Pinned Tweet
Just finished a huge UPGRADE to my Polymarket Arbitrage Trading Bot
📈 $41,514 +EV - [ Right Now]
✳️ 1,032% Spread - [ Right Now]
#⃣ 3X More Arbitrage Opportunities
I had to break a lot of rules to get this to work,
If Polymarket finds out I might be in trouble...
But it was worth it!
I was able to bypass the restrictions that are holding back other Arbitrage Trading Bots
Here’s how it all works...
MARKET MATCHING
The bot is looking for arbitrages across 5 different Prediction Markets
To do that we are indexing millions of individual markets
To try to find the few thousand functionally identical pairs between platforms
That's billions of potential matches
To find the few thousand market pairs that are functionally equivalent between platforms we are using a four part matching system:
1. Keyword extraction, ranking & matching
2. Trigram, Jaccard & Vector hybrid matching algorithm of the market titles, close conditions, alternative titles & outcomes
3. LLM Prompt matching checking for functional equivalence of market rules
-> Incredibly inconsistent, hence the need for part 4, building a system like this at scale will open your eyes to the shortcomings of AI
4. Human verification
+ 99.8% Accuracy
+ 6,320 Markets Matched
Once we have our markets, we move onto..
ARBITRAGE DETECTION - [ UPGRADED ]
To detect if there is an arbitrage we need two things
Market Odds & Orderbooks
Market Odds: This will show us the ‘Spread’, if the sum of Market A YES & Market B NO is less than $1, or vice versa, we have a potential arbitrage
Orderbooks: This will show us the "EV", much we can arbitrage profit we can extract, according to the available liquidity and slippage of both orderbooks
This is where we had been severely limited in the past, due to inadequacies of the WebSocket feeds and API rate limits
Polymarket:
Orderbook initial dumps & entire orderbook price levels missing when connecting hundreds of markets to the WS feed, undisclosed multi-WS rate limits
Opinion:
API rate limits, WS delta updates missing, WS delta updates sent in wrong order, asks sent below best bid, airdrop farming bots posting and filling their own orders breaking WS feed.
Kalshi:
Rate limits and minor book inaccuracies at scale
PredictFun & Probable:
Surprisingly accurate as of current, monitoring how they handle increasing volumes
To get past these limits and scale the Arbitrage Finder we built some advanced new systems
1. Multiple Instances
Instead of scaling vertically we moved to scaling horizontally, a central controller handles the deployment & management of multiple proxied worker instances that each keep a local record of a subset of the market orderbooks and detect arbitrage opportunities as soon as dif updates are received
These worker instances feed the orderbook data back to the main controller which aggregates all information in one place and formats along them with relevant metadata to be fetched by our trading interfaces and applications
2. Handling “Junk Data”
One of the most challenging parts of scaling this application is dealing with the inaccuracies of the data provided by the APIs that we refer to as ‘Junk Data’
Some are easy to deal with:
- Book updates returned in the wrong order required an additional ‘lastTimestamp’ value at each book level which was referenced before any future updates are applied, if diff update timestamp was prior to lastTimestamp the dif update is ignored.
