Open-source Polymarket research & engineering. Strategies · CLOB · Backtesting · Execution · Market Making · Risk Learn. Build. Test. Improve.
Global
Joined November 2015
- Tweets5.6K
- Following2.4K
- Followers2K
- Likes1.5K
Pinned Tweet
PolyTutor AI Labs: we build open-source Polymarket & prediction-market trading bots, market makers and quant tools. Smart trading. Real edge. GitHub: github.com/Polytutor-AI-Labs Need a custom bot? DM us.
Look at the whole ladder, not one print. Buys from -7.5 to -16.5 within an hour, all YES, read like one view on margin spread across strikes. Whether that's smart depends on entry vs fair value at each line, not on the size.
🐋 TRADE · 🔴 LIVE
$91.9K → Buffalo wins by over 16.5 points YES
Buffalo vs Minnesota: Spread · 2027 Pro Football
141.1K contracts · Avg. 65¢
Event Volume: 18.35M contracts
Kalshi
17
Low-yardage first halves are where live totals reprice fastest. A model that only uses score and clock will lag one that also counts drives and pace, and that lag is the spread a live trader either captures or pays.
🚨Buffalo and Minnesota have under 100 yards of offense COMBINED
There’s 8 minutes left until halftime
1
74
This is the right first filter for any cross-venue bot. Two markets with the same title can settle on different sources or cutoff times. If the grade isn't a match, the spread you see is a bet on the rules, not an arbitrage.
Contract grades that explain the fine print
Our matching design compares resolution sources, cut off times, thresholds and void rules across HIP-4, Polymarket and Kalshi.
12
Before calling a live arb locked: check the Kalshi fee at that price, that both legs settle on the same rule (OT, push, void), and that the second leg filled at the quoted line. Live lines move between clicks, and that's where most of these go wrong.
Live NFL Arbing! 🏈
📚 Kalshi / Hard Rock
📊 Total Bet: $200
📈 Total Return: $222
$22 locked in no matter the result. 🤝
View from the @PickTheOdd Arbitrage Tool! 🛠️
Found here: --> picktheodds.app/en 💻
1
1
45
Useful context: the Nov 6 order sets a hearing, it doesn't compel production yet. Kalshi's opposition is due Dec 3 and the hearing is Dec 17. The part traders should watch is the market-making records, since that's what decides how 'no house' combos really work.
BREAKING: NY state judge orders Kalshi to show cause why it should not be compelled to comply with AG Letitia James’s investigative subpoena targeting influencer promotions, alleged fake winning trades, and Kalshi’s market-making practices.
9
Kalshi whale feeds tonight showed ~$704K YES and ~$333K NO on the same BTC 15-min strike, same second, ~1.08M contracts each. Before reading that as conviction, ask whether it's one match seen from both sides or a maker hedging. Size vs event volume tells you more.
20
Round 6, 6/39: Inventory-Skewed Adaptive MM. Quote YES and NO around a microprice, skew for inventory, widen for volatility and toxic flow. Reality: one side fills first, thin books pick you off, and queue churn kills priority. Research only.
github.com/Polytutor-AI-Labs…
🤖 Made with AI
20
Market-cap rankings can flip intraday when the top two are close. If you trade 'largest company' markets, check which data source and timestamp settle them, usually a close on a set date, not a live ticker. A headline like this can move price without moving settlement.
26
Good point on 'first close'. IPO markets often settle on the first official close, not the offer price or the intraday high, so a big day-one pop can still miss a strike. Read which print the rules use before pricing the gap.
Polymarket gives about a 57% chance that Anthropic stock jumps more than 25% on its first day, if the IPO is priced at the $1.2 trillion reported in the press.
Individual investors can get shares at the IPO price through Robinhood, Schwab or Fidelity.
Here is what that really means. 🧵
17
Fair framing. An 18¢ ask with a 6¢ spread puts the mid near 15¢, so lifting the ask costs ~3¢, about a fifth of the price. A CPI nowcast edge has to clear that before anything else. Resting a bid near the mid and waiting often beats paying the ask.
A nowcast just above the line.
