Patrick

@Whizard 100%. AI has sped up a lot of my processes and testing and coding, but I think a lot of people entirely offload all their thinking to it and then we all end up with the proliferation of a bunch of sloppy models. Unfortunate

@manulbets2 Jokes on you, my model runs 1,000,000 simulations! All jokes aside, I can garuntee you it’s worse than you can imagine, 16-17 year old kids are sitting in front of their computer with no statistical background or simple stats knowledge and trying to do this.

Another funny thing thats emerging is the “I ran 100,000 simulations of my model” Of course I love people using distributions instead of just point estimates, but what you sample from matters a lot & I bet people are not making good distribution choices with AI models. Be careful

Poisson requires 1 independent event at a time. Don’t use it for baseball scores Negative binomial requires overdispersion (variance > mean). Generally don’t use it for spreads/MLs Normal distribution generally assumes no skewness/kurtosis. Don’t use it for player props with long tails Monte Carlo sampling requires calibrated inputs

The past year has been the most productive year of my life for my real job, but the most frustrating and humbling year for gambling stuff. Sports betting is hard. At least I’ve learned a lot, I am optimistic to start consistently sharing things I’ve been working on

Lol there is no relationship in the plot at all

@PortmanTracker2 @BDunk975 @BrookieJ07 BrookeJ? More like BrokieJ am I right haha

Official @BrookieJ07 1 month recap Net results -142.39u ROI -43.24% Win rate 2.23% No edge. No accountability. Just losses.

@CallingOurShot “Never again” until day 1 again tomorrow

❌ Day 5 Banchero ends with 16. Bane ends with 12. Both shot 4/14. Never again. Had a chance for OT but the Magic left open the best 3 Point shooter in the league. Dead inside. We climb again soon. 🤦‍♂️

@jakemalasek I met Peyton at coupes one time, his wife (Marshall’s mom) went to UVA if I remember right

Peyton Manning’s son Marshall working out in a Virginia Football shirt.

@WalrusQaunt That totally makes sense, thanks so much!

@manulbets2 I was just looking at positive correlated situations. I will put the code up on GitHub. it dos all correlation testing for all players . And you can see where there is negative correlations .

most exploitable SGP pairs cluster around 8-15% lift over baseline. median is 13%. that means when a player goes over on one prop, the linked prop hits 13 percentage points more often than it would randomly. across 275+ statistically significant pairs. the right tail is what matters. a handful of player-prop combos show 20-28% lift. those aren't noise. those are structural patterns in how certain guys play. when they're active on the boards, their scoring follows. when they're distributing, the threes come. the left side of the distribution is still exploitable. 8-10% lift on a prop that already hits 50% of the time pushes you to 58-60%. books don't adjust SGP correlation pricing that precisely. ran this on every NBA rotation player, 2023-2026, 15-game rolling medians, lagged so there's no lookahead bias. filtered to 25+ MPG and 50+ games. prop test on every pair, p < 0.05. the edge isn't picking winners. it's knowing which props move together and which ones the book priced independently.

@DataBasedBets I tweeted it mostly as a dumb joke tweet but now thinking about it more it probably wouldn’t be that tough for specific cases (the Uber vs Lyft in the quoted tweet, maybe like food delivery or flights)

@manulbets2 Would personally invest in such a product

Odds screens for every industry

Been a minute since I've used the "public model consensus" approach on college basketball, but figured I'd run it and share for these early college basketball games. Gonna play: Ohio State -1.5 (-118, DK) Nebraska -12.5 (-117, Novig) Louisville -4 (-105, ProphetX) This just aligns public models to find agreement

@JayCuda I got it, but some of these were quite surprising to me. Wouldn’t have expected Tennessee > Virginia Furman > UCLA by 100 years Queens > Purdue Idaho > Houston by 50 years

i used 1 simple rule for picking the winners in this bracket....thought it would be easy for u to find the pattern but maybe not

@DailyMantle3 Zelenogorsk ✅

Manul World Cup 2026 (Final) - Zelenogorsk vs Tashi Who will win the Manul World Cup 2026?

Twitter expected a journalist from the Atlantic to write a positive or unbiased portrayal of gambling?

Great article! Thanks for writing it (This isn’t a fully thought out comment, so I may be completely off base here) If I remember right, KL divergence isn’t symmetrical- PvsQ isn’t the same as QvsP. In your case I think that is the difference between “this changed excitement given our win prob model” and “how much does our win model reflect the change in excitement”; both are interesting questions. I am interested in the latter but am not sure how to implement it (what would P be?). But I’ll give it more thought. Reminds me of this- where we know win probabilities (your “true” Q dist) are in themselves sensitive to specific time points x.com/manulbets2/sta…

New (and 1st) post up on Substack. Creating a better way of ranking games with the help of Claude.......Shannon. Borrowing a concept from information theory, KL-divergence, I create a new version of the Excitement Index that better aligns with how people evaluate the games.

@IainMacBets The followers and commenters are also bots, it’s just a new iteration of the dead internet theory. Don’t let it get to you Iain! You are not cooked!

AI girls are taking over the betting space. I’m cooked. Back to the tire factory for me.

@denitsa_tsekova @astraffon @justinaknope @rachaeldottle It’s run that you made the plot look like a bomb

We looked at the average bettor in Iran-related prediction markets. In one of the most popular markets, ~90% of wallets bet $1,000 or less. The 10 accounts that traded $1M+? They only started betting after the initial news broke. W/@justinaknope @rachaeldottle

@JFurKSL The court can’t get wet? That sounds really dumb

One major design flaw (so far) is there’s no halfcourt line during the game except for in the logo, which is tough to see. Also, courtside seats now can’t serve drinks because it’s a digital court.