Built by an actuary.
Judged by the record.
BBMI is an independent sports-modeling project built around a simple question: how predictable are sports, really?
Sports generate enormous amounts of data. They also contain randomness, changing conditions, incomplete information and human behavior. That combination is exactly the kind of problem actuarial modeling is built to study.
Publish before the result
A forecast only counts if it existed before the game was played. Predictions publish with a timestamp and the line at the time.
Keep the misses
Bad predictions stay in the record. Results are posted with win rate, sample size and a confidence interval, so a hot month is not mistaken for skill.
Label the test
Historical backtesting and live performance answer different questions, and each result says which one it is. If a method changes, the old results are labeled, not deleted.
How the models work
Opponent-adjusted strength from the level the sport gives: plate appearances in baseball, efficiency and pace in basketball and football.
Thousands of runs per matchup produce a full distribution of outcomes, not a single number.
A sportsbook consensus is a strong external benchmark. The projection is made first, then measured against the available market number.
Results post to the log with confidence intervals. Every season is an ongoing test on games that were not used to fit or tune the models.
The methods differ by sport. The standard for evaluating them does not.
Questions, corrections, arguments
Common questions
Why should I trust the record?
Because it is not curated. Predictions are published before games are played and the results are recorded afterward, including the ones that went badly. Confidence intervals and sample sizes are shown alongside every win rate, so a good stretch is not mistaken for a settled result.
What does walk-forward mean?
Every season is treated as an ongoing test on games that were not used to fit or tune the models. It is easy to build a model that explains the past; predicting the future is harder, and only the second one counts here.
Is the goal to beat the betting market?
The market is the benchmark, not the goal. A sportsbook consensus represents the combined information of a highly competitive market, so beating a weak baseline is interesting and producing useful information beyond a strong market benchmark is much harder. A large disagreement is a hypothesis, not evidence that BBMI is right.
What does membership add?
Most of BBMI is public: the research, the rankings and the performance record. Premium membership adds access to premium picks and alerts when new picks publish. The historical record stays public, including the misses. Membership gives you access to the picks; it does not buy a cleaner version of the record.
Which sports are covered?
Seven are modeled and six carry a published record. Coverage and the graded count for each are in the panel on this page, and every number there links back to the same model-performance pages it was computed from.
The longer version
Prediction first. Result second.
It is easy to build a model that explains the past. Predicting the future is harder. So BBMI treats every season as an ongoing test on games that were not used to fit or tune the models. Predictions are published before games are played, and results are recorded afterward. Some models will hold up. Some promising ideas will fail. When they do, that belongs in the record too.
What BBMI models
Different sports require different approaches. MLB and college baseball can be modeled from the interaction of pitchers, hitters, parks and game situations. Football and basketball depend more on team efficiency, opponent strength, pace and home advantage.
Those models feed four kinds of output. Rankings measure how strong teams appear to be. Projections estimate future game and season outcomes. Predictions turn projections into a specific forecast before an event occurs. Research uses the same framework to investigate questions that standings and box scores cannot answer.
The market is a benchmark
Sports offer an unusually difficult external benchmark: the betting market. A sportsbook consensus represents the combined information and expectations of a highly competitive market. Beating a weak baseline is interesting. Producing useful information beyond a strong market benchmark is much harder.
For products designed independently of market prices, BBMI generates its projection first and then measures the difference against the available market number. A large disagreement is not evidence that BBMI is right. It is a hypothesis. The result tells us more.
Research is part of the product
Not every interesting question ends with a pick. How much better is a team than its record suggests? What happens to a playoff race when one player is removed? How much of a hot streak is reproducible, and how much is variance? The Research Desk is where those questions get tested. Sometimes the answer confirms the hypothesis. Sometimes it does not. Both are useful.
The record is the point
I do not know in advance which BBMI models will prove valuable. That is part of why the site exists. The experiment gets more interesting as the sample grows: more seasons, more games, more predictions, and more chances to distinguish persistent signal from a good-looking streak. So the forecasts, the results, the methodology and the changes all stay visible.
Don’t trust the claim. Check the record.