- Missing book dumps / levels reduced almost entirely by reducing the number of CLOB tokens per WS connection
- Dif ask/bid flips appearing at impossible levels are not applied
Some were a lot more challenging:
- Missing book updates were only detectable with revalidation & comparison, we don’t know what we don't know until we know we don't know it. More complex revalidation triggers and short recycling periods minimize the issue
With these updates we can scale the number of local orderbooks we are handling at one time:
Before: ~4,000 orderbooks
After: ~10,000 orderbooks
This, along with the improvements in orderbook accuracy, has increased arb density by 3X
Meaning we’re finding 3X the amount of opportunities as before
3. Rate limit bypass
To bypass the API limits that limit the quantity of markets we can subscribe to at once we had to ██████ ███ █ ██████ █████ █████████
TRADING SYSTEMS - [ NEW ]
The data is only as good as you can display it, ultimately the format in which the data is served will determine how efficiently it can be acted upon
We’ve created a system of interconnected tools that enable us to trade these opportunities, each with a different specific use case
1. AlertPilot Trading Terminal
A dashboard displaying all the hundreds of arbitrage opportunities the bot has found across 5 different prediction markets in real time
+ Arbitrage Calculator, showing you exactly how much to bid to take advantage of the arbitrage according to your bankroll, fees & slippage
+ Double Price Chart, which helps traders to estimate how long their arbitrage take to close
+ Strategy Guide, explaining how to execute arbitrage trades most effectively to maximize profits
+ Position Manager, connect your wallets to see your open arbitrage positions, EV, profits & exits
2. AlertPilot Telegram Bot
A system for getting alerts on all new arbitrage opportunities immediately, EV, Spreads & market links
+ Custom Settings, only see the arbitrages you’d want to take with user specific settings
+ Position Sell Alerts, connected to the AlertPilot Position Manager, get alerts to your phone when its time to sell your arbitrage positions
+ All 5 Markets, alerts on all 5 supported prediction markets: Polymarket, Kalshi, Opinion, Probable & PredictFun
3. AlertPilot Discord Bot
Private chat rooms and arbitrage alerts on the AlertPilot Discord Group
+ Custom Alerts, the best arbitrage alerts are sent to the discord channel
+ Support, traders answering your questions on Arbitrage Trading
4. Arbitrage Trading Terminal [ SOON ]
A trading terminal built specifically for Arbitrage
+ Atomic Execution, enter positions on two platforms at the same time
+ Visualize the Arbitrage, trade with both charts in one place, see the gap close as you take your positions
+ Manage positions across multiple prediction markets in one place
Always hiring backend developers with prediction market experience,
Work with a sharp team on a product thats found real PMF in the prediction space (rare)
ALWAYS hiring Talented Frontend / Backend developers
- Work on Prediction Market Adjacent products
- Solving user problems with inventive solutions
- Backgrounds in SaaS development
ALSO, anyone with a history of robbing sports books with +EV Strategies, we want to hire you
Hiring developers who want to work in the Prediction Markets sector
Requirements:
- Handling large datasets
- OR creating complex frontends
- Prev work with Poly or Kalshi API
- Thinks everyone else is mildly retarded at their job and that they could do it better in less time
No agencies
Eastern Europe / Central Asia preferred
Msg on TG with:
- One thing you've built that you're most proud of
- Tech stack
- Location
Someone is creating a sybil farm using my Polymarket referral code
Thanks for the $20, but i assume if this continues they just ban my referral account
Polymarket & Kalshi give this same event different odds 🤔
Prediction Markets are inefficient, these mispricing's appear hundreds of times per day
Some last for seconds, some hours, until an arbitrageur comes in and eats the spread
Pocketing the difference in pure profit
10% referral commission for 30 days only is considerably worse than the old dub referral system
But KOLs aren't allowed to be mad at this as they've now linked their Polymarket accounts to their X
Checkmate!
Referrals now pay just 10%, and stop paying after 30 days
And we never got the Dub referrals transferred to Polymarket referrals as promised
So from very high to very low
Hard to build anything around these systems when they are subject to constant change
LF Consistency
There’s always a catch.
🎰 How Polymarket Forces You to Be Wrong
Take a close look at the formula:
wV = Trade Size × (1 − Entry Price) × Category Weight × Bonuses
If you buy a safe, high-conviction outcome at 90¢ (90% probability), your multiplier is: $1 - 0.90 = 0.10$ (virtually nothing).
But if you buy some wild degen outcome at 5¢ (5% probability), your multiplier is: $1 - 0.05 = 0.95$ (9.5x higher!).
They’ve literally engineered a system where it’s mathematically most optimal to be wrong.
To speedrun your way to the Obsidian tier, you are forced to constantly bet on long shots with a 5-10% chance of winning.
Sure, your bets will go to zero 95% of the time, but you'll get a 50% rebate on your next trades!
Built a Polymarket chrome extension that shows me everyone's PNLs for easy LARP detecting
Also blocks all tweets from Kalshi affiliates
“Missteps”
Egregious fees
Inconsistent resolutions
Ghost fills
Removing active markets
Countless bugs
“Missteps”
Trading volume on Polymarket has declined for the first time in eight months, a setback that comes as its founder acknowledges missteps and its chief rival Kalshi continues to grow bloomberg.com/news/articles/…






















