The 2026 Stanley Cup Final: A Data Analysis

How the analysis works in 30 seconds

Every input is real 2025-26 data, not a reconstruction. Team strength comes from expected goals (xG); goaltending from goals-saved-above-expected (GSAx); goalscorer rates from per-player goals and ice time. The league averaged 3.13 expected goals per game this season, and each team is measured against that baseline.

• Carolina attack 1.17x league, defense 0.92x (below 1 is good: they suppress chances). Season xG share 56%.
• Vegas attack 1.03x league, defense 0.88x. Season xG share 54%.

For each game the model sets expected goals for both sides (adjusted for the opposing starter and home ice), turns that into an exact single-game win probability with two Poisson distributions, and then enumerates the entire best-of-7 outcome space. There is no Monte Carlo simulation: the math is exact, so the numbers are reproducible every time the page builds.

The series after Game 3: Vegas 61%

Game 3 update: Vegas won a double-overtime classic 5-4 to take a 2-1 series lead. The Golden Knights built a 4-0 lead behind a Mitch Marner hat trick, Carolina erased all of it (three goals in 39 seconds in the third, then a Svechnikov power-play tie with 1:42 left in regulation), and Shea Theodore ended it 5:38 into the second overtime. Holding serve at home flips the projection: with Vegas up 2-1 and the math now run over the remaining best-of-four, the recomputed numbers are Vegas 61.1%, Carolina 38.9%. We re-run the model on every result. Carolina stays clearly alive: a Game 7 would be in Raleigh, and the run of play in Game 3 (a four-goal comeback) matched the chance-generation the season numbers describe.

The most likely series outcomes from here, in order:

The series-length distribution (how many games it takes, regardless of winner):

What the 2-1 lead means

Why the lead moves the number so much

The pre-series case for Carolina was home ice: as the better regular-season team (113 points; 29-10 at home), the Hurricanes host a potential Game 7 in Raleigh. That edge is real but small per game; a series lead is worth more. By winning two of the first three, including the only game so far at T-Mobile Arena, Vegas now needs two wins from a possible four to close it out, with two of those at home. The model rates Carolina about 59.0% per home game versus 50.8% on the road, and that home-road gap is exactly why Carolina is still alive at 39%: their two best remaining win-equity spots (Games 5 and 7) are in Raleigh.

The underlying team quality barely moved

Three one-goal-class games do not change a full season of process. Carolina posted a 56% expected-goals share this season (Vegas 54%), and the Game 3 comeback from 4-0 down showed the chance-generation that number reflects. The model still expects close, high-event games. What changed is not the per-game matchup but the scoreboard: Vegas needs two wins, Carolina needs three, and that gap is most of the 61-39 split.

Carolina's path: get to Raleigh

Game 4 is back at T-Mobile Arena, where Carolina sits at that 50.8% road number. Win it and the series is even with the home-ice math fully back in play; lose it and Vegas can serve for the Cup in Game 5. The 61% is a clear favorite, not a lock: Carolina has already won once in Vegas-style hostile spots this run, and a best-of-four with two potential Raleigh games is far from over.

Expected goals: Game 4 in Las Vegas

Game 4 is on the road, where the model expects a near-even game: 3.14 goals from Carolina and 3.09 from Vegas. That is the cost of playing at T-Mobile Arena, and why this is a must-not-fall-behind-3-1 game for Carolina. Back home (Games 5 and 7 if needed), the Hurricanes open a clearer edge: about 3.40 to 2.86 in their favor.

The most likely goalscorers

For each skater, the model blends actual and expected goal rate, regresses small samples toward a position-appropriate baseline (so a hot fourth-liner is not overstated), scales by expected ice time, and adjusts for the opposing defense and goalie. The result is each player's probability of scoring in a representative game.

A note on goaltending (the honest part)

Goaltending is the highest-variance input in playoff hockey, and it is where this analysis is most uncertain, so we are explicit about it. Vegas has gone with Carter Hart, not Adin Hill, so the model uses Hart as the Vegas starter. The two current starters' regular-season goals-saved-above-expected:

• Frederik Andersen (CAR): raw season GSAx/60 -0.094 (slightly below average over 35 games). He was pulled to start the third period of Game 3 after four goals on 16 shots, a strategic move with Carolina trailing rather than an injury. His playoff run remains a small sample.
• Carter Hart (VGK): raw season GSAx/60 -0.247 (below average over 18 games), and he has backstopped two of Vegas's first three wins, including 29 saves in the Game 3 double-overtime win.

Because one season of goalie GSAx has low year-to-year reliability (about 0.35), the analysis regresses both figures to 35% of the observed value rather than taking them at face value. This deliberately mutes the goaltending signal: neither goalie is rewarded or punished much for a noisy season number, and the projection rests on team quality and the series scoreline rather than a bet on which goalie gets hot.

What this analysis is, and is not

It is a transparent, real-data projection, re-run after each result. It is not a guarantee, and it cannot be "right" about a single series: a best-of-7 is one sample, and a 61% favorite still loses 39% of the time. The value is in calibration across many projections, not certainty on one, and a single hot goaltender can break any model in this sport.

Frequently asked questions

What data does this Stanley Cup Final analysis use?

Real 2025-26 regular-season data: official team records from the NHL API, and expected goals (xG), goalie goals-saved-above-expected (GSAx), and per-skater goal rates and ice time from MoneyPuck. Every input traces to a public source, stamped 2026-06-02.

How is the series projection calculated?

Each team gets an attack and defense rating from its season xG relative to the league average of 3.13 expected goals per game. For each of the seven games, the model sets expected goals for both sides (adjusted for the opposing starter and home ice), computes an exact single-game win probability from two Poisson distributions, then enumerates the entire best-of-7 outcome space analytically. There is no simulation, so the numbers are exact and reproducible.

Has the projection changed during the series?

Several times, which is the point of re-running it. Pre-series: Carolina 64% on home ice. Vegas won Game 1 to flip it; Carolina won Game 2 in OT to flip it back; Vegas won Game 3 in double OT to take a 2-1 series lead. The recomputed projection is now Vegas 61% and Carolina 39%, computed over the remaining best-of-four from the 2-1 scoreline. Carolina stays live mainly because a Game 7 would be in Raleigh.

How does the analysis handle goaltending?

Carefully, because it is the highest-variance input in playoff hockey. The model uses the actual playoff starters (Frederik Andersen for Carolina, and Carter Hart, who started Game 1 for Vegas) but regresses their season GSAx heavily toward league average, because one season of goalie GSAx has low year-to-year reliability (about 0.35). It applies 35% of the observed figure.

How reliable is a single-series projection?

A best-of-7 is a single sample; a 61% favorite still loses 39% of the time. The value of a model like this is calibration across many projections, not certainty on one. The goalscorer figures are the most testable: across a full slate, players given a 30% chance to score should score about 30% of the time.

Data sources: NHL API (api-web.nhle.com/v1/standings) for official records; MoneyPuck.com season summaries (teams, goalies, skaters) for xG, GSAx, goal rates, TOI. Season: 2025-26 regular season. Stamped 2026-06-02.