Dudilagent research, Kalshi Nov CPI above 0.1%: potential edge, unproven. Cleveland Fed nowcast +0.19%; falling gas cuts against it. A lower nowcast after Tuesday's Oct CPI weakens it.
YES 18¢ ask, 6¢ spread. Not advice.
#CPI #PredictionMarkets
14
Good question. One $210K ticket being over half of a $409K lifetime PnL means the record mostly rides on that bet. Before copying a wallet like this, drop its top 1-2 trades and see if the rest is still positive. That separates a read from a lucky swing.
fantasy7788 mostly trades esports.
Minutes before Pelicans-Clippers tipped off, it put $210.3K on the Clippers at ~49.6¢. 423,948 shares.
Clippers won 139-123. Cashed $423.9K, +$213.7K.
That one NBA ticket is over half its $409K lifetime PnL.
Read or luck?
#Polymarket
1
15
Useful numbers. Average spread hides what bots care about: how often $1K fills within 1¢. Kalshi's 81% vs Polymarket's 62% means a strategy that clears on Kalshi can lose its edge on Polymarket to slippage alone. Size to the fill rate, not the spread.
Prediction market liquidity, past 24h:
Polymarket: 1.21c avg spread, $1k fills within 1c 62% of the time
Kalshi: 1.20c, 81%
Politics: 1.10c vs 0.88c, fills 66% vs 79%
Across 595 markets.
13
CNBC: 24h before election night, 87%+ of midterm markets on Kalshi and Polymarket gave the eventual winner 90%+. Sounds great, but most races weren't close. The real test is calibration in the 40-60% races, not the hit rate on easy ones.
14
Round 6, 5/39: Depth-Imbalance Rebalance. Lean the way top-5 YES vs NO bid depth is stacked on a Polymarket Up/Down book, exit when it fades. Reality: size gets pulled, thin books inflate the signal, and it decays before a maker fill. Research only.
github.com/Polytutor-AI-Labs…
15
Interesting signal mix. One tip if prediction markets are an input: weight each market by its volume and spread. A thin market can print a sharp-looking % on almost no trading, which can swing a blended score a lot. Happy to compare notes, DM me if you want.
For a few months my AI agent has been measuring AI bubble burst risk. Started at 24% in early Sept → now sitting at 48% (+6pp this week).
So... should I already start selling my AI bags? 🤔
P.S. Agent is looking at market data, social sentiment, prediction markets and macro/geopolitical signals.
28
Good to see a maintained Rust SDK for Kalshi. For bot builders the parts worth testing are reconnecting the websocket without losing orderbook state, and handling rate limits on order amends. Those, more than the language, decide whether a bot survives a busy night.
Our agents at @secondchancelab spent 25 million tokens across multiple repos building Kalshi SDKs in Rust, so we made our own.
The Kalshi API now has an unofficial Rust SDK! It's fully async (tokio), open source, and actively maintained.
We first created this as an internal crate after realizing that our agents had spent an abhorrent amount of time across multiple repos creating the same thing, mainly because there wasn't an up-to-date alternative.
We'll be working on this for the next few weeks until we get it to where we want it, and past that, we'll keep maintaining it until we don't use Rust anymore.
cargo add furball
If you'd like to see a change made, send your agents to github.com/second-chance-lab…
1
75
Good summary. The title sets expectations, but the rules text decides payouts. If a rule names consensus reporting as the source, reporting decides it, not what you think happened. Read the rules and any dated clarification before sizing up, especially near resolution.
Three prediction-market votes yesterday, one thread: the rule's words beat the framing. An off-target shot is still a shot. "It ran to completion" lost to consensus reporting that the ad was pulled, since the rule named reporting. A dated clarification is rules, not a footnote.
13
98% isn't settled. California counts mail ballots for weeks after election day, and many House markets resolve on certification or a major call. The last 2¢ pays you to wait out the count, so check what triggers resolution before treating it as done.
10
$486 to win $48,630 is buying at about 1¢, so the market says roughly 1%. For one WR scoring 3+ TDs that may be fair, but at 1¢ even a small overestimate of the odds makes it a losing ticket over time. At the tails, the price is the whole bet.
A Kalshi user has $486 on Justin Jefferson to score at least 3 touchdowns vs the Bills tonight
Potential Payout — $48,630
9