The Model Went Undefeated Through Its First 14 MLB Picks. How Good Is That?
Thirteen wins. No losses. One push.
Through the first 14 MLB picks that settled Friday, BBMI was 13-0-1. The full night finished 14-1-1 across 16 picks: first-five moneyline went 3-0-1, full-game moneyline went 4-1, and the run line went 7-0. The probability calculation below focuses on that undefeated opening stretch.
The lone loss came on Seattle's full-game moneyline, with the Mariners falling 7-5 to the Angels. That result stays on the record alongside the 14 wins and the push.
That's a night worth enjoying. Baseball usually makes us work considerably harder for considerably less.
It's also the kind of night that deserves a little perspective. The model didn't suddenly become unbeatable. It had an excellent night in a long season of trying to find small, repeatable advantages.
Still, there's a natural question: how unusual was it?
Using each product's historical record entering Friday and assuming independent results, the combined estimate for 13 wins, no losses and exactly one push is roughly 1 in 1,300, or about 0.08%. Both figures are rounded.
That estimate combines the results across products and includes the push. It comes with a substantial caveat: several picks can depend on the same baseball game. We'll get to that, because it matters.
First, here's the complete September 25 record from the updated picks log:
F5 spreads contributed no picks to Friday's record. After Friday, first-five moneyline stands at 128-88-37 for the season, full-game moneyline at 115-68, and the run line at 249-136.
The math
For the probability calculation, we remove Friday's results. Using those wins to estimate the chance of those same wins would give the hot streak a vote in its own evaluation.
Then we multiply across products.
The push needs a little care. Entering Friday, first-five moneyline had 125 wins, 88 losses and 36 pushes across 249 settled picks. Its 58.7% win rate excludes pushes. To calculate three wins and one push in four picks, we instead use 125/249 for a win and 36/249 for a push, allowing the push to occur in any of the four spots.
For one possible composition of that 13-0-1 start, the calculation is three F5 moneyline wins and a push, four full-game moneyline wins, and six wins on the run line:
The log records dates and results, but not the order in which picks settled. Depending on which product supplied the remaining win after the 13-0-1 start, the combined estimate ranges from 0.07% to 0.08%, about 1 in 1,370 to 1 in 1,280. That's why we're calling it roughly 1 in 1,300 rather than pretending the log supports a more precise reconstruction.
A big caveat: these picks are not independent. A first-five moneyline and a full-game moneyline can cover the same baseball game. A strong start can help both cash, while a bad start can hurt both. Picks on the run line can overlap those games too. Multiplying the probabilities assumes each pick's outcome is independent of the others, which is not true of this slate. The 1-in-1,300 figure is an independence benchmark, not the actual probability of Friday's streak. Shared outcomes can make a sweep more likely than that calculation suggests; the size and direction of the effect depend on the sides and markets involved. We have not estimated that adjustment.
The historical rates are estimates too. They aren't precise probabilities for these particular matchups. The moneyline products remain labeled in calibration, and their tracked histories include changes in the rules that determine which picks fire.
We also noticed the streak after it happened. Across a long season, there are many opportunities for an eye-catching stretch. A small probability for one specified set of picks is not the probability of ever seeing a great night.
None of that takes away the wins. It keeps the explanation honest.
Winning isn't the same as beating the book
There is another distinction that matters even more over the long haul: winning a pick and beating the sportsbook economically are different things. Prices matter. Winning 60% of bets at -200 would still lose money over time. A winning record has to clear the cost of the odds attached to it.
That's why BBMI keeps the longer record in view. Forecast accuracy, the prices available, and results across hundreds of picks tell us more than one Friday night can.
Most of a baseball season is a slog. A late bullpen collapse turns a sensible forecast into a losing ticket. Sometimes several close games break our way at once.
The work is to keep testing whether the model's advantages hold up through all of it. One big night doesn't finish that work, and one bad night doesn't erase it. Neither is a reason to abandon discipline or suddenly change the stakes.
So yes, we'll enjoy 14-1-1. We'd be lying if we said otherwise.
Then we'll log it, keep the loss and the push beside the wins, and get back to work. The goal is a record that holds up over seasons. Friday was a very good day along the way.
Source: BBMI's complete MLB picks log, checked September 26, 2026. The opening 13-0-1 sequence is based on Andy's contemporaneous account; the completed 14-1-1 tally and product records are confirmed by the updated log. Calculations use unrounded historical frequencies and assume independence.
Disclosure: analytical and educational content about model performance. Not betting advice. Rankings, predictions, and probabilities are model outputs, not financial or betting advice, and past performance does not guarantee future results.
Related research
Schlittler Pulls Clear; Misiorowski and Yamamoto Lead the NL Race
September 16: Cam Schlittler leads the AL runs-saved model at 99.8%. Jacob Misiorowski leads the NL at 57.4%, with Yoshinobu Yamamoto at 37.2% and Chris Sale at 5.4%. Updated inputs, 500 simulated finishes.
Cam Schlittler 99.8% AL · Jacob Misiorowski 57.4% NL · 500 simulated seasonsSep 16 · 4 minAwards · MVPCrow-Armstrong Pulls Ahead in the September MVP Update
September 16: Pete Crow-Armstrong leads the NL value model at 97.0%, up from 66.2%. Yordan Alvarez again finishes first in all 500 AL simulations. Updated batting, pitching and Statcast fielding inputs; the same player-value framework.
Yordan Alvarez 100.0% AL · Pete Crow-Armstrong 97.0% NL · 500 simulated seasonsSep 16 · 4 minAwards · Cy YoungSchlittler, Misiorowski Hold Cy Young Leads as Sale Surges
Cam Schlittler and Jacob Misiorowski answered late-August pressure with wins, and the model still has both atop their leagues at 67.0% and 58.8%. The movement is behind them: Parker Messick has climbed to 20.2%, Dylan Cease to 11.8%, and Chris Sale to 11.8% on the back of a nine-inning shutout of the Dodgers. Plus the July and August calls, graded.
Schlittler 67.0% AL · Misiorowski 58.8% NL · 500 simulated seasonsSep 2 · 6 minSee it live
This is the “we show our work” side. The Prediction Record is the other half — every call locked before the game and graded in public.