This project challenges the limitations of traditional point-estimate predictions by modeling NBA team scoring as a probabilistic distribution rather than a single value. Using historical game logs and opponent defensive metrics, we engineer features to fit Negative Binomial distributions for entire teams, accounting for both their average offensive output and inherent variance (consistency). By calculating the Cumulative Distribution Function (CDF) for specific betting lines (e.g., Team Total Over/Under 112.5 points), the model generates a "true" win probability for every game. This allows us to systematically identify and exploit market inefficiencies where the sportsbook’s implied probability diverges significantly from the team's actual statistical profile.