40: Players signed away. 12.7: Projected WAR lost. 7/24: Our “safe” calls that walked. 38%: Signed who were UFAs In May we ran every existing PWHL roster through our value model and published projected expansion-protection lists at /front-office/expansion . The expansion is now finished. Detroit, Hamilton, Las Vegas, and San Jose each built a ten-player roster out of the eight existing teams across three signing windows, Phase 2 through Phase 4, in June. 40 players left their old clubs. This post grades our model against that. One rule governs everything below, so we lead with it. We can see who was signed. We cannot see who was protected. Teams did not publish their protection lists. A player an expansion team signed was, by definition, left unprotected. A player who was not signed tells us nothing: she may have been protected, or exposed and simply passed over. So we only ever count confirmed signings. We never compute a recall, a precision, or a hit rate over our predicted losses, because the thing those numbers divide by is invisible. What we can do is fair, and it is enough. Where the talent went, and through which door Start with pure fact. For each existing team, add up the projected WAR (wins above replacement, our single-number value estimate) of every player it actually lost, and split it by the contractual door she left through: under contract or restricted, versus a pending unrestricted free agent. This makes no prediction. Every player in the chart signed with an expansion team. The only thing ours is the valuation. Across the league, 25 of the 40 signings were players under contract or restricted , and 15 were pending unrestricted free agents (38% by headcount). That split matters for our model. The simulator pools only contracted and restricted players when it guesses who gets signed away, on the logic that a pending UFA is a separate problem (you cannot keep her without a new contract). The data says that instinct was directionally right: the larger share of the talent did leave through the contracted door, which is the one the model watches. The trouble is what the model did with the other door. Where the model said “safe” and was wrong This is the audit the censored data does allow. Our model named a three-player Phase 1 protection for each team, 24 players it judged safest to keep. 7 of them were signed away by an expansion team , which means their old club did not protect them. These are clean errors: the model said keep, the team exposed, the player left. Nicole Gosling (MTL, under contract, 0.80 projected WAR) : our model slotted her into MTL's three-player Phase 1 protection. Hamilton signed her in Phase 2.. Daryl Watts (TOR, pending UFA, 0.69 projected WAR) : our model slotted her into TOR's three-player Phase 1 protection. Detroit signed her in Phase 2. She received one of the three Expansion Foundational Offers, the binding instrument a player cannot refuse without a matching deal.. Britta Curl-Salemme (MIN, under contract, 0.58 projected WAR) : our model slotted her into MIN's three-player Phase 1 protection. Detroit signed her in Phase 2.. Allyson Simpson (NY, under contract, 0.58 projected WAR) : our model slotted her into NY's three-player Phase 1 protection. Hamilton signed her in Phase 4.. Brianne Jenner (OTT, pending UFA, 0.54 projected WAR) : our model slotted her into OTT's three-player Phase 1 protection. Hamilton signed her in Phase 2.. Anne Cherkowski (NY, RFA, 0.53 projected WAR) : our model slotted her into NY's three-player Phase 1 protection. San Jose signed her in Phase 2.. Hilary Knight (SEA, pending UFA, 0.47 projected WAR) : our model slotted her into SEA's three-player Phase 1 protection. Las Vegas signed her in Phase 2. She received one of the three Expansion Foundational Offers, the binding instrument a player cannot refuse without a matching deal. The pattern is the contract column. 3 of the 7 were pending free agents, and the model protected them anyway, ranking them on talent as if keeping them were free. It is not. A pending UFA has to be re-signed to be protected, and the most coveted UFAs drew binding Expansion Foundational Offers they could not simply turn down. Ranking a star UFA into a costless protection slot is the single clearest thing this model gets wrong, and it is exactly the fix our roster-value notebook started on: price the re-sign, do not assume it. The other 4 (Nicole Gosling, Britta Curl-Salemme, Allyson Simpson, Anne Cherkowski) were under contract or restricted, players the model valued above where their own team did. Those are honest talent-versus-priorities disagreements, the same kind the draft post turned up. Where the model was right The flip side, and the only other thing the data lets us claim. Some players the model flagged as exposed (either a simulated Phase 2 loss or the single most painful cut it could not protect) did in fact sign with an expansion team. We name the confirmed ones and stop there. We do not report how many of our predicted losses came true, because the ones that did not sign are not misses, they are unknowns. Kendall Cooper (MIN, 0.55 projected WAR) : flagged by the model as exposed, signed by Las Vegas in Phase 2.. Mae Batherson (MIN, 0.44 projected WAR) : flagged by the model as exposed, signed by Las Vegas in Phase 2.. Rory Guilday (OTT, 0.44 projected WAR) : flagged by the model as exposed, signed by San Jose in Phase 2. What this read isn't Not a hit rate. Because protection lists are private, a player we predicted to lose who was not signed is an unknown, not a miss. In fact several players our model expected to be lost were publicly confirmed as protected (Abby Newhook, Susanna Tapani, Hannah Miller, Kali Flanagan), which is exactly why we refuse to score that side of the ledger.. Not a claim about which team signs whom. Our simulator assigns losses to a generic pool of expansion teams; it does not predict that a specific club signs a specific player. The expansion team named beside each player is the real signing, not a model output.. Our valuation, real signings. WAR is our estimate of value and carries real error, especially for goaltenders and small samples. The signings are facts. When the two disagree, that is a flag to examine, not a verdict.. Contract status is frozen pre-expansion. Every UFA or contracted label reflects a player's standing going into the expansion window, not after it.. The model's appetite is an assumption. How many players the expansion teams would sign, and how that spreads across clubs, is a tuning choice in our simulator, not a league rule. The one hard rule we lean on is the league's stated cap: an existing team could lose at most four players who were under contract for 2026-27. Methodology The model. Each team's Phase 1 protections are its top three players by regressed, pre-aging projected WAR. The simulator then pools every unprotected contracted or restricted player league-wide and assigns the most valuable as expansion signings, before each team protects three more. Full machinery at /front-office/expansion .. The signings. All 40 players signed by Detroit, Hamilton, Las Vegas, and San Jose across Phases 2 to 4, taken from the PWHL.com phase-completion releases. The three Expansion Foundational Offers (Daryl Watts, Hilary Knight, Kristin O'Neill) are flagged as such.. The valuation. Projected WAR is each player's estimated next-season wins above replacement. For the talent-drain chart we floor it at zero, so a below-replacement depth signing does not subtract from a team's loss total.. Matching. Signed players are joined to our value model by normalized name; on this build all 40 matched, so there is no coverage gap to discount.. Numbers regenerate on every build. A WAR recalibration or a contract-status update reshapes the chart and the named lists on the next export. The live protection simulator is at /front-office/expansion . Our companion grade of the model on the 2026 entry draft is at the draft post-mortem . The next version of the expansion model should price the cost of re-signing a pending free agent instead of protecting her for nothing.