98%: NCAA skaters matched. 0.51: Rank agreement (confident). 3: Top picks model deferred on. 0: PWHL shifts played On June 17 in Detroit, twelve PWHL front offices spent six rounds and seventy-two picks turning college and pro prospects into draft rights. Our Big Board at /draft had spent the spring ranking the same NCAA pool on a single number: projected PWHL points per game. The draft is the first time we get to hold the two side by side. Before anything else, the honest framing: this is not a report card. Not one of these players has taken a PWHL shift, so there is no outcome to grade. All we can measure today is how closely our order matches the order the GMs drafted in, and what to make of the places it does not. Start with coverage, because it bounds everything that follows. Our board ranks NCAA skaters only. Of the 55 NCAA skaters drafted, we matched 54 (98%). The skaters we cannot speak to are the 9 drafted out of European pro leagues and U Sports, plus every goaltender (eight of them), who live on a different projection scale entirely and are left out of every number here. So this is a read on the NCAA skater class, not the whole draft, and it is quietest exactly where our data runs out: the European and U Sports picks that cluster in the later rounds. How much does our number track the room? The single summary stat is rank agreement, measured as a Spearman correlation: 1.0 means our board and the draft put the players in the exact same order, 0.0 means no relationship, and negative would mean we were backwards. Across every matched skater it is 0.40 . But that number is dragged down by a handful of players the model openly flagged as low-confidence (more on them below). Restrict to the projections the model actually stood behind, the high and medium confidence skaters, and agreement rises to 0.51 : 0.66 for forwards and 0.60 for defenders. That is a moderate, positive relationship. The board and the GMs are looking at a lot of the same thing, which is reassuring, and disagreeing plenty, which is the interesting part. Agreement is loudest at the very top. Caroline Harvey (pick 1), Abbey Murphy (pick 2), Laila Edwards (pick 4), Lacey Eden (pick 5), Kirsten Simms (pick 8) are players our board and the draft put in nearly the same place. When a prospect is genuinely elite, a points model and a room full of scouts tend to land together. The disagreements live in the middle, where most of a draft actually happens. Where the model deferred, and the GMs did not The loudest apparent disagreements are not really arguments. They are the model shrugging. Our projection works by translating college scoring into a PWHL rate using players from the same conference who already made the jump. When a conference has sent almost no one to the PWHL, there is nothing to translate through, so the model shrinks those players hard toward zero and stamps the projection low-confidence. The board then buries them, even when their college production was excellent. The draft is a clean reality check on exactly these players, because the GMs were not limited to a thin translation sample. Tessa Janecke (F, CHA, pick 3) : best NCAA season 1.55 P/GP, projection 0.07 P/GP at low confidence. The board is not saying she is bad. It is flagging that it has too little history from her conference to translate her scoring with any confidence.. Leah Stecker (D, CHA, pick 27) : best NCAA season 0.51 P/GP, projection 0.16 P/GP at low confidence. The board is not saying she is bad. It is flagging that it has too little history from her conference to translate her scoring with any confidence.. Madelyn Christian (F, CHA, pick 33) : best NCAA season 0.90 P/GP, projection 0.04 P/GP at low confidence. The board is not saying she is bad. It is flagging that it has too little history from her conference to translate her scoring with any confidence. The pattern is not random. The deferrals cluster in College Hockey America (CHA), the conference Penn State plays in. CHA has sent a tiny handful of players to the PWHL, so the model's CHA-to-PWHL translation for forwards rests on a single-digit sample and lands near zero. Feed a genuine CHA scorer through it and you get the deferrals above: Tessa Janecke , Madelyn Christian . The third overall pick is the headline. Tessa Janecke scored better than a point and a half per game as a Penn State senior and was a steady CHA producer for four years, yet our number sits near the bottom of the forward board at low confidence. That is not a read on Janecke. It is the model telling you, through the confidence flag, that it cannot see CHA. The GMs taking her third is strong evidence the flag was right and the rank was meaningless, which is exactly why these players are kept out of the agreement numbers above. It is also a concrete fix for next offseason: a better prior for thin conferences. The real disagreements, among projections the model stood behind Strip out the deferrals and you are left with the honest disagreements: players the model rated with real confidence, where our board and the room still parted ways by a wide margin. These are the ones worth arguing about, because both sides actually had a conviction. Drafted well ahead of our board. When a GM reaches past a points projection, it is usually for something the projection cannot encode: a left-shot defender in a league short on them, a power forward's frame, national-team pedigree, a two-way role, or simple signability. Avi Adam (F, ECAC, pick 24, 0.15 projected