I found an error in my baseball model code. It's been there for three years.
For the betting nerds, I have returned a +27.5 Unit performance in the 20 days since.
On a financial basis, a $1,000 bankroll would have profited $1,199 in that timeframe.
This is my explanation of what exactly happened, since I think it's easier to explain with a little demonstration.
In addition, I've begun tracking MY performance on baseball wagers since June 30th (when my code fixes occurred)...been comparing it to other so-called experts. Expect a video series of me critiquing those other accounts. They aren't all frauds, but they aren't running a tight ship.
I didn't think Model 9 could be a thing, but here I am. No coworkers, so I have to speak this into some void. I'm not at the point where I think it flies, but it seems to work. The numbers are smooth.
- Broke the collection of outs into many categories. Before, it was just "out" or strikeout. Now there's groundouts, lineouts, flyouts, and popouts. Turns out they behave very differently under different settings.
- Temperature/Wind calcs are added. I had previously collected 10 years of stats categorized by temps and winds, but I never put them into the meat grinder to find the curve of performance. Temps adjust stats from 35-105 degrees, and winds are measured from one of 8 directional...directions.
- The only reason I felt encouraged to try this was due to figuring out a way to speed up my simulators. Simulating a game 5,000 times has been cut down by 70-80%. 90 seconds to 20-30 seconds changes the scope on many things.
- Pitcher stats. Behind the scenes, I spent 45-60 minutes updating these dudes on a WEEKLY basis. With the improved simulator, I can update pitcher stats daily in about 15 minutes, that includes weather stats.
- Idk, if nothing else it seems cool.
Range from lowest to highest value for:
No Temp Adj - 0.177 runs per game
W/ Temp Adj - 0.118
For the purposes of making starts comparable for a guy who starts in all hot weather vs. a guy in cold weather, I'd say this is a good launch point.
The image below are simulated 10-day average, league-wide ERAs for pitchers in 2025, broken out into "months" of that season.
The left column for each month is the raw data.
The right column has a Temperature Adjustment to scale results down to 72 degrees.
The image below are simulated 10-day average, league-wide ERAs for pitchers in 2025, broken out into "months" of that season.
The left column for each month is the raw data.
The right column has a Temperature Adjustment to scale results down to 72 degrees.
I've also ripped out a lot of how I calc'd pitcher data (it was stupid).
I have added Wind/Temperature factors to game data. So if a game is played in 95 degrees with the wind blowing to CF, simulation results would change.
This is an "update" on my baseball model work (Model 9?)
I found a way to cut the simulation time for a game down. A lot. 5x faster.
A small-scale simulator I use for pitchers previously took 18-20 seconds to simulate 5,000 games' worth of data. It's humming at 3.5 seconds.
If a righty batter faces winds blowing towards/away from leftfield, the results are clearly better for winds blowing out.
That said, excessive winds do not present a greater opportunity for scoring besides perhaps a fluctuation in sample sizes on the extreme ends.
Wind data is more segmented, but nothing stands out as significant.
Heavy winds blowing in/out does not create any smooth improvement in batter OPS.
Even HR% is scarcely affected.
Stiff crosswinds do not impact the rate of defensive errors either.
Wind data is more segmented, but nothing stands out as significant.
Heavy winds blowing in/out does not create any smooth improvement in batter OPS.
Even HR% is scarcely affected.
Stiff crosswinds do not impact the rate of defensive errors either.
Does weather have an impact on MLB scoring?
Using play-by-play data of 750,000+ PAs from 2022-2025, batter OPS rose with temps.
Wind? Not so much.
Wind shockingly doesn't move the needle as predicted. 20+ MPH winds to leftfield doesn't help righty batters in a clear way.
The most curious thing in this data from 2023-2025 is that the top base stealers aren't necessarily more successful at their attempts.
The difference is the top base stealers simply attempt it more often.
When did these SB attempts occur? The image shows all the "settings" of an MLB game.
71% of all SB attempts happened with a Runner on 1st, any outs.
87% of all SB attempts happened with either a Runner on 1st or Runners on 1st/3rd, any outs.
If we look at the most common base-stealing scenarios (Runner on 1st, and Runners on 1st/3rd), there is a notable jump in aggressiveness when the outs climb.
20.5% of attempts happen with no outs.
32.5% with one out.
47.0% with two outs.
When did these SB attempts occur? The image shows all the "settings" of an MLB game.
71% of all SB attempts happened with a Runner on 1st, any outs.
87% of all SB attempts happened with either a Runner on 1st or Runners on 1st/3rd, any outs.
In 2023, the MLB implemented larger bases, limited pickoff attempts, and the pitch clock. It led to the following:
SB attempts per year
2022: 3,297 (75.4% success rate)
---
2023: 4,369 (80.2%)
2024: 4,578 (79.0%)
2025: 4,429 (77.7%)
2026: On pace for 4,219 (75.2%)
In 2023, the MLB implemented larger bases, limited pickoff attempts, and the pitch clock. It led to the following:
SB attempts per year
2022: 3,297 (75.4% success rate)
---
2023: 4,369 (80.2%)
2024: 4,578 (79.0%)
2025: 4,429 (77.7%)
2026: On pace for 4,219 (75.2%)
Be at a packed college bar in downtown Providence, on a Saturday night, with no TV audio playing on the speakers, and spend the next 20 minutes as a borderline mob because everyone thought they screwed up the call at home plate.
The winner needs to reach hearts beyond the insular PC community.
I have spent many years as an Ohio man battling for PC hoops relevancy amongst the loud OSU football culture.
I follow almost no one. If you rise to the top in my feed, it means you are a true burner sicko.
I was a junior during Bryce Cotton's preternatural excellence. I once went to rec center and he was the only other hooper. I left. I couldn't disgrace him.
Weak region. None appeared on my feed. Great burners can capture the hearts of the 30+ Ohioan alumni.
This is where we test the limits of the PC burnerverse. Can you appeal to a Midwest resident, 2015 graduate? Always Sunny in Providence is the only one to trickle into my feed, and I frantically rely on this world for my Friar heartbeat.
Region of death. Letter swap names rule this field. Deno Kavis? Gicket Taines?
No one is funnier than Dark Pinnis. I don't know if it's actually funny. The name is so funny it blinds me.
I follow 16 accounts. PC Shepherd is the only one I followed because he frantically asked me if my 33 year old friend from my days at PC had a burner. Dude was frothing at the mouth. How do I vote.
In defense of the guy, he does know baseball.
During his run, he's gotten a question that was basically "who did Bob Feller play for in the 1940s" and he recently got one where Ty Cobb was the correct answer.
Big 12 Schools is one of the categories on Jeopardy tonight.
There's a guy who has won 29 straight Jeopardy games.
He was stumped when asked about a college in Waco, Texas