Bayesian playoff sims

A basketball forecast starts with one question — who wins one game — and this playoff model turned that into a full bracket on social media with published assumptions and simulation counts. (nba.com) The posts described a Bayesian logistic regression, a version of a win-probability model that estimates the chances of a yes-or-no outcome and updates those estimates with prior beliefs and observed data. Logistic regression is widely used for binary predictions, and Bayesian versions are built to express uncertainty instead of a single fixed answer. (turinglang.org) For the playoffs, the model used player impact inputs, adjusted for home and road performance, converted each matchup into game-level win probabilities, and then ran 100,000 simulated brackets to estimate each series result. Monte Carlo simulation — repeated random trials using the same probabilities — is a standard way to turn one-game odds into series odds. (sports.sites.yale.edu) (bayesrulesbook.com) The timing matters because the 2026 SoFi NBA Play-In Tournament ran from April 14 to April 17, and the first round of the 2026 NBA Playoffs began on April 18. By April 20, the bracket was live and official series matchups were set across both conferences. (nba.com 1) (nba.com 2) The official bracket shows eight first-round series, including Detroit-Orlando, Boston-Philadelphia, New York-Atlanta and Cleveland-Toronto in the East, plus Oklahoma City-Phoenix, San Antonio-Portland, Denver-Minnesota and the Lakers-Houston in the West. The higher seed has home-court advantage in the NBA’s 2-2-1-1-1 format, which makes home-road adjustments a concrete input rather than decoration. (nba.com) (espn.com) That setup differs from a simple power ranking or a single “pick.” A simulation model can show, for example, that one team is favored in Game 1, less favored on the road in Game 3, and still only modestly ahead over a seven-game series after all those paths are counted. (sports.sites.yale.edu) (sciencedirect.com) Sports analysts have used logistic regression on basketball games for years because the method is interpretable: each input shifts the odds up or down rather than disappearing inside a black box. Recent academic work has also used Bayesian logistic modeling in basketball settings to estimate win probabilities while keeping uncertainty visible. (github.com) (frontiersin.org) The social-media thread drew attention partly because it showed the machinery — model type, matchup adjustments and simulation volume — instead of posting unexplained percentages. In a playoff week filled with picks from television panels, sportsbooks and team sites, that made the forecast readable to fans who wanted to know how the numbers were produced. (x.com 1) (x.com 2) The next test is simple and public: every game starting April 18 will either support the model’s probabilities or expose where its assumptions missed. That is the appeal of a transparent playoff sim — the math is posted before the ball goes up. (nba.com)