We Simulated Every 2016 Warriors Game — Here's What the Data Says
The 2015-16 Golden State Warriors won 73 regular season games. That number has become shorthand for the ceiling of what an NBA team can accomplish over a full season. We wanted to know how accurately the OBM simulator reproduces that dominance — so we ran all 82 games through the engine and compared the results.
The setup was straightforward. We loaded the Warriors' 2016 classic roster into the simulator alongside each opponent's roster from that season. One game per matchup, using the lineup order and roles that reflect how Steve Kerr actually deployed his rotation. Then we let it run.
What the numbers looked like
The simulation gave Golden State 68 wins. That is five short of the real record, which sounds like a miss until you think about what the engine is actually measuring. It plays each matchup once, with no home-court advantage modifier, no momentum from a winning streak, no fatigue from a back-to-back. The real Warriors had all of those things working in their favor across the course of a season. A simulated win rate of around 83 percent for that roster, against those opponents, is actually a reasonable reflection of their underlying talent gap.
Steph Curry's numbers in the simulation tracked close to his real averages in most games. The engine captures his three-point tendency and off-ball movement well because both are reflected in his shooting attributes and assist frequency. Where the simulation undersells him is in the intangible — the way a real defense collapses around him changes the game for everyone else in ways a per-possession model can only partially account for.
The losses were interesting
Of the 14 simulated losses, about half came against teams the real Warriors also struggled with that year — Oklahoma City, San Antonio, and Cleveland. The engine correctly identified those as the most competitive matchups. A few of the simulated losses were genuine surprises, including one against a mid-table Eastern Conference team that happened to have a dominant interior player whose defensive rating neutralized the Warriors' paint attack for one game.
What this tells us about the simulator
The OBM engine is a possession-by-possession model that runs every play individually. It does not know the 2016 Warriors are historically great — it just processes the ratings. The fact that it reproduces roughly the right win distribution, identifies the right opponents as dangerous, and generates realistic box scores is the real validation. Try it yourself at the simulator page.