• Three goalies, three different contexts — Bobrovsky seventeen NHL games played for Florida, Stolarz seven NHL games, Akhtyamov an AHL playoff run. No cross-goalie rate in this deck is like-for-like, so this report does not compare the three to each other. Each is presented on his own against the same pool.
• Bobrovsky’s sample is his Florida body of work. He was acquired by Toronto in the 2026 offseason; all 331 tagged shots are from his 2025–26 season with the Florida Panthers. It is shown here to brief his new club. His games are never rolled into a Toronto team total anywhere in this deck, and there is no Toronto team total.
• Stolarz’s sample is small — 118 decided shots on 14 goals, against 327 for Bobrovsky and 401 for Akhtyamov. One goal moves his overall save % by about .008, so every rate on his slides is fragile.
• Two data-quality flags from import both land in Stolarz’s set (03.30.2026_ANA P3: an exact-duplicate save row, and two goals tagged at the same clock). If that second goal is spurious, GA is 13 and Sv% is .889, and ST1/ST3 shift with it. Both are to be resolved before publishing.
• Baseline (“NHL pool”) = the seven other goalies we track in data.json — 2,005 shots, includes backups — not the league.
• That pool is all NHL. For Akhtyamov’s AHL sample it is not a like-for-like peer set, and GEM holds no AHL peer set at all. Every pool comparison on his slides is indicative of shot-type profile only, never of level.
• Akhtyamov’s strength column is untagged. All 407 of his shots read 5v5, where every other goalie in the corpus runs 15–25% non-5v5 (Bobrovsky 17%, Stolarz 25%). Twenty AHL playoff games with no penalty-kill shots faced is not possible, so the column never got tagged and is sitting at its default. Two consequences: no strength split of any kind is shown for him (his slides carry an explicit placeholder where the penalty-kill row would be), and his nominal “5v5” bucket would actually contain the special-teams shots every other goalie’s 5v5 excludes — so the deck publishes no 5v5 figure for him and never sets one beside another goalie’s. His Overall and every other split shown are all-situations on both sides of the comparison and are unaffected. On the others’ rates, roughly 45–70 of his 407 shots are not 5v5.
• The panels below list the games in the analyzed sample. They carry no shots-against, goals-against or save-percentage column, and no won-lost record: we hold no box score for either goalie, so nothing of the kind is shown.

“Holds up best” is relative to his own sample, not to the pool. He is .865 overall here against the pool’s .907, so most rows on this slide still sit below the baseline — these are the situations where the gap closes or reverses, not a list of strengths against NHL peers. Only the top two rows are above the pool, and both rest on fewer than 30 shots. Pool = the seven other goalies we track in data.json (2,005 shots, includes backups) — NHL, so a like-for-like level of play for him, but not the league.

Florida sample. These are the situations from his 2025–26 Florida season, shown to brief Toronto staff. On 327 decided shots most of these rows rest on more shots than their counterparts on Stolarz’s slides — the rush row is 75 shots and the slot row 108, against 10–44 there — so the margins carry more weight, though they are still a Florida sample. Bar length is the printed number as a fraction.

Small sample. Every row above rests on 10–44 decided shots out of 118 in total. Pool = the seven other goalies we track in data.json (2,005 shots, includes backups) — NHL, so a like-for-like level of play, but not the league. A single clip moves most of these rows by several points; the one-timer row would fall to .950 on one goal.

Small sample — and it cuts both ways. The upper-glove row is 18 shots and the quick-release row 13; on populations that size a difference of this size is a reason to open the clips, not a conclusion. Two import flags sit inside this set: if the duplicated goal in 03.30.2026_ANA is spurious, GA is 13, overall Sv% is .889, and the routine and should’ve-had rows move with it. Bar length is the printed number as a fraction; the should’ve-had row is one where low is good.

AHL sample against an NHL pool. Akhtyamov played this sample in the AHL. The pool column is the seven other goalies we track in data.json (2,005 shots, includes backups) — all NHL, and not the league. GEM holds no AHL peer set, so these rows describe how his shot-type profile sits beside NHL tagging, not how he would grade against AHL or NHL peers. Read the shape, not the margin.

AHL sample against an NHL pool. As on the previous slide: the pool is the seven other goalies in data.json, all NHL, and there is no AHL peer set to substitute. A negative margin here is a prompt to review video, not evidence that he is below a peer group. Bar length is the printed number as a fraction — magnitude, not quality; the should’ve-had row is one where low is good.