Inside PWHL Roster Construction

The Average Salary System

The PWHL does not use a hard salary cap. Instead, it uses an average salary system: each team's average player salary must fall within 10% of a league-set target. For 2025-26, that target was $58,350 per player on a 23-player active roster, creating an effective team spending band of roughly $1.21M to $1.46M.

There is no individual salary maximum. Emily Clark earns $126,090 and that's fine, as long as Ottawa's team average stays within the band. But the system has structural constraints that force a specific roster shape:

• At least 6 players per team must earn $80,000+, established through mandatory three-year contracts in the inaugural season.
• No more than 9 players can earn the league minimum ($37,132).
• All figures escalate 3% annuallythrough the CBA's 2031 expiration.

The result is a system that constrains roster shape, not just total spending. You cannot stack stars and fill with minimum-salary players. You cannot avoid investing in top-end talent. You have to build a middle class. The question is how teams navigate the space within those rules.

How Eight Teams Built Their Rosters

In April 2026, the PWHLPA published every player's base salary. This was the first time a professional women's hockey league had full salary transparency. Across 194 players, the range is $37,132 to $126,090. Half the league earns under $50,000. Only ten players earn six figures.

Within these constraints, teams took noticeably different approaches:

Boston and Minnesota are the balanced builds. Both hit exactly 6 players at $80K+, maintain a deep middle tier (9-10 players in the $45-80K range), and use 9 roster spots at the bottom. Their top five players consume about a third of payroll. Both finished in the top three.

Ottawa and New York went top-heavy. Ottawa's top five players (Clark, Jenner, Hughes, Philips, Larocque) consume 37.6% of payroll, the highest concentration in the league. New York has 14 players under $45K. Both teams paid for star power at the top and filled the bottom of the roster with near-minimum contracts.

Seattle went the opposite direction. Only two players earn $80K+ (Knight at $106K, Carpenter at $90K), but 16 players sit in the $45-80K middle tier. That is the largest middle class in the league. Seattle's median salary ($54K) is the highest of any team. They spread money evenly rather than concentrating it. They also finished last.

Vancouver invested at the top differently. Seven players at $80K+, the most in the league, but no single player above $90K. Instead of one or two marquee contracts, they distributed their high-end spending across more players. The tradeoff: 10 players under $45K.

Outperforming the Contract

With WAR derived in Part 4, we can measure contract efficiency in wins, not just goals. We express value as WAR per $100K of salary. Higher is better.

Claire Butorac earned $40,000 and produced +0.99 WAR. That is 2.48 wins of value per $100K spent. Amanda Boulier at the same salary produced +0.93 WAR. Both delivered nearly a full win above replacement at minimum salary. Nylén Persson led the league in total skater WAR (+1.05) at $54,000.

Minnesota again placed two players (Butorac and Curl-Salemme) in the top ten for value. Their balanced roster construction pays off: invest enough in the middle tier to find productive players at reasonable salaries.

Goalies dominate the per-dollar rankings even more dramatically:

Kirk at +3.16 WAR/$100K leads all players. She earned near-minimum salary ($37K) and produced +1.18 WAR behind a middling Toronto defense. Desbiens (+2.96) and Frankel (+2.51) provided even more total value at higher salaries. Goaltending remains the single most efficient way to buy wins in the PWHL. Osborne at the other end: $39,000 for -0.77 WAR, the league's worst goaltending value.

Falling Short of the Contract

Ten skaters earn $100,000 or more. None produced more than +0.40 WAR at 5v5. Most are near zero or negative.

Read this with context. WAR measures 5v5 impact only. Fillier led the league in individual xG (9.9). Clark and Jenner are among the league's most productive scorers. Their value comes in part from power play production, which this model excludes.

Stars also draw the toughest defensive assignments. RAPM includes quality of competition as a control variable, but with heavily regularized coefficients across 120 games, it cannot fully separate matchup difficulty from individual performance.

What we can say: at 5v5, the highest-paid players are not producing the highest WAR. Poulin (+0.40) and Keller (+0.38) are the only six-figure skaters producing even a third of a win above replacement. Roque (-0.36) and Fast (-0.19) are the clearest cases where salary and 5v5 production diverge.

The structural reason this gap is limited: when the highest salary is 3.4x the lowest, the most anyone can be “overpaid” is about $89,000. In the NHL, that gap can be $9 million. The PWHL's compressed salary structure limits how much any team can gain through value identification. Butorac outproduced Clark by nearly a full win at one-third the salary. Real money for the player, but not the kind of roster arbitrage that wins championships in salary-capped leagues with wider ranges.

Team Payroll Efficiency

How efficiently did each team convert its payroll into wins above replacement?

Boston spent $159K per win above replacement. Seattle spent $286K. The difference is not total spending (payrolls range from $1.33M to $1.45M, a narrow band). The difference is how well each team identified productive players within the CBA's constraints.

Boston's efficiency comes from Frankel (+2.32 WAR at $93K) plus near-minimum contributors like Maloney and Eldridge. Montréal's comes from Desbiens (+2.67 WAR at $90K) and Boulier (+0.93 WAR at $40K). Both teams found their surplus value in goaltending and in the middle of the roster, not at the top.

Limitations

• 5v5 only. WAR excludes power play and penalty kill. Stars who derive significant value from special teams are systematically undervalued. A PP/PK extension is feasible but would have noisier coefficients in a 120-game league.
• Salary reflects more than on-ice value. Poulin, Knight, and Fillier sell tickets, drive media coverage, and attract sponsors. Their salaries reflect market value that has nothing to do with 5v5 WAR. That is the salary structure working as intended, not a flaw.
• The CBA constrains optimization. The structural requirements (6 at $80K+, max 9 at minimum) limit how much teams can exploit market inefficiencies even when they identify them. The 3.4x salary spread means the financial stakes of any individual evaluation decision are small.
• 2025-26 rosters only. With four expansion teams entering for 2026-27 and significant player movement, these rosters no longer exist. The framework applies to a snapshot.

What We Built

This series set out to build a player evaluation framework for a two-year-old league with limited data. Starting from 11,625 shots of play-by-play data:

• Part 1 built an xG model that predicts goals within 0.3% of actual (993.2 vs 990).
• Part 2 used that model in a ridge regression (RAPM) to isolate each skater's 5v5 impact, controlling for linemates, opponents, score state, and venue.
• Part 3 evaluated goaltending independently via GSAx, revealing that three of five protected goalies were genuinely elite and two were replacement-level.
• Part 4 derived WAR from first principles: Pythagorean expectation, goals per win, and replacement level calibrated to the PWHL. The combined framework explains 93% of standings variance and pool-balances at the league level.
• Part 5 connected WAR to salary. The best-value players are near-minimum skaters and elite goalies. Boston and Montréal spent roughly $165K per win above replacement. Seattle and Vancouver spent about $285K. The salary structure is too compressed to allow dramatic roster arbitrage, but the teams that identified the right players within those constraints won.

The framework is reproducible. When the 2026-27 season generates another 180+ games of play-by-play data with four new expansion teams, every number in this series can be updated and the WAR conversion recalibrated. That is the point: not a one-time ranking, but a system that improves with more data.