Polymarket · Bybit · Python. Building my edge in public.

[Marcus Vance @0xvancee]

@polytutor already walking the ladder, not the top. SOL was 0.190 at the top, 0.223 avg on $200, so a third of the edge gone. wasn't logging the mid though. it is now, on every pass. DMing you

[Marcus Vance @0xvancee]

I used Claude Opus 5.5 to build an agent that hunts mispriced strikes on Polymarket

Paper result: +$128.90 on day 1

44 markets watched. 7,919 quotes priced. 2 trades passed every check. 1 bug caught on the first run.

I called it Fairline. Here's how it works:

FEED pulls Binance spot every 20 seconds plus two volatility reads: 4 hours of 1m candles and 7 days of 1h candles. It keeps the bigger one.

PRICER turns spot, strike and minutes left into a fair probability for every "BTC / ETH / SOL / XRP above ___" strike.

BOOK walks the real Polymarket order book for a $200 ticket. The fill is whatever the ladder gives, not the top ask.

GATE trades only when fair value beats the all-in price by 4+ points after fees. Nothing in the last 10 minutes. Nothing more than 8 hours out.

SETTLE grades every trade off the source named in the rules: the Binance 1m candle that closes at 12:00 ET.

Day 1 tape:

Nov 8, 15:49 UTC. BTC at $76,060. Market had "above $76K" at 63.7%. Fairline said 56.1%. Paper-bought NO at 0.365.
Close $75,780.22 → +$339.77

Nov 8, 15:49 UTC. SOL at $89.69. Market had "above $90" at 19%. Fairline said 31.6%. Top ask 0.19, but the book was thin and the $200 fill averaged 0.223. Slippage ate a third of the edge.
Close $89.06 → −$210.87

The bug:

v1 only looked at 4 hours of 1m volatility. A quiet afternoon made a 25-hour move look impossible, so it "found" 9–15 point edges on the next day's BTC and ETH strikes. The model was wrong there, the market was right. All 4 trades voided.

The math:

P(above K) = Φ( ln(S/K) / σ√t − σ√t / 2 )
σ = max(σ_4h, σ_7d) × 1.15
edge = fair − avg fill − fee per share

No $68 → $750K. Two fills, every number logged.

[Marcus Vance @0xvancee]

I used Claude Opus 5.5 to build an agent that hunts mispriced strikes on Polymarket

Paper result: +$128.90 on day 1

44 markets watched. 7,919 quotes priced. 2 trades passed every check. 1 bug caught on the first run.

I called it Fairline. Here's how it works:

FEED pulls Binance spot every 20 seconds plus two volatility reads: 4 hours of 1m candles and 7 days of 1h candles. It keeps the bigger one.

PRICER turns spot, strike and minutes left into a fair probability for every "BTC / ETH / SOL / XRP above ___" strike.

BOOK walks the real Polymarket order book for a $200 ticket. The fill is whatever the ladder gives, not the top ask.

GATE trades only when fair value beats the all-in price by 4+ points after fees. Nothing in the last 10 minutes. Nothing more than 8 hours out.

SETTLE grades every trade off the source named in the rules: the Binance 1m candle that closes at 12:00 ET.

Day 1 tape:

Nov 8, 15:49 UTC. BTC at $76,060. Market had "above $76K" at 63.7%. Fairline said 56.1%. Paper-bought NO at 0.365.
Close $75,780.22 → +$339.77

Nov 8, 15:49 UTC. SOL at $89.69. Market had "above $90" at 19%. Fairline said 31.6%. Top ask 0.19, but the book was thin and the $200 fill averaged 0.223. Slippage ate a third of the edge.
Close $89.06 → −$210.87

The bug:

v1 only looked at 4 hours of 1m volatility. A quiet afternoon made a 25-hour move look impossible, so it "found" 9–15 point edges on the next day's BTC and ETH strikes. The model was wrong there, the market was right. All 4 trades voided.

The math:

P(above K) = Φ( ln(S/K) / σ√t − σ√t / 2 )
σ = max(σ_4h, σ_7d) × 1.15
edge = fair − avg fill − fee per share

No $68 → $750K. Two fills, every number logged.

[Marcus Vance @0xvancee]

@0xVishnya @Plasma plasma the best card

[Marcus Vance @0xvancee]

@Katanaxhv what is this?

[Marcus Vance @0xvancee]

@shredderAI cool video