P/GP) : a Cornell forward whose point totals the model reads as middle of the pack, taken in the second round. The likely gap: the room is paying for a role and a two-way game that a points projection does not capture. Ranked higher by our board than by the room. These are the model's quiet bets: confident projections that the draft passed on. Some will look smart in two years. Some are the model trusting a gaudy rate built against a soft schedule that scouts correctly marked down. Alexis Petford (F, ECAC, pick 53, 0.26 projected P/GP) : a Colgate forward (she transferred up from Stonehill) the board rates as a real bet on the strength of her ECAC scoring, who slid to the back of the fourth round. If the model is right, this is a value pick; if the room is right, it saw something in her game the numbers do not.. Taylor Otremba (F, WCHA, pick 71, 0.17 projected P/GP) : graded above her draft slot by our board. Could be value the room passed on, could be a schedule-strength or signability read we cannot see.. Reichen Kirchmair (F, Hockey East, pick 64, 0.21 projected P/GP) : graded above her draft slot by our board. Could be value the room passed on, could be a schedule-strength or signability read we cannot see.. Emerson O'Leary (F, ECAC, pick 56, 0.21 projected P/GP) : a Princeton forward the board graded above her draft slot. A confident projection the room was cooler on, for reasons that may live off the scoresheet.. Gracie Gilkyson (D, ECAC, pick 50, 0.21 projected P/GP) : a puck-moving ECAC defender (Yale) our board liked more than the room did. Defenders are the hardest group for a points model to rank, so treat this as a soft disagreement. Ottawa's draft, through the model's eyes The Charge made 6 selections. Here is how our board saw the ones it could see (her European picks sit outside our NCAA data, so the model has no opinion there): Vivian Jungels (D, pick 11, WCHA, 0.11 projected P/GP, high confidence). Jordan Ray (F, pick 23, ECAC, 0.23 projected P/GP, high confidence). Tory Mariano (D, pick 47, Hockey East, 0.03 projected P/GP, medium confidence). Taylor Otremba (F, pick 71, WCHA, 0.17 projected P/GP, high confidence) Vivian Jungels at 11 is the instructive one. Our board, which pays for points, has her as a depth defender; Ottawa took her in the first round for a defensive, projectable game a college scoring line does not capture. Jordan Ray at 23 is the opposite kind of pick, a Yale forward our model already liked, so the board and the front office agreed. That split is the whole post in miniature: a points model and a hockey-ops department are measuring overlapping but different things. What this read isn't Not a verdict on anybody. No PWHL games have been played by this class. Every line above is about agreement between two pre-season opinions, not about who was right. We will come back to this post after the 2026-27 season with actual production and see which disagreements the ice settled.. Not a complete draft. The board is NCAA-only, so European pro and U Sports picks are missing by construction, and goalies are excluded entirely because their save-percentage projection is not comparable to a skater's point rate. Coverage is best at the top of the draft and thinnest in the later rounds, which is exactly where we are least entitled to a strong opinion.. Not the GMs being wrong. Draft position encodes things our model never sees: position and handedness scarcity, age, frame, national-team reps, character, medicals, and whether a player will actually sign. A reach against our board is the rational price of information we do not have. Disagreement is the expected state, not an error.. Not a frozen pre-draft snapshot. The board recomputes from live NCAA data on every export. Because the college season is over, the inputs are stable, but this is a comparison against our current board rather than a sealed pre-draft file. Methodology The board. Skaters are ranked by projected PWHL points per game: college production translated through a conference-strength factor and an age adjustment, with sample-size shrinkage toward a position prior. Each projection carries a confidence level (high, medium, low) that reflects how much clean data backed it. Full board at /draft .. Matching. Drafted players are joined to the board by normalized name, with a small alias table for known spelling variants and a surname fallback. European pro and U Sports picks do not appear on an NCAA board and are reported as coverage misses, not model failures.. Agreement. Spearman rank correlation between our board rank and actual overall pick, computed over matched skaters. The headline number uses only high and medium confidence projections, so a few low-confidence collapses do not masquerade as the model disagreeing with the room. Forwards and defenders are correlated within position; the two are never ranked against each other.. Flagging. A divergence is highlighted only when a confident projection sits at least ten places off its within-position draft order. Smaller gaps live inside the noise of a flat projection middle and are not worth a take.. Numbers regenerate on every build. A scraper refresh, a translation-factor recalibration, or a fix to a single projection all reshape these results on the next static export. The full prospect Big Board lives at /draft . Our companion look at how the 2026 expansion reshaped existing rosters is at /front-office/expansion . We will revisit this draft with real PWHL production after the 2026-27 season.