Shai Gilgeous-Alexander walked off the Paycom Center floor with the Bill Russell trophy in his hands at 11:47 PM local time on June 22, 2025, the Oklahoma City crowd six rows deep behind the player who had just finished a Game 7 with 29 points, 12 assists, and the cleanest individual line of any Finals MVP since LeBron’s 2016 closeout. The final was Thunder 103, Pacers 91. The series had been the rare Finals where the favorite was tested across seven games without ever losing the underlying numbers — Indiana’s depth had stretched the series to its limit and stopped exactly there, with Tyrese Haliburton on the bench in the fourth quarter of Game 7 with a torn Achilles he had been hiding for two games and OKC closing out the game the way the regular-season models had said they would.
The 2025 Thunder won 68 regular-season games. They had the league’s best net rating, the best defensive rating, the MVP, and a roster construction that the analytics community had been treating as the modern blueprint for an entire calendar year before the playoffs started. None of that guaranteed the championship. The Finals are the rare sample where the favorite still loses about 30% of the time. OKC won because the underlying numbers were correct about them across both the regular season and the playoff path, and because Indiana’s depth model — which I covered in a piece three days after the Finals ended — could not survive a healthy top-three opponent at the league’s ceiling.
What follows is what made the OKC title an unusually clean alignment between regular-season indicators and championship outcome, where the alignment broke down in specific moments, and what the title actually tells us about how the modern NBA is being won.
The regular-season signals that traveled cleanly
OKC entered the playoffs with a regular-season net rating of +12.7 per 100 possessions, the highest in any season since the 1996 Bulls. The defensive rating was 106.4, also the league’s best, and the on/off split for SGA was +9.2 — the largest in the league for any player with at least 2000 minutes. The model-implied championship probability heading into the playoffs sat around 35%, which is the highest pre-playoff number any team has carried into the postseason since the late-Warriors era.
Those signals predicted the playoff path with unusual precision. OKC swept their first-round opponent. They handled the second round in five games. They closed the Western Conference Finals in six. By the time the Finals started, the Thunder had played 17 playoff games, won 12 of them, and held a net rating of +9.8 across the bracket — only slightly lower than the regular-season figure, which is unusual for a deep playoff run.
The Indiana series was the test. The Pacers had built around a different model — depth, ball movement, Haliburton — that had carried them through three rounds against the East’s middle class. The Finals matchup was the year’s cleanest experiment in “does star-plus-defense beat depth-plus-pace?” The answer was yes, in seven games, with the caveat that Haliburton’s Achilles injury removed the depth model’s safety valve in the fourth quarter of Game 7.
Where the alignment between regular-season indicators and championship outcome broke down
The regular-season models had OKC closing out the series in five or six games. The series went seven. That gap is not a model failure — Finals models are calibrated for 70-30 favorites to lose roughly 30% of the time — but it surfaces where the alignment was not perfect.
| Indicator | Regular-season number | Finals number | What changed |
|---|---|---|---|
| Net rating | +12.7 | +1.2 | Indiana’s depth produced two unexpected wins |
| SGA TS% | 63.7% | 57.4% | Indiana’s halfcourt defense was better than expected |
| Bench points/G | 34.1 | 26.8 | Pacers depth compressed OKC bench impact |
| Three-point % | 38.4% | 34.1% | Finals variance, partially Indiana defense |
The Finals number for each indicator was worse than the regular-season number, by a meaningful margin in three of the four. That is what playoff variance does even to favorites. The series went seven because OKC’s regular-season ceiling had been built against a league average that was not the same as Indiana’s specific defensive scheme. The Pacers had figured out enough of the SGA double-team coverage to drag the series past the model’s expected length. They had not figured out enough to actually win it.
Where this gets weird
The clean “OKC’s numbers were right all along” reading misses three nuances that the broader Finals coverage flattened.
The first is that Haliburton’s Achilles injury fundamentally changes how we should read Game 7. The injury appears to have happened in Game 6, with Haliburton playing through it. The Pacers’ Game 7 game plan had been built around Haliburton’s expected fourth-quarter usage; with him at compromised fitness in the third quarter and unavailable in the fourth, Indiana’s depth model lost its highest-impact piece. A counterfactual Game 7 with a healthy Haliburton was probably still an OKC win — the regular-season numbers point that way — but the margin would have been smaller and the series win probability for the Thunder in that specific moment lower than the 100% it eventually became.
The second is that the OKC depth model is itself one of the under-discussed aspects of the title. The Thunder’s bench produced the second-most points per game of any team in the 2025 playoffs (Indiana was first). The framing of the Finals as “Indiana’s depth vs OKC’s stars” was structurally wrong. OKC had star-plus-depth. Indiana had depth-plus-Haliburton. The depth advantage that the Pacers narrative implied did not actually exist at the series level. OKC matched it.
The third is that the +12.7 regular-season net rating, which the analytics community has been citing as historically dominant, is partly a function of opponent quality. OKC played the Western Conference in a year where the conference was historically deep — six teams won 50+ games — but they also went 32-2 against teams with sub-.500 records. The padding against bad teams inflated the season-long number. The Finals net rating of +1.2 is closer to the version of OKC that shows up against playoff-caliber opposition. Both numbers are real. The first one is regular-season-quality-adjusted; the second is Finals-opponent-adjusted. The 2025 Thunder were great. They were not as historically dominant as the +12.7 alone suggests.
What the title means for the 2025-26 NBA season
- OKC is the 2025-26 favorite by a margin not seen since the prime Warriors years. The roster returns intact. Wilson age curve aside (he is 26), the team should improve by another two-to-three points of net rating. The model already has them as a 35% championship favorite for the season that has not yet started.
- The depth model needs star context to win titles. Indiana’s run was the cleanest test of “can depth alone beat star-plus-depth?” The answer is no, with footnotes for injury context.
- The Western Conference compression continues. Six 50-win teams in one conference is the highest density in fifteen years. The 2026 race is going to look similar.
- SGA is the league’s best player and is not going anywhere. The on/off split, the Finals MVP, the clutch metrics, the model-favorite status — every layer of the evaluation points the same direction. The MVP race in 2026 is going to be between SGA and Jokic again, and the voter drift conversation I covered in the March piece is going to compound.
The callback
That moment at 11:47 PM in Oklahoma City, when SGA walked off the floor with the Bill Russell trophy six rows deep into a home crowd that had waited fifteen years for it, was the cleanest model-aligned title in modern NBA history. The Thunder did not steal the championship from the underlying data. They confirmed it across 19 playoff games against an Indiana team that had built the deepest playoff rotation in twenty years and lost anyway. The +12.7 regular-season net rating was inflated by opponent quality. The +1.2 Finals net rating was honest. Both are part of the picture. The next time someone tells you analytics never predict championships, point them at the 2025 OKC season. The numbers had this right from Game 1 of the regular season. They were going to have it right at Game 7. The path from one to the other took 96 games to play out, and the model survived every single one of them. The playoff scaling piece covers the broader framework. SGA closed the deal. The Thunder are the dynasty the model said they were.
Series data via Basketball Reference; advanced impact metrics via Cleaning the Glass; clutch tracking via NBA.com.
