We apply a hybrid, data-driven and physics-based method to predict the most likely futures of gas production from the largest mudrock formation in North America, the Marcellus shale play. We first divide the 100,000+ square miles of the Marcellus formation into four regions with different reservoir qualities: The Northeast and Southwest cores, and the noncore and outer areas. Second, we define four temporal well cohorts per region, with the well completion dates that reflect modern completion methods. Third, for each cohort, we use the Generalized Extreme Value statistics to obtain historical well prototypes of average gas production. Fourth, cumulative production from each well prototype is matched with a physics-based scaling model, and extrapolated for two more decades. The resulting well prototypes are exceptionally robust. If we replace production rates from all wells in a given cohort with their corresponding well prototype, time-shift the prototype well according to the date of first production from each well, and sum the production up, this summation matches rather remarkably the historical gas field rate. The summation of production from the existing wells yields a base or do nothing forecast. Fifth, we schedule the likely future drilling programs to forecast infill scenarios. The Marcellus is predicted to produce 85 Tscf of gas from 12,406 existing wells. By drilling ∼3,700 and ∼7,800 new wells in the core and noncore areas, EUR is poised to increase to ∼180 Tscf. In contrast to EIA, we show that drilling in the Marcellus Outer area is uneconomic.