Kodo HockeyRole Shifts
Whose power-play role genuinely changed, when it changed, and how sure we are. A model reads every skater's share of his club's power play game by game and reports the probability his role began within his last ten games, the game it most probably began on, and his level before against his level since.
?as_of=2026-01-15 to the address to see what a manager would have seen on any day.| # | Player | Changed? | Since | PP share before | PP share now | Δ share (pts) | Δ TOI |
|---|---|---|---|---|---|---|---|
| 1 | Devon Toews COL · D | 100% | 5 Apr · 4 gm | 22% | 66% | +45 | — |
| 2 | Ryan Poehling ANA · C | 99% | 4 Apr · 6 gm | 3% | 27% | +24 | +2.4 |
| 3 | Thomas Harley DAL · D | 87% | 11 Apr · 3 gm | 30% | 63% | +34 | — |
| 4 | Jack Roslovic EDM · C | 87% | 2 Apr · 6 gm | 19% | 54% | +35 | -2.9 |
| 5 | Vasily Podkolzin EDM · R | 82% | 8 Apr · 3 gm | 20% | 74% | +54 | +3.9 |
| 6 | Easton Cowan TOR · R | 78% | 30 Mar · 5 gm | 27% | 56% | +29 | +5.9 |
| 7 | Justin Sourdif WSH · C | 73% | 8 Apr · 3 gm | 22% | 59% | +37 | +4.5 |
| 8 | Carter Verhaeghe FLA · L | 57% | 20 Mar · 5 gm | 38% | 65% | +27 | — |
| # | Player | Changed? | Since | PP share before | PP share now | Δ share (pts) | Δ TOI |
|---|---|---|---|---|---|---|---|
| 1 | Michael Bunting DAL · L | 91% | 28 Mar · 4 gm | 27% | 0% | -27 | -3.9 |
| 2 | Jamie Benn DAL · L | 89% | 28 Mar · 8 gm | 39% | 18% | -22 | — |
| 3 | Oliver Bjorkstrand TBL · R | 76% | 21 Mar · 9 gm | 54% | 26% | -28 | +7.2 |
| 4 | Dmitry Orlov SJS · D | 59% | 11 Apr · 3 gm | 49% | 20% | -30 | — |
Changed? is the probability that his current role began within his last ten games, not the probability he is good. Since is the most probable first game of that role and how many games he has played in it. A change is listed only when it is at least even odds, at least ten points of share, seen over at least three games and no older than five weeks, so one loud night does not make the list and a man who rarely dresses cannot stay on it from November to March.
Power-play share is his power-play time over the club's, and the club's is its skaters' man-seconds over five, because five men are on the ice for every second of it. Share rather than minutes, because share is what the coach decides and minutes are what the schedule hands out. The method is Bayesian online changepoint detection: the model carries a distribution over how long the current role has run and updates it one game at a time, so five ordinary games at a new level count for far more than one extraordinary one. Calibrated on series with known changepoints, a real promotion is found inside eight games nine times in ten and a single big night is mistaken for one about seven times in a hundred. Nothing here is a projection of what happens next; for that, see Usage Trends and the player's own page.
Power-play share is the player's power-play time over the club's, and the club's is its skaters' man-seconds over five. 'Since' is the most probable first game of the current role; the probability is that the role began within his last ten games. Nothing here is a projection of what happens next.