3-5: Final (SEA). 39-34: Shots on goal. 57%: Ottawa Corsi. 3.24-3.24: xG (excl. EN) Tuesday night at TD Place, the Ottawa Charge out-attempted the Seattle Torrent 39-34 on shots, won the Corsi battle 57%-43% , and lost 5-3. That's the setup for the argument every coach has heard: the possession metrics say we played well, the scoreboard says we didn't, somebody got unlucky. PWHL-calibrated xG, the model fit on 16,709 non-EN PWHL shots in the previous post , grades this one differently. Possession isn't the story. Chance quality was close to a dead heat. Finishing, and one hook with 3:17 left, did the damage. Strip the empty net and it's a draw Raw xG: Seattle 3.24, Ottawa 3.24 . A 0.94 gap, which looks like Seattle generated the clearly better chances. The wrinkle is where that gap lives . Cayla Barnes 's empty-net goal at 19:41 scored off a shot the model prices at 0.00 xG . An open net is a near-certain goal by construction, and any xG model that didn't have empty-net as a feature would be broken. Remove that single open-net attempt and the xG battle collapses: Seattle non-EN xG: 3.24. Ottawa xG: 3.24. Difference: 0.00, well inside the noise of a single-game xG estimate. Stripped of the empty-netter, the two teams manufactured essentially the same quality of chance across sixty minutes. Corsi flattered Ottawa because more of their attempts came from the perimeter. High-danger count tells the same story xG does: Seattle 10 HD chances, Ottawa 8 . More total attempts, fewer from the inside. The game in cumulative xG The scoring timeline P1 1:54 · Fanuza Kadirova · xG 0.11 · 1-0. P1 13:09 · Jada Habisch · xG 0.11 · 1-1. P2 5:31 · Danielle Serdachny · xG 0.15 · 1-2. P2 9:46 · Kateřina Mrázová (PP) · xG 0.19 · 2-2. P3 2:02 · Danielle Serdachny · xG 0.06 · 2-3. P3 8:51 · Gabbie Hughes · xG 0.17 · 3-3. P3 18:39 · Natalie Snodgrass (PP) · xG 0.07 · 3-4. P3 19:41 · Cayla Barnes (EN) · xG 0.00 · 3-5 Six of the eight goals were scored at 5-on-5 on chances the model priced at 0.04 to 0.17 xG . Neither team cashed a gimme from the slot; both converted on secondary looks and broken plays. That's an Ottawa-style game on paper. The Charge's scoring profile this season has leaned on middle-danger volume, not high-danger bursts, and it's the kind of game where finishing rate on small samples becomes the difference. The 3:17 mark Tied 3-3 after Gabbie Hughes 's 5-on-5 equaliser at 8:51 of the third, the game flowed end-to-end for six scoreless minutes. Then, at 16:43 , Emily Clark took a hooking minor. Natalie Snodgrass redirected a point feed past Gwyneth Philips for the go-ahead goal 1:56 later, four seconds before the penalty expired. The model priced the shot at 0.058 xG : a low-percentage look that she finished. Ottawa pulled the goalie with 55 seconds left, and Cayla Barnes found the open net 36 seconds after that. This is the single sequence a video coach would open up first. Not because the goal is particularly instructive on its own. A redirected point shot at the end of a successful kill is exactly the kind of play you accept on the PK. The penalty that led to it is the clearest coachable moment in the game. 3-3 with 3:17 left, a shift-length battle for territory, and a stick gets into the hands. That's the clip. Finishing, not deserving Ottawa scored on 3 of 39 non-empty-net shots, a 7.7% rate on 3.24 xG, which works out to 0.93× expected . The model called their output almost exactly right. Seattle scored 4 on 33 non-EN shots, 12.1% on 3.24 xG, or 1.23× expected . That's the entire margin. Over 82 games it averages out; over one game it's the difference between a tie and a regulation loss. Ottawa's highest-danger misses The five shots the model graded as Ottawa's best chances that didn't score. Every xG in the list is an attempt the average PWHL shooter converts between 10% and 33% of the time. On another night, one of these goes in and the finishing ratios look different. P2 7:15 · Rebecca Leslie · xG 0.35. P1 19:28 · Emily Clark · xG 0.16. P1 15:10 · Rebecca Leslie · xG 0.16. P2 2:04 · Sarah Wozniewicz · xG 0.15. P2 17:10 · Rebecca Leslie · xG 0.13 What this read isn't Not a claim Ottawa deserved to win. They didn't. An even xG battle with worse finishing is a loss; there's no asterisk on that scoreboard.. Not a claim xG is the only number that matters. Goalie-specific inputs, traffic in front, pre-shot puck movement past three seconds, and handedness all change the chance quality read. PWHL feed coverage on these is thin; see the xG methodology post .. Not a single-game indictment of anyone. Single-game xG vs. goals has wide variance on both sides. The value of the read is in a season-long pattern of over- or under-finishing, not one night.. Not an argument against the penalty call. The Emily Clark hook at 16:43 was called. The coachable moment is the stick position with 3:17 left at 3-3, not the official's judgement. Methodology xG model : PWHL-calibrated logistic regression on 16,709 non-EN shots. Feature set: distance, angle, rebound (≤3 s), empty net, penalty shot, power play, six shot-type dummies. Held-out 5-fold log-loss lift vs. league-base-rate null is 12.4%.. Corsi / Fenwick / HD counts : derived from the same PBP feed. Corsi counts blocked shots; Fenwick strips them; HD is the feed-provided shot-quality tag ("Quality on net") agreeing with the model for this game.. Finishing ratio : observed goals divided by summed xG for that team's attempts, excluding empty-net in the denominator. Single-game finishing ratios are noisy by construction; treat the number as a description of one night, not a forecast.. Numbers regenerate on every build. If the PBP is re-synced or the xG model is recalibrated, every figure in this post re-computes from scratch on the next static export. The full play-by-play, line-by-line, is at /games/102 , including shift-by-shift overlays and the Sportscode XML download.